> For the complete documentation index, see [llms.txt](https://docs.p0.dev/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.p0.dev/change-logs/p0-controller/2026.md).

# 2026

This page has all the current & past feature, updates and changes to the p0 app for 2026.

## November 2026

### New Features

#### P0 Management

* **Auditor role.** Give someone read-only access to Policy Studio, Agentic Access, access requests, integrations, and P0 settings. Auditors can't change anything, and can't view IAM assessment results. If someone also needs to view assessment results, assign them the Assessment Viewer role as well. Assign the role to users or identity-provider groups in **P0 Management** > **Access control**. See [Role-based access control](https://docs.p0.dev/p0-management/role-based-access-control.md).

#### Integrations

* **Access Logs (Beta).** See what a person did with the AWS or Google Cloud access a request gave them. The request's details page shows an **Access Logs** card with the CloudTrail events and Cloud Audit Logs entries P0 attributes to the request, in one timeline, with notes on anything that kept P0 from seeing all of it. Anyone who can view the request can see it. P0 reads the entries from your own logs through a connector you deploy, and doesn't store them. Install the new AWS CloudWatch Logs and Google Cloud Logging log sources to turn it on. The card replaces the CloudTrail Lake session history. Access Logs is in beta. Contact P0 to turn it on for your organization. See [Access Logs](https://docs.p0.dev/integrations/access-logs.md).
* **Cloudflare MCP server (Beta).** Give an AI agent Cloudflare permission groups on one account for a limited time. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), an agent requests account groups, or zone groups on one zone or all zones, with `p0 request cloudflare-ai permission`. P0 keeps one account API token for each agent with only the approved groups, and deletes it when the agent's last grant ends. The agent reaches the account through Cloudflare's hosted MCP server, and P0 never grants the groups that manage tokens, sign-in, billing or account settings. P0 acts through an install token you create and a connector you deploy on AWS Lambda or Google Cloud Run, so the install token stays in your own secret manager. This integration is in beta. See [Cloudflare MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/cloudflare.md).
* **Miro MCP server (Beta).** Give an AI agent a role on one Miro board for a limited time. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), an agent requests `viewer`, `commenter` or `editor` on a board with `p0 request miro-ai board`. P0 adds the requester's own Miro user to the board for the approved duration, then puts back the role they had before. The agent reaches the board through Miro's hosted MCP server, signed in as the requester, and P0 never grants the tools that share boards or reach across the team. P0 acts through a connector you deploy on AWS Lambda or Google Cloud Run, so the Miro token stays in your own secret manager. This integration is in beta. See [Miro MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/miro.md).
* **GitGuardian MCP server (Beta).** Give an AI agent scoped access to your GitGuardian workspace for a limited time, instead of sharing a long-lived personal access token. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), an agent requests scopes such as `incidents:read` or `scan` with `p0 request gitguardian-ai permission`. P0 creates a GitGuardian API token that holds only the approved scopes, and revokes it when the agent's last grant ends. P0 acts through a connector you deploy on AWS Lambda or Google Cloud Run, so the admin token stays in your own secret manager. You can install it with the P0 Terraform provider version 0.93.0 or later. See [GitGuardian MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/gitguardian.md). This server is in beta.
* **Oracle Cloud (Preview).** Give people just-in-time membership in the groups of an Oracle Cloud Infrastructure identity domain. A person requests a group with `p0 request oci group <domain-ocid> <group-ocid>`, or from Slack or the web app. P0 adds the person to the group for the approved duration, then removes them. You install each identity domain in the P0 Dashboard with a service user and API key that you create in Oracle Cloud. See [Oracle Cloud](https://docs.p0.dev/integrations/resource-integrations/oracle-cloud.md). This integration is in preview.
* **Database Activity in Access Logs (Beta).** See the SQL statements a person ran during a PostgreSQL or MySQL access request, on the request's Access Logs panel. P0 reads them from the database's own audit log on AWS RDS, Aurora, or Google Cloud SQL, through the AWS CloudWatch Logs or Google Cloud Logging integration for the database's account or project. You turn on auditing for the database yourself, and the PostgreSQL and MySQL installers say what to turn on. Contact P0 to enable Access Logs. See [Database auditing](https://docs.p0.dev/integrations/access-logs/database-auditing.md).

#### User Experience

* **Maximum Duration on Policy Actions.** Route requests by how long they ask for from the console, instead of only through the API or Terraform. In the policy editor, check **Match policy action by requested duration** on an action and enter a **Maximum duration** as a number and a unit, up to your organization's maximum. The action then handles only requests of that length or shorter, and the policy and access bundle lists show the limit on each action. See [Configuring access policies](https://docs.p0.dev/policy-studio/policy-studio/configure-your-first-access-policy.md#limit-access-duration-for-specific-resources).

### Enhancements

#### Integrations

* **Integrations Catalog Availability.** Splunk HTTP Event Collector, Zscaler Private Access (Preview), Remote Desktop Protocol (RDP), and File transfer (Beta) now appear in the P0 Dashboard integrations catalog without P0 enabling them for your organization. The **Docs** button on their pages, and on AWS OIDC, AWS Secrets Manager, Databricks, Harbor, and Google Cloud Secret Manager, now opens the matching docs page. See [Splunk HEC setup](https://docs.p0.dev/integrations/siem-integrations/splunk-hec-setup.md).
* **Agentic Gateway Generally Available.** The AI Gateway, MCP server, A2A bridge, and log provider components are now generally available, along with the P0-hosted domain, MCP tool and tool-annotation access, and the OAuth, federated, and token exchange credential sources. The Agentic Access page no longer shows a **Beta** tag. Vaulted credentials and service account impersonation are in preview, and Google Cloud Logging as a log provider is in beta. See [Gateway](https://docs.p0.dev/agentic-access/agentic-gateway/gateway.md).
* **Gateway Install Uses the Latest Chart and Shorter Terraform.** The [Install from the P0 console](https://docs.p0.dev/agentic-access/agentic-gateway/gateway/deploy/install-from-the-p0-console.md) installer now looks up the latest stable gateway chart version and puts it in the Terraform it generates, instead of a version fixed in the console. It also leaves out values that match the chart's defaults, so the snippet you review is shorter. If P0 can't look up the version, the installer shows an error and doesn't generate the Terraform.

#### User Experience

* **Event Details Title and Delegation Flow.** The [Event log](https://docs.p0.dev/agentic-access/using-the-gateway/event-log.md) drawer now opens with a title that names the event's risk, or the call when it has none, the person or agent the event relates to, and the client it came through, instead of a generic "Event Details" heading. The **Delegation Chain** now reads top to bottom, from the originating subject through each agent it delegated to, and ends at the MCP server or A2A bridge the call reached. A request's **Agentic Details** uses the same order.

#### API & CLI

* **Access Bundles in Terraform (Preview).** Manage an access bundle as code with the new `p0_access_bundle` resource, instead of building it in Policy Studio. It takes a label, description, requestor, approval and two to fifteen items, and imports an existing bundle by its ID. Set `enable_preview_resources = true` in the provider block and use P0 Terraform provider version 0.90.0 or later. `p0_access_policy` now rejects names that start with `bundle:`, because P0 keeps each bundle's policy under that name. This resource is in preview. See [Access bundles](https://docs.p0.dev/policy-studio/policy-studio/access-bundles.md).
* **Microsoft Entra ID Terraform Resources.** Install Microsoft Entra ID with Terraform instead of the P0 Dashboard installer. Stage the tenant with `p0_entra_id_staged`, create P0's app registration with its federated identity credential and Microsoft Graph permissions, then install it with `p0_entra_id`. Set `email_field` to `userPrincipalName`, `mail` or `otherMails`, and `sovereign_cloud_id` to `AzureCloud` or `AzureUSGovernment`. You deploy the security perimeter in Azure separately. This needs P0 Terraform provider version 0.92.0 or later. See [Microsoft Entra ID](https://docs.p0.dev/integrations/directory-integrations/microsoft-entra-id.md).
* **AWS Lambda Notifier Terraform Resources.** Install an AWS Lambda notifier with Terraform. Stage the function caller with `p0_aws_function_caller_staged`, which returns the IAM role to create in the function's account, install it with `p0_aws_function_caller`, then create the notifier with `p0_aws_lambda_notifier`. This needs P0 Terraform provider version 0.91.0 or later. See [AWS Lambda notifier](https://docs.p0.dev/integrations/notifier-integrations/custom-notifiers/aws-lambda-notifier.md).

### Stability & Fixes

* Fixed the error shown when an existing invalid policy blocks you from saving or deleting another policy or access bundle rule. It now names the invalid policy instead of reading as if the change you made were at fault.
* Fixed SSO sign-in errors appearing as a browser alert with technical error codes. A failed sign-in now shows a short message on the login page. If your organization already has an account for your email that uses a different sign-in method, the message tells you to sign in the way you did before, or to ask your P0 administrator to remove the old account.
* Fixed an access bundle item that listed more than one value for a parameter, such as two roles, returning an internal error. The API now says a bundle item grants one value and asks for one item for each role.
* Fixed the MCP server and A2A bridge install forms failing without a message when an ID starts with the reserved `p0-` or `a2a-` prefix. The ID field now shows the error.
* Fixed the Cloud SQL, custom application, Google Cloud service account, and Google Drive install checks on Google Cloud failing when the connector was still deploying. P0 now looks up the connector's Cloud Run service again before it reports an error.
* Fixed the first Google Cloud resource request that needed a new P0 custom role, such as a role on a bucket or a Compute Engine instance, failing because Google Cloud hadn't made the role available yet. P0 now waits for the role, so the request succeeds without you submitting it again.
* Fixed `p0 request ssh group --destination <instance>` failing on Google Cloud with "SSH groups have not been configured" when you gave a destination and no group name. P0 now finds the group from the destination, and a destination with no group returns "The selected destination does not belong to a group."
* Fixed saving an access bundle in the Policy Studio editor erasing a description that was set through the API or Terraform.

## October 2026

### New Features

#### Integrations

* **ClickHouse Cloud (Beta).** Give people just-in-time access to the services of your ClickHouse Cloud organizations: a read, write or admin level on a whole service, a database, a table, a view or a dictionary, or a console role such as SQL console read-only. A person requests it with `p0 request clickhouse sql` or `p0 request clickhouse console`, or from Slack or the web app. For each approved request, P0 creates a custom role with the person's own ClickHouse Cloud login as its only member, and deletes it when access ends. P0 acts through an API key you create and a connector you deploy on AWS Lambda or Google Cloud Run, so the API key stays in your own secret manager. See [ClickHouse Cloud](https://docs.p0.dev/integrations/resource-integrations/clickhouse.md). This integration is in beta.
* **Tailscale (Beta).** Give people just-in-time network access through the P0 postures your tailnet's policy file defines. Your own grants require a posture with `srcPosture`, and a person requests it on one of their devices with `p0 request tailscale posture <posture> --device <device>`, or from Slack or the web app. P0 sets the posture's device attribute on that device for the approved duration, then removes it. P0 never writes your policy file, so it works with a policy file you manage with GitOps or Terraform. Your access policies decide who can request which posture. The integration needs a Tailscale Premium or Enterprise plan. P0 acts through a connector you deploy on AWS Lambda or Google Cloud Run, which signs in to Tailscale with a federated identity, or with an OAuth client whose secret stays in your own secret manager. This replaces the previous Tailscale integration: install each tailnet again, and for access people requested as a device, a tag or Tailscale SSH, write a grant that a P0 posture gates. `p0 request tailscale device`, `tag` and `ssh` are removed, and `p0 request tailscale posture` takes `--tailnet` in place of `--organization`. See [Tailscale](https://docs.p0.dev/integrations/resource-integrations/tailscale.md). This integration is in beta.
* **Grafana IRM (Beta).** Approve just-in-time access based on who is on call in Grafana IRM. P0 auto-approves a request when the requestor is on call for one of the IRM teams you select. A requestor can also escalate a pending request to page the IRM teams in your access policy, and anyone on call for those teams can approve it. P0 resolves the alert group once someone approves or denies the request. See [Grafana IRM](https://docs.p0.dev/integrations/approval-integrations/grafana-irm.md). This integration is in beta.
* **Auth0 (Beta).** Give people just-in-time access in your Auth0 tenants: a tenant role, a permission on an API, or membership of an organization, optionally with a role in it. A person requests it with `p0 request auth0 role`, `p0 request auth0 permission` or `p0 request auth0 organization`, or from Slack or the web app. P0 assigns the access to their existing Auth0 user for the approved duration, then takes back only what it assigned. P0 acts through a Management API application you create and a connector you deploy on AWS Lambda or Google Cloud Run, so the application's client secret stays in your own secret manager. See [Auth0](https://docs.p0.dev/integrations/resource-integrations/auth0.md). This integration is in beta.
* **Argo CD (Beta).** Give people just-in-time roles in your [Argo CD](https://argo-cd.readthedocs.io) instances. A person requests a role on a whole instance, such as `role:readonly`, with `p0 request argocd global-role <role>`, or a role that one project defines with `p0 request argocd project-role <project> <role>`, or from Slack or the web app. P0 gives their SSO user the role for the approved duration, then takes it away. P0 acts through an Argo CD account you add in Git and a connector you deploy on AWS Lambda or Google Cloud Run, so the account's API token stays in your own secret manager. See [Argo CD](https://docs.p0.dev/integrations/resource-integrations/argocd.md). This integration is in beta.
* **1Password (Beta).** Give people time-limited access to one item in a 1Password vault. A person requests an item with `p0 request 1password item <account> <vault> <item>` or from Slack or the web app. Once the request is approved, P0 makes a read-only share link to the item that opens only for the requestor's email, and that expires after the approved time, rounded to one of 1Password's link durations. P0 acts through a 1Password service account for each 1Password account, with access to the vaults people may request items from, and a connector you deploy on AWS Lambda or Google Cloud Run, so the service account token stays in your own AWS Secrets Manager or Google Secret Manager. See [1Password](https://docs.p0.dev/integrations/resource-integrations/1password.md). This integration is in beta.
* **GitHub (Beta).** Grant members of your GitHub organization just-in-time roles on individual repositories and memberships of its teams. A member requests a role on one repository, such as Write or Maintain, with `p0 request github role <repository> <role>`, or a membership of one team with `p0 request github team <team> <role>`, or from Slack or the web app. P0 makes them a direct collaborator or a team member for the approved duration, then puts back what they had before. P0 acts through a GitHub App you create and install, and a connector you deploy on AWS Lambda or Google Cloud Run, so the app's private key stays in your own AWS Secrets Manager or Google Secret Manager. See [GitHub](https://docs.p0.dev/integrations/resource-integrations/github.md). This integration is in beta.
* **GitHub Force-merge Access (Beta).** Let a member of your GitHub organization merge pull requests into one protected branch without its required reviews, for a bounded time, with `p0 request github force-merge <org>/<repo>/<branch>` or from Slack or the web app. P0 adds them to the branch's list of people who can bypass required reviews, and takes them off it when access ends. The branch needs a classic branch protection rule for its exact name. Rulesets, required status checks and other rules still apply. It uses the permission the GitHub App already has. See [GitHub](https://docs.p0.dev/integrations/resource-integrations/github.md#force-merge-access). This access is in beta.
* **Docker Hub (Beta).** Grant just-in-time member, editor or owner roles in your Docker Hub organizations. A person requests a role with `p0 request docker role <organization> <role>` or from Slack or the web app. P0 changes the role of someone already in the organization for the approved duration, then puts back the role they had before. Someone who isn't in the organization gets an invite, and loses the invite or their membership when access ends. P0 acts through a connector you deploy on AWS Lambda or Google Cloud Run, so the organization access token stays in your own AWS Secrets Manager or Google Secret Manager. See [Docker Hub](https://docs.p0.dev/integrations/resource-integrations/docker.md). This integration is in beta.
* **Harbor (Beta).** Give people just-in-time roles on the projects of your [Harbor](https://goharbor.io) instances. A person requests a role on one project, such as Developer or Maintainer, with `p0 request harbor project-role <instance> <project> <role>` or from Slack or the web app. P0 makes their existing Harbor user a direct member of the project for the approved duration, then puts back the lower role they had before or removes the membership. P0 acts through a dedicated Harbor admin account from your identity provider and a connector you deploy on AWS Lambda or Google Cloud Run, so the account's credential stays in your own secret manager. Harbor must sign people in with OIDC. See [Harbor](https://docs.p0.dev/integrations/resource-integrations/harbor.md). This integration is in beta.
* **Harbor System Administrator Access (Beta).** Give people Harbor's system administrator access on one instance for a bounded time, with `p0 request harbor system-admin <instance>` or from Slack or the web app. P0 refuses someone who is already a system administrator, and clears the administrator flag when access ends. Every Harbor instance offers it: P0 changes the flag as the instance's admin account, which patched Harbor requires. See [Harbor](https://docs.p0.dev/integrations/resource-integrations/harbor.md#system-administrator-access). This access is in beta.
* **SaaS MCP servers (Beta).** Give AI agents OAuth-authenticated, policy-scoped access to popular SaaS tools through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), instead of distributing a vendor token to each agent. P0 now ships predefined MCP servers for tools such as Atlassian, Notion, Linear, Stripe, and PayPal. Choose **OAuth** as the credential source and **P0** as the definition, then pick the server. The OAuth credential source is in preview. Each user signs in with OAuth on first use, and calls run through the gateway under your access policy. See [SaaS MCP servers](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/saas.md).
* **Databricks MCP servers (Beta).** Grant an AI agent temporary access to one Unity Catalog catalog, schema, table, view, or volume in your Databricks accounts on AWS, at a `read` or `write` level, instead of handing it a long-lived Databricks token. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), the agent uses Databricks' Genie and Databricks SQL MCP servers as a service principal that P0 creates for that grant alone and deletes when the grant ends. P0 acts through a connector you deploy on AWS Lambda, which signs in to Databricks without a stored secret. See [Databricks MCP servers](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/databricks.md). This integration is in beta.
* **Google Drive MCP server (Beta).** Grant an AI agent scoped access to a single Google Drive folder or file, instead of sharing a drive broadly or handing over a long-lived service-account key. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), P0 grants the agent `reader` or `writer` access to only the folder or file you choose. This integration is in beta. See [Google Drive MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/google-drive.md).
* **Snowflake MCP server (Beta).** Give an AI agent temporary access to a Snowflake account through Snowflake's managed MCP server, as a Snowflake user of its own. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), an agent requests an existing role, or `SELECT`, `INSERT`, `UPDATE` or `DELETE` on a database, schema, table or view. P0's connector in your Google Cloud project creates a user and role for each grant and removes them when access ends. This integration is in beta. See [Snowflake MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/snowflake.md).
* **PostgreSQL on AWS RDS MCP server (Preview).** Let an AI agent run SQL against an Amazon RDS for PostgreSQL database through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md) without holding database credentials. The agent requests a scoped Postgres role on an RDS instance, and the gateway connects with short-lived RDS IAM credentials for the session. This server is in preview.
* **Agentic tool-call risk scoring.** See the risk of each agent action at a glance. The [Event log](https://docs.p0.dev/agentic-access/using-the-gateway/event-log.md) now shows a **Risk** level (Critical, High, Medium, or Low) for a tool call. P0 derives the level from the tool's own annotations and, for the GCP Storage, IAM, BigQuery, and Compute Engine MCP servers, grounds it in the actual IAM permission behind the call. A tool that reports itself as read-only can still score High if the permission behind it can read sensitive data.
* **Databricks (Beta).** Give people just-in-time Unity Catalog privileges, workspace admin and account admin in your Databricks accounts. P0 grants the privilege to the Databricks user whose username is their P0 email, removes what it added when access ends, and never removes a privilege they already held. P0 doesn't assign the user to a workspace or give them compute, so they need both already. You can install it with the P0 Terraform provider version 0.64.0 or later. See [Databricks](https://docs.p0.dev/integrations/resource-integrations/databricks.md). This integration is in beta.
* **Jira MCP server (Beta).** Give an AI agent time-bound access to Jira tools in one project of your Jira site. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), P0 leases each agent session an account from a pool you provide and adds it to a role in the project. P0's connector runs on Google Cloud, and you can install it with the P0 Terraform provider version 0.72.0 or later. See [Jira MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/jira.md). This server is in beta.
* **GitLab MCP server (Beta).** Give an AI agent time-bound roles on the projects of your GitLab.com top-level group. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), an agent requests a role such as Reporter or Developer on a project. P0's connector creates a GitLab service account for the person, adds it to each approved project with the highest role requested, and deletes it when access ends. The Owner service account's token stays in your own secret manager. You can install it with the P0 Terraform provider version 0.78.0 or later. See [GitLab MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/gitlab.md). This server is in beta.
* **Tableau MCP server (Beta).** Give an AI agent time-bound read or write access to projects on your Tableau site. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), P0's connector creates a Tableau user for each grant and gives it the approved level on the approved projects. The projects must have their permissions locked to the project, and Tableau emails an invitation to each new user. You can install it with the P0 Terraform provider version 0.76.0 or later. See [Tableau MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/tableau.md). This server is in beta.
* **ServiceNow MCP server (Beta).** Give an AI agent a time-bound ServiceNow role, such as `itil`, on one of your ServiceNow instances. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), P0 creates a ServiceNow user with the approved role for each agent session, and the gateway calls the ServiceNow REST API as that user. Only an agent session can request it, and `--instance` can be left out when one instance is installed. You can install it with the P0 Terraform provider version 0.75.0 or later. See [ServiceNow MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/servicenow.md). This server is in beta.
* **Amazon SQS (Beta).** Send P0's audit log events to an Amazon SQS queue in your AWS account, so any tool that reads from SQS can consume them. P0 sends each event as one JSON message to a standard or FIFO queue. You create an IAM role named `P0RoleIamEventSender` that trusts the service account P0's AWS integration already uses, so P0 stores no secret, and you give P0 the queue ARN and an optional KMS key ARN. An event larger than 1 MiB isn't sent. If P0 can no longer send, for example because the queue was deleted or access was denied, it skips events until you resume the integration. You can install one queue per organization with the P0 Terraform provider version 0.79.0 or later. See [Amazon SQS setup](https://docs.p0.dev/integrations/siem-integrations/amazon-sqs-setup.md). This integration is in beta.
* **LaunchDarkly (Preview).** Give people just-in-time base roles in your LaunchDarkly accounts. A person requests `reader`, `writer` or `admin` with `p0 request launchdarkly base-role <account>/<role>`, or from Slack or the web app. P0 raises the base role of their existing LaunchDarkly member for the approved duration, then puts back the role they had before. P0 acts through an access token you create and a connector you deploy on AWS Lambda or Google Cloud Run, so the token stays in your own secret manager. You can install it with the P0 Terraform provider version 0.86.0 or later. See [LaunchDarkly](https://docs.p0.dev/integrations/resource-integrations/launchdarkly.md). This integration is in preview.
  * **Action required:** If you set `plan` on `p0_launchdarkly_account` with P0 Terraform provider version 0.83.0 to 0.85.x, remove it before you upgrade to 0.86.0 or later.
* **Looker MCP server (Preview).** Give an AI agent time-bound access to one LookML model in your Looker (Google Cloud core) instance. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), an agent requests `read` to explore the model, run queries, and view Looks and dashboards, or `edit` to also save them. P0's connector creates a Looker service account for each grant and deletes it when access ends. The connector runs on Google Cloud Run, and you can install it with the P0 Terraform provider version 0.85.0 or later. See [Looker MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/looker.md). This server is in preview.
* **Amazon DynamoDB MCP server (Beta).** Give an AI agent time-bound read or write access to one DynamoDB table and its indexes, in one AWS account and Region. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), P0 adds an IAM policy statement for the table to a role in your AWS OIDC role pool, and removes it when access ends. Only agents can request it. It needs P0's AWS IAM management and AWS OIDC installed for the account, and you can install it with the P0 Terraform provider version 0.86.0 or later. See [Amazon DynamoDB MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/dynamodb.md). This server is in beta.
* **MongoDB Atlas MCP server (Beta).** Give an AI agent time-bound roles in one of your MongoDB Atlas projects, such as `GROUP_DATA_ACCESS_READ_ONLY`. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), P0's connector creates an MCP configuration in the project with the approved roles and stores its token in your secret manager. The gateway calls MongoDB's hosted Atlas MCP server with that token, and P0 deletes the configuration when access ends. The connector runs on AWS Lambda or Google Cloud Run. See [MongoDB Atlas MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/mongodb-atlas.md). This server is in beta.
* **Amazon Redshift MCP server (Beta).** Give an AI agent time-bound access to a Redshift provisioned cluster or Serverless workgroup. Through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), an agent requests existing Redshift roles, or `SELECT`, `INSERT`, `UPDATE` or `DELETE` on a database, schema, table or view. For each grant, P0 creates an IAM role and a database user that signs in only through it, and removes both when access ends. P0 acts through a connector you deploy on AWS Lambda, and you can install it with the P0 Terraform provider version 0.87.0 or later. See [Amazon Redshift MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/redshift.md). This server is in beta.
* **Amazon S3 Audit Logs (Beta).** Send P0's audit log events to an Amazon S3 bucket in your AWS account, so tools that read log files from S3, such as Splunk and Amazon Athena, can ingest them. On a 15-minute schedule, P0 writes a gzipped file with one JSON event per line, in date folders that Athena can discover. You create an IAM role named `P0RoleIamS3Writer` that can only add new files to your bucket. P0 writes events from the moment you finish the install, and supports one bucket per organization. You can install it with the P0 Terraform provider version 0.84.0 or later. See [Amazon S3 (audit logs) setup](https://docs.p0.dev/integrations/siem-integrations/amazon-s3-audit-setup.md). This integration is in beta.

### Enhancements

#### Integrations

* **SSH Without Tailscale Network Access.** SSH installs no longer offer Tailscale as a **Network-access provider**. An SSH install that used Tailscale now grants SSH access without network access. To give people network access to your hosts, install P0 for your tailnet, and have people request Tailscale access separately. See [Tailscale](https://docs.p0.dev/integrations/resource-integrations/tailscale.md).
* **AWS Secret Write Access.** When you request access to a single secret in AWS Secrets Manager, you can now choose read and write access, scoped to that secret's ARN, instead of read-only.
* **Cloud SQL Shell Install (Preview).** Install the Google Cloud SQL connector with a generated `gcloud` shell command as an alternative to Terraform, so teams that don't use Terraform can deploy it. See [Installing a Cloud SQL instance](https://docs.p0.dev/integrations/resource-integrations/postgresql/installing-a-cloudsql-database.md). This option is in preview.
* **Snowflake and Tableau MCP Servers on AWS.** Install the Snowflake and Tableau agentic connectors on AWS Lambda with AWS Secrets Manager, instead of only on Google Cloud Run with Google Secret Manager. Choose **AWS Secrets Manager** at install and P0 generates the Terraform that deploys the connector as a Lambda function in your AWS account. With the P0 Terraform provider, set `secret_manager.type` to `aws-sm`, which needs version 0.80.0 or later for Snowflake and 0.81.0 or later for Tableau. See [Snowflake MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/snowflake.md).
* **Agentic Integrations Move to Beta.** The PagerDuty, GitHub App, Datadog, Salesforce and Google Drive agentic integrations are now in beta and appear on the **Integrations** page without P0 enabling them for your organization first. The **Vaulted** credential source for MCP servers, which the PagerDuty, Datadog and GitHub App servers use, now shows in the MCP server form and is in beta too.
* **More Predefined OAuth MCP Servers.** The P0 catalog of OAuth MCP servers now includes dozens of SaaS tools, such as Cloudflare, Grafana Cloud, PostHog, Calendly and monday.com. Choose **OAuth** as the credential source and **P0** as the definition, then pick the server. Each person signs in with OAuth on first use. An admin of monday.com or Miro may need to approve the connection first. See [SaaS MCP servers](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/saas.md).
* **Google Cloud Logging and Firestore MCP Servers.** Give an AI agent access to Cloud Logging and Firestore through Google's hosted MCP servers, alongside the other Google Cloud MCP servers. An agent requests a role on the project, such as `datastore.viewer` for Firestore, and P0 grants it for the approved duration. Firestore works only with Native mode databases.
* **GitHub MCP Server Tool List.** The GitHub MCP server now offers 49 tools instead of 40, so an agent can reply to a pull request review comment and edit an issue comment. See [GitHub MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/github.md).
* **AI Gateway Installer Updates.** The AI Gateway installer in the P0 console now generates Terraform for version 1.3.0 of the gateway module, which adds Valkey to the gateway's storage. The **Gateway storage** help names an encrypted storage class for each cloud, such as `standard-rwo` on GKE and `managed-csi` on AKS, and warns when you enter `gp2`. An optional **Identity provider issuer** field lets you sign in with a provider other than Google, such as Okta or Microsoft Entra ID, and a non-Google issuer turns off the verified-email check. With no persistent storage, OAuth credentials are lost when the gateway's pods restart. See [Install from the P0 console](https://docs.p0.dev/agentic-access/agentic-gateway/gateway/deploy/install-from-the-p0-console.md).
* **Install Item IDs With Special Characters.** An install item ID can now contain `/`, `?`, `#` or spaces. Before, P0 rejected an ID with `/`, and a link to an item whose ID contained `%` opened the wrong item. Cloudflare SCIM, Grafana Cloud, the agentic MCP server and A2A bridge, and custom inventory still reject `/`.
* **Extend Policies for AWS Resource Requests on Merged Identity Center.** An AWS resource request on the merged Identity Center login now offers **Extend policies**, as policy requests already did, and `extends` works from the CLI and API. Permission set lookups no longer need the separate Identity Center login type installed. When two requests that extend the same permission set end together, P0 now detaches the extended policies. Merged Identity Center doesn't copy the extended permission set's inline policy. See [Identity Center (merged)](https://docs.p0.dev/integrations/resource-integrations/aws/identity-center-merged.md).
* **Custom Application Connector URL (Preview).** When a custom application's connector runs on Cloud Run, you can enter a **Connector URL**, an `https` origin with no path. P0 sends every request to it, including the install test, instead of the `run.app` URL, so you can put your own load balancer in front. See [Deploying a connector to Google Cloud Run](https://docs.p0.dev/integrations/resource-integrations/custom-application/deploying-a-connector-to-google-cloud-run.md). This option is in preview.
* **Clearer Kubernetes Access Requests.** In the Kubernetes resource pickers, a namespace now reads "Namespace object: default" instead of "Namespace / default", so it isn't mistaken for everything inside the namespace. Slack and approval messages lead each resource with its kind, such as `ClusterRole / x`. The pickers offer only role and resource combinations that match, and P0 rejects a role with no rules for the requested resource when you make the request, instead of approving it and failing later. The match ignores API groups. See [Requesting Kubernetes access](https://docs.p0.dev/integrations/resource-integrations/kubernetes/requesting-access.md).

#### User Experience

* **Per-Rule Access Limits.** Set a maximum access duration on an individual approval rule in an access policy through the [Access policies API](https://docs.p0.dev/policy-studio/policy-studio/access-policies-api.md). P0 enforces it on every request, granting the smaller of the matching rule's limit and your organization's maximum. You no longer need to include a precomputed duration value. P0 derives it from the time and unit you set.
* **Agentic Access Event Log Opens First.** The [Event log](https://docs.p0.dev/agentic-access/using-the-gateway/event-log.md) tab now loads first, so you land on recent agent activity instead of the client list.
* **Clearer OAuth Install Errors.** A failed OAuth integration install, such as PagerDuty, Slack, Microsoft Teams, Google Workspace, or Microsoft Entra ID, now shows the provider's own error message, and an install that finishes in a different browser tab shows a retry message instead of appearing stuck.
* **Retry a Failed Expiry.** When P0 can't remove access as a request expires, retry the removal with **Retry revoke** on the request's page in the web app or through the [Access requests API](https://docs.p0.dev/access-management/access-management/just-in-time-access/just-in-time-api/access-requests-api.md), the same way as a failed revoke, instead of removing the access by hand. A successful retry ends the request as expired, and its `permission-requests.expired` audit event includes a `revocationDetails` object that identifies who retried it. A retry through the bulk revocation endpoint records `api` as the `revocationSource`, and one made with an API key records the key's ID and name instead of a user's. See [/pages/5i3rIBsphOdpqWrBMnXH#permission-requests.expired](https://docs.p0.dev/pages/5i3rIBsphOdpqWrBMnXH#permission-requests.expired "mention").
* **MCP Server Form Starts With the Server.** When you add a P0-defined MCP server, you now choose the server first, and the credential options narrow to those that work with it. Before, you chose the credential first and got "Server is incompatible with" the server you picked. The credential source now defaults to Native instead of AWS, and the server list is sorted by name. See [MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server.md).
* **Agent Client Owners.** Each agent client can now have an owner, the email of a P0 user. Registering a client in the web app or the API makes you its owner, and clients a gateway registers on its own have none. The client list has a searchable **Owner** column, and anyone who can update clients can change the owner in the client's drawer or with `PATCH /agentic/clients/<client ID>` and an `owner` field. In a federated client's drawer, **Register** under OAuth credentials is disabled unless you own the client. See [Agentic client registration API](https://docs.p0.dev/agentic-access/using-the-gateway/connect-an-mcp-client/client-registration-api.md).
* **Revoke Client OAuth Credentials.** Revoke the OAuth credentials of a federated agent client from the client's details drawer. The drawer lists each credential, and **Revoke** asks you to confirm. The API offers `GET /agentic/clients/<client ID>/credentials` and `DELETE /agentic/clients/<client ID>/credentials/<credential ID>`, and each revoke appears in the audit log. You need permission to update clients. A credential whose MCP server was uninstalled is still listed, and revoking it returns an error. See [OAuth servers from a headless agent](https://docs.p0.dev/agentic-access/using-the-gateway/headless-oauth-credentials.md).
* **Event Log Risk Filter and Delegated-By Identity.** Filter the [Event log](https://docs.p0.dev/agentic-access/using-the-gateway/event-log.md) by **Risk** level, and hover a **Risk** value to see what set it. The **Identity** column now shows the person or upstream agent the client acted for, followed by "via" and the client. The risk filter runs after P0 reads the log, so a filtered page can hold fewer rows than you expect.
* **Role Groups From Your Directory.** On the access control settings page, each role's **Groups** field is now a searchable list of groups from your directory when your sign-in provider has a matching directory installed, Microsoft Entra ID or Okta. Before, you typed a value that had to match the token's groups claim, such as a GUID for Entra ID. Organizations without a matching directory keep the text field. See [Group-based access control](https://docs.p0.dev/p0-management/group-based-access-control.md).
* **Bundle Conflict Checks.** P0 now refuses an access bundle with two items that ask for different roles on the same resource, such as two roles on one GitHub repository, one Docker Hub organization or one Harbor project. The check runs when you save the bundle and when someone makes a request. See [Access bundles](https://docs.p0.dev/policy-studio/policy-studio/access-bundles.md).
* **Back Button in the Integration Installer.** Return to the identifier form from the setup-instructions step to fix a mistyped identifier or option, instead of leaving the installer and typing everything again. The form opens with your entries filled in. If you submit a different identifier, P0 warns you and removes the item you came back from. Anything you already set up in your own environment stays in place.
* **Component Link in Install Breadcrumbs.** On an install item's page, the component name in the breadcrumbs is now a link to that component's table, so you can go back in one click instead of returning to the integration page and choosing the component again.

#### API & CLI

* **Terraform Access Policy Max Duration.** Set `max_duration` on an approval in a `p0_access_policy` resource to route requests by how long they ask for, with a `time` and a `unit` of `s`, `m`, `h`, `d` or `w`. An approval matches a request only when the requested duration is no longer than its `max_duration`. This needs P0 Terraform provider version 0.65.0 or later.
* **Terraform Agent-to-Agent Access Policy Requestors.** Write a `p0_access_policy` that matches one agent acting for another agent, instead of only an agent acting for a person. In `requestor`, set `user.type` to `agent` and describe the agent it acts for in `user.agent`, using `any`, `agent-client`, `agent-owner`, `owner-group` or `provider`. The `agent` block still describes the agent making the request. This needs P0 Terraform provider version 0.85.0 or later. See [Agentic access policies](https://docs.p0.dev/agentic-access/agentic-access-policies.md).

### Stability & Fixes

* Fixed AWS Identity Center assignment changes failing with a conflict while another change was in progress. P0 now retries before it reports an error.
* Fixed grants into the AWS account that hosts Identity Center failing with `Error assigning user permission` on the merged Identity Center login. Existing installs must rerun the setup commands or re-apply `p0_aws_midc_staged` to get the fix.
* Fixed `p0 request aws resource <arn> <policy name>` being approved and then failing at grant with a policy ARN length error. P0 now resolves the name to its ARN when you make the request, and an unknown name fails with `Policy '<name>' does not exist`.
* Fixed Slack and web AWS resource requests sending "Expected resource ARN" when you searched policies before choosing a resource. The **Policies** field now appears after you choose a resource, and **Extend policies** with it.
* Fixed Kubernetes request errors that didn't say what was wrong. A missing role now names the role and cluster and points to `p0 ls k8s resource role`, and an unreachable Kubernetes proxy no longer tells people to reinstall.
* Fixed Google Kubernetes Engine control-plane upgrades failing and rolling back on clusters that use the P0 permission boundary. The install manifest in the console now calls P0's admission webhook only for P0's own requests, which needs Kubernetes 1.28 or later. Existing installs need `matchConditions` added to the `p0-admission-controller` webhook configuration.
* Fixed the PostgreSQL on RDS setup error when the connector can't connect during `terraform apply` of `p0_postgres`. It now lists the SQL statements to run as the master user.
* Fixed PostgreSQL on RDS logins failing with `PAM authentication failed` for a principal whose database username passed Postgres's 63-byte limit, such as an AWS role ARN from outbound identity federation. P0 now falls back to a shorter name. MySQL on RDS names between 33 and 63 characters still fail.
* Fixed the PostgreSQL on AWS RDS MCP server rejecting every write with "Error validating query", even for an agent that held a write role such as `pg_write_all_data`. The role's privileges now decide what an agent can run, and the server no longer runs in a read-only mode of its own.
* Fixed the custom application install check on Google Cloud failing with "P0 is not permitted to read the Cloud Run service". The setup steps now also grant P0's service account `roles/run.viewer`, and an install that granted only `roles/run.invoker` needs to add it.
* Fixed custom application errors on AWS that all read "P0 could not reach the connector". A missing Lambda function, or one P0 isn't allowed to invoke, now shows its own message.
* Fixed connector install checks failing with `500 unknown error` or a raw `lambda:InvokeFunction` denial right after you deployed a connector. P0 now retries while permissions apply, and the message says what to check.
* Fixed the Databricks request form failing in Slack with "P0 encountered an unknown error". Slack caps placeholder text at 150 characters, so P0 now shows the detail as hint text.
* Fixed the Agentic Access Servers tab and `GET agentic/servers` listing SaaS MCP servers, such as Atlassian, as container servers. They now show as external.
* Fixed replacing the public key of an installed AI Gateway failing on save. The change now saves, and the gateway reconnects after it restarts.
* Fixed approvers who belong to Microsoft Entra ID groups not receiving the Slack DM for requests routed to their group. P0 now also matches people by their mail aliases when the user principal name differs from their Slack email.
* Fixed the group picker on the policy tab staying on "Loading groups..." when one installed directory failed to list its groups. It now shows the groups that loaded.
* Fixed Google Cloud assessments failing when a deleted service account still appeared in Cloud Asset inventory. P0 now skips the account and its keys.
* Fixed **Show credentials like this** in the IAM assessment failing when you clicked the used or unused credential counts.
* Fixed Slack and Teams standing-access messages repeating the last item when they named three or more, as in "a, b, c, and c".
* Fixed an approver's Slack DM leaving a request waiting when Slack didn't respond. P0 now times out Slack calls after 13 seconds instead of waiting indefinitely.
* Fixed P0 Terraform provider plans failing when P0 rate-limited the provider, with an `invalid character '<'` error. One rate-limit response now pauses all requests, and the provider honors `Retry-After` and retries for up to 5 minutes. This needs version 0.69.0 or later.
* Fixed the approval dialog cutting the grant after 32 characters. It now shows the whole grant, with long ARNs wrapping. The request list, filters and CSV export still shorten it.
* Fixed **Access Properties** on a request's page leaving out the access P0 provisions separately to support the request, such as the Google Cloud access behind a Cloud SQL request. Each one now has its own section, named for its integration. For a bundle request, **Access Properties** shows what was granted for each item, instead of a separate card with the bundle's current definition.
* Fixed resource search in Slack and the web app listing an exact match after other options with the same score. An option whose value equals your search term now sorts first, so searching for an S3 bucket name no longer lists it after EKS resources whose ARNs only contained the term.
* Fixed the integrations catalog row named PostgreSQL (new). It is now PostgreSQL, and the older integration is now PostgreSQL (Legacy), which stays hidden unless P0 has enabled it for your organization.
* Fixed Azure SSH and RDP access requests failing with a raw Azure `ResourceNotFound` error when the target virtual machine no longer exists. The request now names the missing VM and asks you to confirm it exists. When the configured jump host or Azure Bastion host is gone, it tells you to ask your P0 admin to reconfigure it.
* Fixed IAM assessment runs that kept getting interrupted staying in progress and holding up your other runs. A run that is interrupted repeatedly, or makes no progress for 12 hours, now ends as failed with a message that says it was interrupted.
* Fixed requesting Cloudflare access ending in "An unexpected error occurred while checking grants" when Cloudflare rejected P0's email lookup for the requestor. P0 now falls back to scanning users, and shows an integration's own error instead of the generic one.

## September 2026

### New Features

#### Integrations

* **P0-hosted AI Gateway Domain (Beta).** Stand up a self-hosted P0 AI Gateway (see [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md)) without configuring DNS by hand. When you install the [Gateway](https://docs.p0.dev/agentic-access/agentic-gateway/gateway.md), you can now choose a P0-hosted domain: P0 assigns the gateway a unique hostname, and once you submit your load balancer address, P0 creates the matching DNS record in its own zone. If you run your own domain instead, the installer now generates a ready-to-apply Terraform snippet for the gateway deployment and tells you which DNS record type to create. This option is in beta.
* **GitHub MCP Server (Preview).** Give AI agents policy-scoped access to GitHub through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), instead of handing them a personal access token. The predefined GitHub MCP server reads its token from a secret P0 provisions into your vault for each grant, and an agent requests access to a specific organization, repository, and permission level, for example `contents:read`. This server is in preview.

### Enhancements

#### Integrations

* **AI Gateway Management Tunnel (Beta).** Register a self-hosted [Gateway](https://docs.p0.dev/agentic-access/agentic-gateway/gateway.md) with P0 even when it has no public DNS record or TLS certificate. Every gateway deployment now runs an outbound-only tunnel client (`braekhus-client`); paste its public key into the console during install, and P0 routes configuration sync, entitlement, and activity-log calls, plus gateway JWKS verification, through the tunnel instead of requiring a direct connection. Registering the tunnel is required for all new gateway installs. Pre-existing installs keep working exactly as before. This capability is in beta.
* **VictoriaLogs Log Provider (Beta).** Read [AI Gateway](https://docs.p0.dev/agentic-access/agentic-gateway.md) tool-call activity from a VictoriaLogs instance deployed alongside your gateway, as an alternative to Google Cloud Logging. Choose **VictoriaLogs (via gateway)** when you configure the [Log provider](https://docs.p0.dev/agentic-access/agentic-gateway/log-provider.md) component. It needs no separate credential: P0 authenticates with the service account it already uses to manage the gateway.
* **Tool and Annotation Access Control for Custom MCP Servers (Preview).** Gate a [Custom MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/custom.md) by MCP annotation (read-only, destructive, or neither) or by individual tool name, instead of requiring the upstream server to enforce access through its own IAM. This closes the gap for a custom server that has no IAM of its own to scope. Both access types are in preview.
* **Gateway Install Without a Storage Class (Preview).** Deploy the [Gateway](https://docs.p0.dev/agentic-access/agentic-gateway/gateway.md) component on a Kubernetes cluster with no dynamic storage provisioner, for a proof-of-concept install, instead of the form requiring a storage class name. Choosing this option turns off the gateway's bundled PostgreSQL and VictoriaLogs persistence. Naming a storage class still installs exactly as before. This option is in preview.
* **Gateway Name Validation.** The [Gateway](https://docs.p0.dev/agentic-access/agentic-gateway/gateway.md) install form now rejects a name Kubernetes will not accept as you type it, instead of letting the install look successful and then failing later when you apply the generated Terraform. A gateway name must use lowercase letters, digits, and hyphens, start and end with a letter or digit, and be at most 53 characters.
* **Identity Provider Replay Protection (Preview).** Choose how an [Identity provider](https://docs.p0.dev/agentic-access/identity-provider.md) handles a repeated token: reject any repeat, allow every repeat, or allow a repeat within a grace period you set. The grace period option covers issuers that legitimately re-serve the same signed token, such as SPIRE or GCP service accounts.
* **Splunk Stack Cache Management.** The Splunk Stack integration installer now has a **Clear Cache** action for its cached resource lists, so you can force a refresh after deploying a new version of your Lambda handler instead of waiting up to 40 minutes for the cache to expire. P0 also stops caching an empty result from a misconfigured Lambda, so access requests recover as soon as the Lambda does.
* **Agentic Gateway Components Move to Beta.** The AI Gateway (see [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md)), MCP server, A2A bridge, and Log provider components move from preview to beta, along with the P0-hosted domain option and the VictoriaLogs log provider. The Agentic Access page's status tag now reads **Beta** to match. MCP tool and tool-annotation access, the container hosting type for custom servers, the OAuth and vaulted credential sources, Google Cloud Logging as a log provider, and installing the gateway without a storage class remain in preview.

#### User Experience

* **Identity Provider Issuer Allowlist Warning (Preview).** The [Identity provider](https://docs.p0.dev/agentic-access/identity-provider.md) install flow now shows the exact Terraform line needed to add a configured issuer to your gateway's allowlist, and warns if an installed gateway doesn't have that issuer yet. Previously nothing told you this second step existed, so a newly configured identity provider could silently never authenticate any agent.
* **Gateway Prerequisite Warning.** The MCP server and A2A bridge install forms now disable the **Gateway** field and explain that a [Gateway](https://docs.p0.dev/agentic-access/agentic-gateway/gateway.md) component is required, instead of showing an empty dropdown with no explanation, when no gateway is installed.
* **Event Log Time Range and Empty States.** The [Event log](https://docs.p0.dev/agentic-access/using-the-gateway/event-log.md) tab now offers a **Last 7 days** or **Custom range** selector instead of a bare, easy-to-misread date picker, and tells apart an empty result from a missing log provider. Previously the two looked identical, so you couldn't tell whether there were no calls to show or nothing was configured to read them.
* **Org Duration Floor Validation.** P0 now blocks lowering an organization's maximum or standing access duration below any existing policy's duration, instead of accepting the change and leaving that policy unable to grant its own configured duration. The error names the policy and approval rule that needs to change first. See the [Just-in-time settings API](https://docs.p0.dev/p0-management/management-api/just-in-time-settings-api.md).
* **Request Duration Capping.** Requesting a duration longer than your organization's maximum access duration no longer rejects the request outright. P0 caps it to the organization's maximum and creates the request instead. See the [Just-in-time settings API](https://docs.p0.dev/p0-management/management-api/just-in-time-settings-api.md).

#### API & CLI

* **Bulk Revocation API.** Revoke many access requests in a single call — either every currently-active request for a principal or an explicit list of request IDs — instead of revoking them one at a time. The new endpoint is restricted to owners and API keys, supports a `dryRun` preview, and is capped at 1,000 request IDs per call. See the [Access requests API](https://docs.p0.dev/access-management/access-management/just-in-time-access/just-in-time-api/access-requests-api.md).
* **`p0 claude mcp add` Failure Reporting.** [p0 claude mcp add](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-claude-mcp-add.md) now exits with an error when the underlying `claude mcp add` step fails, instead of reporting success and leaving you to find out later when the server fails to authenticate.
* **`--reason` in `p0 aws rds generate-db-auth-token`.** [p0 aws rds](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-aws-rds.md) now forwards `--reason` to the access request it creates. Organizations that require a reason for access previously had every RDS token request through this command rejected, since the flag was accepted but its value was never used.

## August 2026

### New Features

#### Integrations

* **Federated Agent Access Policies (Preview).** Write [Agentic access policies](https://docs.p0.dev/agentic-access/agentic-access-policies.md) that match AI agents by their identity provider, instead of registering and naming every agent individually. A new Federated rule (`type: provider`) matches agents that authenticate to the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md) by federating a token from an external provider, keyed on the provider ID and an optional subject pattern. Because federated agents act without a bound human user, the rule always pairs with "No user." This rule is in preview.
* **Automatic Agent Registration (Preview).** Enroll an external identity provider with the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md) and let P0 register matching agents on their first connection, instead of pre-registering every agent client by hand. The [Identity provider](https://docs.p0.dev/agentic-access/identity-provider.md) integration's **Dynamic registration** toggle controls whether P0 registers an agent automatically when it first presents a valid token, or requires the agent identity to be registered in advance. Every accepted request is still authorized against your access policies. This integration is in preview.
* **GCP Compute Engine Resource-Scoped Access (Preview).** Grant an agent least-privilege access to a single Compute Engine VM instead of the whole project. The [Compute Engine MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/gcp/compute.md) now supports resource-scoped access to one VM with a role such as `compute.instanceAdmin.v1`, alongside project-wide access with a role such as `compute.viewer` for tools that must span resources. This server is in preview.
* **Custom MCP Servers (Preview).** Put your own MCP server behind the [AI Gateway](https://docs.p0.dev/agentic-access/agentic-gateway.md) when P0 doesn't ship a predefined server for it, so agents reach it through the same policy checks and short-lived credentials as P0's built-in servers. A [Custom MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/custom.md) can authenticate to its upstream with a gateway-minted native credential rather than user-delegated OAuth, and gate each call through an installed, requestable P0 integration. Custom servers also accept Google Cloud Workload Identity Federation credentials, and you can add custom JWT claims to the native credential with Liquid templates when an upstream needs extra claims such as scopes. This capability is in preview.

### Enhancements

#### User Experience

* **AI Agent Request Attribution.** Access requests initiated by an AI agent now show the agent's identity in the web UI, so approvers can see which agent, and which human on its behalf, a request came from. The request details page groups the agent, gateway client, and delegation fields into distinct sections and shows the request reason.
* **Agentic Activity Tab.** The [AI Gateway](https://docs.p0.dev/agentic-access/agentic-gateway.md) Activity tab shows the caller beneath each gateway client ID, keeps the timestamp column fixed, and records a session ID for every agent tool call, so you can trace which agent identity performed which MCP action. You can also create a policy to block a specific MCP action directly from the activity log.
* **Agentic Access Policy Refinements (Preview).** [Agentic access policies](https://docs.p0.dev/agentic-access/agentic-access-policies.md) now evaluate agent rules against only the acting agent rather than the full delegation chain, so a rule matches based on the agent that actually makes the call. The `mcp-client` agent rule is now named `agent-client`. Existing policies that use `mcp-client` continue to work.
* **Policy Studio Installed Resources.** [Policy Studio](https://docs.p0.dev/policy-studio/policy-studio.md) now lists only the integrations you have installed when you build a policy, instead of every resource type P0 supports, so you no longer scroll past resources you cannot grant.

#### API & CLI

* **Revocation Attribution in Audit Logs.** The `permission-requests.revoked` audit event now includes a `revocationDetails` object naming who revoked a request, so agents and requestors can see who ended their access. This distinguishes a *revoke* (another user ends the request) from a *relinquish* (the recipient ends their own), and covers revocations made through the web app, Slack, Microsoft Teams, and email. See [Audit log format](https://docs.p0.dev/integrations/siem-integrations/audit-log-format.md).
* **Azure CLI SSH Extension Hint.** [p0 ssh](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-ssh.md) to an Azure VM now surfaces an actionable hint when the Azure CLI `ssh` extension is missing, telling you to run `az extension add --name ssh`, instead of failing with the raw Azure CLI error that made the P0 CLI look broken.

## July 2026

### New Features

#### Integrations

* **Google Cloud SQL for PostgreSQL.** Manage just-in-time access to Google Cloud SQL for PostgreSQL databases, closing the gap for teams whose PostgreSQL runs on Google Cloud rather than Amazon RDS. P0 provisions a dedicated IAM database user for each identity and grants the requested role during access. This connection option is now available in the UI without a feature flag. See [Installing a Cloud SQL instance](https://docs.p0.dev/integrations/resource-integrations/postgresql/installing-a-cloudsql-database.md).
* **Azure Custom Jump Host.** Reach Azure VMs for SSH access through your own existing jump host instead of Azure Bastion, so you can use P0 in environments that already standardize on a bastion VM. When you configure Azure SSH, you choose either an Azure Bastion or a jump host as the connection path, and the P0 CLI supports both. See [Custom jump host](https://docs.p0.dev/integrations/resource-integrations/microsoft-azure/configure-bastion-host-integration/custom-jump-host.md).
* **Azure Jump Host Management (Beta).** Send privileged commands, such as terminating a live SSH session, to jump host VMs in your Azure tenant through a P0-managed connector. Configure the optional [Jump host management](https://docs.p0.dev/integrations/resource-integrations/microsoft-azure/jump-host-management.md) component with the app registration P0 authenticates as and the Azure Function App that dispatches commands. This component is in beta.
* **GCP MCP Servers (Preview).** Give AI agents policy-scoped, credential-free access to Google Cloud APIs through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md), instead of handing agents long-lived service-account keys. P0 ships one predefined [GCP MCP server](https://docs.p0.dev/agentic-access/agentic-gateway/mcp-server/gcp.md) per service—Compute Engine, Cloud Storage, BigQuery, Cloud Monitoring, and IAM—so you install only the ones your agents need. Each call is authorized per session against your P0 access policy, and the gateway mints a short-lived token using Workload Identity Federation rather than issuing standing credentials. These servers are in preview.
* **GCP Workload Identity Federation (Preview).** Let agents reach Google Cloud through the [P0 Runtime Access for AI](https://docs.p0.dev/readme/agentic-control-plane.md) without any long-lived credentials by federating the gateway's identity into your project. Installing the [GCP Workload Identity Federation](https://docs.p0.dev/agentic-access/credential-sources/gcp-wif.md) component creates a Workload Identity Pool and OIDC provider that trust tokens signed by your gateway, so P0 grants and revokes short-lived, policy-scoped Google Cloud access per session. It builds on the base Google Cloud integration with IAM management and is the prerequisite for the GCP MCP servers. This component is in preview.

#### Posture

* **Azure IAM Assessment Monitors.** Surface unused, excessive, and unused-privileged Azure role assignments, bringing Azure to parity with the AWS and Google Cloud assessments. New monitors flag role assignments that go unused, grant more than is needed, or carry unused privileged access, and the Azure assessment now analyzes federated access risk. Install the assessment from [IAM assessment](https://docs.p0.dev/integrations/resource-integrations/microsoft-azure/iam-assessment.md). At launch these monitors report findings only; automated fix generators are not yet available, so remediation is manual.

#### API & CLI

* **Inventory Export API.** Pull your cloud [Access inventory](https://docs.p0.dev/inventory/access-inventory.md) programmatically over a new API instead of clicking **Export as JSON** in the app, so you can feed identities, grants, credentials, and resources into data warehouses, compliance workflows, or custom integrations on a schedule. List your P0 environments, then export a chosen environment's inventory as grouped JSON; exported findings now also carry risk scoring and timestamps. See the [Inventory export API](https://docs.p0.dev/inventory/inventory-export-api.md) reference.

### Enhancements

#### User Experience

* **Microsoft Entra ID PIM Eligibility in Inventory.** The [Access inventory](https://docs.p0.dev/inventory/access-inventory.md) grants table now shows an `Eligible` tag beside Microsoft Entra ID roles assigned through Privileged Identity Management, so you can tell eligible (PIM) assignments apart from active ones at a glance. Active assignments no longer add a redundant tag, and the full status remains available in TSV and JSON exports.

#### API & CLI

* **File Transfer Audit Logs.** Uploads, downloads, and deletions performed with [p0 file-transfer](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-file-transfer.md) now emit audit-log events, giving you a complete record of file movement for compliance. Each event records the request, file transfer, and bucket involved.
* **File Transfer Install Verifier.** Installing the [File transfer](https://docs.p0.dev/integrations/resource-integrations/file-transfer.md) integration now verifies the target S3 bucket and derives its region and partition automatically, removing manual configuration steps that were error-prone.

## June 2026

### New Features

#### Integrations

* **GitHub Without an Okta Directory.** The [GitHub](https://docs.p0.dev/integrations/resource-integrations/github.md) integration no longer requires Okta. P0 now identifies GitHub users through a verified organization email domain, so teams on any identity provider can manage GitHub team access with P0.

### Enhancements

#### Documentation

* **Access Inventory and Posture Getting-Started Guides.** New guides walk you through [Explore who has access to what with Access Inventory](https://docs.p0.dev/getting-started/getting-started-with-access-inventory.md) and [Find risky access with Posture](https://docs.p0.dev/getting-started/getting-started-with-posture.md), including a beginner's guide to the [Query language basics](https://docs.p0.dev/inventory/query-search/query-language-basics.md) and a how-to for [Create a custom monitor](https://docs.p0.dev/posture/create-a-custom-monitor.md).
* **IAM Assessment Install Guides.** New install guides cover the IAM assessment: [IAM assessment](https://docs.p0.dev/integrations/resource-integrations/aws/iam-assessment.md), [IAM assessment](https://docs.p0.dev/integrations/resource-integrations/google-cloud/iam-assessment.md), and [IAM assessment](https://docs.p0.dev/integrations/resource-integrations/microsoft-azure/iam-assessment.md).
* **Microsoft Entra ID Security Perimeter Logs.** A new guide explains how to view security perimeter logs for Microsoft Entra ID. See [Viewing security perimeter logs](https://docs.p0.dev/integrations/directory-integrations/microsoft-entra-id/viewing-security-perimeter-logs.md).

#### API & CLI

* **Automatic Google Cloud Login.** Connect to a Google Cloud target with [p0 ssh](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-ssh.md) without running `gcloud auth login` first. When your gcloud credentials are missing or expired, P0 runs the login automatically, matching the existing AWS and Azure behavior instead of failing partway with a manual login hint.
* **Fish Shell Support.** Use the credential output from [p0 aws role assume](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-aws-role-assume.md), `permission-set`, and `rds` in the fish shell. These commands previously printed only bash and zsh syntax, so fish users ended up with unset credentials and failing AWS calls; the CLI now detects fish and prints the matching syntax.
* **Clearer GCP SSH Errors.** `p0 ssh` now explains that a rejected Google Cloud connection is most often caused by OS Login not being enabled, instead of a generic "Permission denied (publickey)" error. The message tells you how to enable OS Login and notes that a just-granted role can take a moment to propagate. See [p0 ssh](https://docs.p0.dev/p0-cli/troubleshooting/p0-ssh.md).
* **File Transfer Improvements.** File transfer reuses your existing SSH request instead of opening a second one, removes the uploaded object from the S3 bucket once the download completes, and refreshes AWS credentials mid-transfer so long uploads no longer fail when temporary credentials expire.
* **MCP Server OAuth Registration.** `p0 claude mcp add` now registers MCP servers using OAuth, so the servers it adds to Claude Code authenticate through the standard OAuth flow.

### Stability & Fixes

* **API Key Creation.** Fixed an error that prevented creating an API key. See [Generating an API key](https://docs.p0.dev/p0-management/generating-an-api-key.md).
* **GCP SSH Group Resolution.** Fixed group resolution so access targets the correct group when multiple Google Cloud SSH groups are installed.

## May 2026

### New Features

#### API & CLI Enhancements

* **File Transfer Command.** New `p0 file-transfer` command uploads files directly to S3 with multipart support, progress tracking, and presigned download URLs. Supports arbitrarily large files with automatic retry on transient failures.
* **AI Gateway integration.** New `p0 claude mcp add` command registers MCP servers with Claude Code, enabling AI-assisted access management through the P0 AI Gateway.

#### Security & Compliance Workflows

* **Agentic Access Policies.** [Agentic access policies](https://docs.p0.dev/agentic-access/agentic-access-policies.md) can now match on the AI agent or MCP client that initiated a request, enabling fine-grained rules based on agent identity and agent owner group.
* **On-Call Schedule Selection for Auto-Approval.** Auto-approval rules for [Incident.io](https://docs.p0.dev/integrations/approval-integrations/incidentio.md) now let you select specific on-call schedules, matching the existing [PagerDuty](https://docs.p0.dev/integrations/approval-integrations/pagerduty.md) behavior for more granular control.

#### Cloud Platform Integrations

* **RDS PostgreSQL Generally Available.** The Amazon RDS [PostgreSQL](https://docs.p0.dev/integrations/resource-integrations/postgresql.md) integration is now generally available.
* **Azure Bastion SSH Generally Available.** Azure Bastion SSH is now generally available, with session audit logging for tracking active connections. See [Configure bastion host integration](https://docs.p0.dev/integrations/resource-integrations/microsoft-azure/configure-bastion-host-integration.md).

### Enhancements

#### User Experience

* **Policy Studio Redesign.** Compact pill-style selectors replace vertical radio groups so the entire policy definition fits on one screen in [Policy Studio](https://docs.p0.dev/policy-studio/policy-studio.md), with per-section status icons for better readability.
* **Compact Access Request Lists.** Access request lists display with reduced spacing, making it easier to scan multiple requests at once.
* **Session Timeout Handling.** The "Stay Logged In" button hides when your identity provider caps the session length, and dismissing the session warning now correctly resets the session timer.

#### Reliability & Performance

* **Improved AWS IDC Error Messages.** When a permission-set assignment fails because the target role exceeds the managed-policy quota, P0 now explains the cause and suggests remediation steps.
* **Automatic Connector Install Retry.** Database connector installation retries IAM permission validation automatically, eliminating manual retries during initial setup.

#### API & CLI Enhancements

* **AWS GovCloud SAML Support.** The CLI now supports SAML assertions for AWS GovCloud (`aws-us-gov`) partitions, enabling [p0 aws role assume](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-aws-role-assume.md) in regulated environments.

### Stability & Fixes

* **Web Request Modal.** Fixed an error when changing the resource field after selecting an access type in the [Web request modal](https://docs.p0.dev/access-management/access-management/just-in-time-access/requesting-access/web-request-modal.md).
* **Auto-Approval Expiry.** When multiple auto-approval rules apply, P0 now selects the shortest expiry timestamp to ensure least-privilege durations.
* **`p0 ssh` Polling.** Fixed a URL construction issue that caused [p0 ssh](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-ssh.md) polling requests to fail with 404 errors.
* **CLI Error Handling.** Fixed an error-handling path that could crash during Okta login when reading an HTTP response body twice.

## April 2026

### New Features

#### Integrations & Approvals

* **Incident.io On-Call Approvals & Escalation.** Users on-call in [Incident.io](https://docs.p0.dev/integrations/approval-integrations/incidentio.md) can now approve access requests. Requestors can also escalate by paging the on-call responder directly from P0.

#### Database Access

* **New PostgreSQL Integration.** A redesigned, connector-based [PostgreSQL](https://docs.p0.dev/integrations/resource-integrations/postgresql.md) integration for AWS RDS. Install, list databases, and manage just-in-time access with improved reliability and setup.

#### Access Management

* **Custom Durations for Auto-Approved Requests.** Auto-approved requests now honor the requested duration instead of defaulting to one hour.
* **Configurable Pending Request Timeout.** Administrators can set how long access requests remain pending before automatic denial, with a configurable default. See [Requesting access](https://docs.p0.dev/access-management/access-management/just-in-time-access/requesting-access.md).
* **GCP and Azure SSH Parent Requests.** Request [SSH](https://docs.p0.dev/integrations/resource-integrations/ssh.md) access to all nodes in a GCP project or Azure subscription using `p0 request ssh parent`.

### Enhancements

#### Cloud Platform Integrations

* **Simplified Azure SSH Setup.** Azure SSH no longer requires the full Entra ID directory integration, and the interactive admin consent flow has been removed, significantly reducing installation complexity. See [Install SSH access](https://docs.p0.dev/integrations/resource-integrations/microsoft-azure/install-ssh-access.md).
* **Okta Group Assignment Now GA.** [Okta](https://docs.p0.dev/integrations/directory-integrations/okta.md) group assignment is out of beta and available.
* **Google Cloud Labels for Grouping.** Google Cloud labels are now used as a fallback for grouping tags when organizing resources.
* **AWS Identity Center Email Matching.** P0 now matches on Identity Center user email in addition to username, resolving user-not-found errors for customers whose IDC usernames differ from their email.

#### User Experience

* **Request Modal Improvements.** [Slack](https://docs.p0.dev/integrations/notifier-integrations/slack.md) and web request modals now pre-populate dropdown options on focus, and the system preserves form values during slow responses.
* **Integration Labels on Notifications.** Notifications and request details now display which cloud platform (AWS, GCP, Azure) a request targets, making multi-cloud approvals clearer.
* **Pagination for Active Requests.** Active and pre-approval request pages now support pagination with configurable page sizes. See [Pre-approving access](https://docs.p0.dev/access-management/access-management/just-in-time-access/approving-access/pre-approving-access.md).
* **Faster GCP Resource Listing.** Listing GCP resources in Slack, CLI, and the web UI is significantly faster for large environments.

#### CLI Improvements

* **Google Account Selection.** `p0 login` with Google Workspace now forces account selection, preventing accidental login with the wrong account.
* **Multi-Org SSH Configuration.** [p0 ssh](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-ssh.md) proxy now supports the `P0_ORG` environment variable for multi-organization SSH configurations via `~/.ssh/config`.
* **SSH Debug Commands.** `p0 ssh` now displays reproducible CLI commands for all providers (AWS, GCP, Azure) to simplify troubleshooting.

### Stability & Fixes

* **CLI Network Resilience.** The CLI now retries polling for request status on network interruption without creating duplicate requests.
* **CLI Pipe-Friendly Output.** Spinner output from [p0 ls](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-ls.md) and `p0 request` is suppressed when piped to other tools.
* **`--size` with `--json` Fix.** The `--size` flag now correctly limits results when combined with `--json`.
* **Access Request Hang Fix.** Fixed an issue where access requests would hang indefinitely while waiting for approval.
* **SSH Pre-Approved Sudo Fallback.** Fixed [p0 ssh](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-ssh.md) failing to fall back to non-sudo access when sudo was not pre-approved.
* **AWS SSH from New Machine.** Fixed [SSH](https://docs.p0.dev/integrations/resource-integrations/ssh.md) access failing when connecting from a new machine or after key regeneration.
* **Pre-Approval Date Validation.** Fixed date validation errors that occurred in certain time zones when creating pre-approvals.

## March 2026

### New Features

#### On-Premises Access

* **Linux Connector for SSH.**\
  SSH access to on-premises Linux machines now uses a lightweight connector instead of requiring the P0 agent on each target host. Includes session recording and security assessment support.
* **On-Premises RDP Access.**\
  Request and provision just-in-time access to Windows machines via Remote Desktop Protocol, including session recording and playback.
* **`p0 sudo` Command.**\
  New CLI command to request just-in-time sudo access and run commands as another user on Linux machines.
* **Windows Admin Integration.**\
  Manage Windows user accounts and passwords through P0, with Active Directory user listing and password reset lifecycle.

#### Integrations

* [Cisco Secure Access](https://docs.p0.dev/integrations/resource-integrations/cisco-secure-access.md) **Integration.**\
  JIT network routing for SSH through Cisco Secure Access.
* **Zscaler Private Access Integration.**\
  JIT SSH network routing through Zscaler Private Access.
* **incident.io Auto-Approval.**\
  Access policies can now use incident.io on-call schedules as automatic approvers for emergency access workflows.
* **Webex Notifier.**\
  Send access request notifications and approvals through Webex.
* **Splunk Stack Access Integration.**\
  Manage access to Splunk Stack environments with manifest-based lifecycle and typed Lambda API.

#### Session Recording & Audit

* **Unified Terminal Replay.**\
  Session recordings across all integrations now use a terminal replay player with full ANSI color support, playback controls, and search.
* **Splunk Session Audit.**\
  Read SSH session recordings directly from Splunk for centralized audit trail.
* **Self-Hosted SSH Session Recording.**\
  View SSH session evidence including terminal replay and audit trail for [Self-hosted](https://docs.p0.dev/integrations/resource-integrations/ssh/self-hosted.md) instances.

#### Vault & Secrets

* **Secrets Vault Viewer.**\
  New vault page in the UI to view and manage secrets created by P0, with filtering, caching, and management tags.
* **GCP Secret Manager Access.**\
  Request access to GCP Secret Manager secrets via resource requests, with project-level IAM policies. See [Google Cloud](https://docs.p0.dev/integrations/resource-integrations/google-cloud.md).

#### Security & Compliance

* **SSH Security Assessment.**\
  Assess SSH security posture across AWS SSM and self-hosted instances, with risk findings.
* **Workload Identity Federation (WIF) Support.**\
  WIF identities can authenticate, request access, and manage policies through Policy Studio.
* **SSH Risk Findings in Assessment.**\
  SSH security assessments now include actionable risk findings.
* **Self-Hosted SSH Assessment.**\
  Run security assessments against self-hosted SSH instances.

#### Cloud Platform Integrations

* **RDP Proxy Access in CLI.**\
  New RDP proxy access support in the P0 CLI for remote desktop connections. See [p0 commands and usage](https://docs.p0.dev/p0-cli/p0-commands-and-usage.md).

### Enhancements

#### Search & Navigation

* **Regex Search in Inventory and Monitors.**\
  Use regex patterns (`/pattern/`) in the [Query search](https://docs.p0.dev/inventory/query-search.md) bar for inventory and monitor queries.
* **SSH Targeting by FQDN and Public IP.**\
  Target GCP, Azure, and AWS SSH instances by FQDN, public IP, or custom tags instead of instance ID. See [Google Cloud](https://docs.p0.dev/integrations/resource-integrations/google-cloud.md) and [Install SSH access](https://docs.p0.dev/integrations/resource-integrations/microsoft-azure/install-ssh-access.md).
* **Hostname-Based Access Requests.**\
  Users can request SSH access by hostname instead of requiring a UUID.
* **Linux User Listing.**\
  List users on Linux machines directly from the P0 UI for access requests.

#### Authentication & Access

* **Self-Hosted SSH Key Authentication.**\
  Self-hosted SSH switches from certificate signing to public key push for broader compatibility.
* **Sudo Access via Groups.**\
  Sudo access provisioning now uses groups, removing the need to start a new shell session after access is granted.
* **Azure Bastion Premium SKU.**\
  Azure Bastion integration now supports premium SKU. See [Configure bastion host integration](https://docs.p0.dev/integrations/resource-integrations/microsoft-azure/configure-bastion-host-integration.md).
* [p0 ssh](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-ssh.md) **Retry on Unknown Port Errors.**\
  Automatically retries connection failures during self-hosted SSH access propagation.

#### Reliability & Performance

* [PostgreSQL](https://docs.p0.dev/integrations/resource-integrations/postgresql.md) **Duplicate Provisioning.**\
  PostgreSQL no longer errors when provisioning access to a user who already has it.
* **Improved CLI Network Resilience.**\
  The CLI retries transient network errors with exponential backoff instead of failing immediately.

#### User Experience

* **Dashboard Empty State.**\
  The posture dashboard panel shows clear empty results instead of an infinite loading state when no findings exist.
* **Removed Stale Announcement Banners.**\
  Cleans up outdated Policy Studio announcement banners from the UI.
* **Fixed Cisco in SSH Installer.**\
  [Cisco Secure Access](https://docs.p0.dev/integrations/resource-integrations/cisco-secure-access.md) now appears correctly in the SSH network access provider list during installation.

#### Stability & Fixes

* **Slack OAuth Token Handling.**\
  Fixes stale [Slack](https://docs.p0.dev/integrations/notifier-integrations/slack.md) OAuth tokens that accumulated errors silently.
* **RDP Session Recordings.**\
  Fixes RDP session recordings not appearing in production.
* **`p0 scp` Error Messages.**\
  Suppresses spurious "SSH session terminated" messages for [p0 scp](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-scp.md) and surfaces missing file and permission errors.
* **GCP Access Conditions.**\
  Fixes condition chaining for GCP access requests across multiple grants.
* **Terraform Null Field Crash.**\
  Fixes a crash when optional Terraform provider fields are set to null.
* **Integration Error Surfacing.**\
  Integration errors from Cloud Run now surface to end-users with actionable messages instead of failing silently.

***

## February 2026

### New Features

#### SIEM Integrations

* **Datadog Audit Log Integration.**\
  Send P0 audit logs directly to Datadog with a self-service installer in the UI. See [Datadog setup](https://docs.p0.dev/integrations/siem-integrations/datadog-setup.md).
* **S3 Audit Log Archival.**\
  Automatically archive audit logs to an Amazon S3 bucket on a 15-minute schedule.

#### Cloud Platform Integrations

* [AWS RDS](https://docs.p0.dev/integrations/resource-integrations/aws/managed-services/aws-rds.md) **Token Generation.**\
  New `p0 aws rds generate-db-auth-token` CLI command to generate database authentication tokens using P0 grants.
* [AWS RDS](https://docs.p0.dev/integrations/resource-integrations/aws/managed-services/aws-rds.md) **Terraform Install.**\
  Set up AWS RDS integration via Terraform directly from the web UI installer.
* [MySQL](https://docs.p0.dev/integrations/resource-integrations/mysql.md) **and RDS Terraform Installers.**\
  Set up MySQL and AWS RDS integrations via Terraform directly from the web UI.
* **GCP Security Perimeter Region Configuration.**\
  GCP security perimeter Cloud Run deployments are no longer hard-coded to `us-west1`; specify your preferred region during installation. See [Security perimeter](https://docs.p0.dev/integrations/resource-integrations/google-cloud/security-perimeter.md).
* **SSH Alternative Name Targeting.**\
  SSH nodes can be targeted via hostname, FQDN, instance ID, or custom tags using `p0 ls` and `p0 ssh`. See [p0 ssh](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-ssh.md).

#### Authentication & Access

* [Okta](https://docs.p0.dev/integrations/directory-integrations/okta.md) **Groups as Security Reviewers.**\
  Assign Okta groups as Security Reviewers in P0 Management, not individual users only.
* **AWS IDC Delegated Admin Accounts.**\
  Manage access in AWS Identity Center environments using delegated admin accounts.
* **SSH Host Key Caching.**\
  The CLI caches SSH host keys locally for faster connections and detects host key mismatches with guided recovery.
* **SSH Password Authentication Disabled.**\
  The CLI now enforces public key authentication for [p0 ssh](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-ssh.md) and [p0 scp](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-scp.md), preventing fallback to less secure methods.

#### User Experience

* **Refreshed Navigation.**\
  We redesigned the sidebar navigation with a light theme and outlined icons for a cleaner look.
* **Integration Documentation in UI.**\
  Integration pages now display contextual documentation, descriptions, prerequisites, and use-case guidance directly in the setup flow.

### Enhancements

#### User Experience

* **Audit Logs for Request Creation.**\
  The `api.jit.permission-requests.created` event now fires for all request sources: UI, CLI, Slack modal, and slash command.
* **MySQL Connectivity Instructions.**\
  [Slack](https://docs.p0.dev/integrations/notifier-integrations/slack.md) notifications for approved MySQL/RDS access now include connection instructions.
* **AWS Permission Set Validation.**\
  Requests for AWS Identity Center permission sets with customer-managed policies are validated at submission time.

#### Authentication & Access

* [Kubernetes](https://docs.p0.dev/integrations/resource-integrations/kubernetes.md) **EKS Auto-Mode Support.**\
  The Kubernetes installer now supports EKS clusters with Auto-Mode enabled.

#### Stability & Fixes

* [Microsoft Entra ID](https://docs.p0.dev/integrations/directory-integrations/microsoft-entra-id.md) **Group Caching.**\
  Fixes a bug where Entra ID groups were not listed in the Slack access request modal.
* **Device Code Authentication.**\
  Fixes the device code login flow that was always showing "User Not Logged In."
* **Database Connector Error Messages.**\
  Database connector commands now provide actionable error messages and suggestions instead of misleading privilege errors.
* [p0 kubeconfig](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-kubeconfig.md) **Windows Fix.**\
  Fixes `p0 kubeconfig` to use OS-safe command execution on Windows.
* **Slack Pre-Approval Validation.**\
  Fixes a bug where re-requesting access from [Slack](https://docs.p0.dev/integrations/notifier-integrations/slack.md) bypassed pre-approval validation.
* **GCP API Error Handling.**\
  Fixes error handling in GCP API calls that caused issues for Google Cloud requests.
* **Magic Link Login.**\
  Fixes a 500 error when clicking the email magic link for newly provisioned organizations.
* **Okta Token Exchange Errors.**\
  Surfaces specific [Okta](https://docs.p0.dev/integrations/directory-integrations/okta.md) token exchange errors instead of a generic "session expired" message.
* [p0 scp](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-scp.md) **Error Surfacing.**\
  Surfaces missing file/directory and permission denied errors during `p0 scp` operations.
* [p0 kubeconfig](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-kubeconfig.md) **Bug Fix.**\
  Fixes incorrect parameter ordering in `p0 request k8s resource` generated by `p0 kubeconfig`.

***

## January 2026

### New Features

#### Core Platform Capabilities

* **Policy Studio.**\
  New [Policy Studio](https://docs.p0.dev/policy-studio/policy-studio.md) tab consolidates access policies and approval policies into a single management view.
* [Microsoft Entra ID](https://docs.p0.dev/integrations/directory-integrations/microsoft-entra-id.md) **Manager Approvals.**\
  Approval workflows can now route to the requestor's manager via Entra ID manager lookup.
* **Approvers Can Revoke Pre-Approvals.**\
  Designated approvers, not only admins, can now revoke standing pre-approvals. See [Pre-approving access](https://docs.p0.dev/access-management/access-management/just-in-time-access/approving-access/pre-approving-access.md).
* **In-App Announcement Banner.**\
  New in-app banner for surfacing product updates and important notices to users.
* **AWS SSH IP Address Search.**\
  Search for SSH targets by IP address in the AWS integration.
* **GCP Role Removal Suggestions.**\
  Posture management now surfaces suggestions to remove unused GCP roles.
* **P0 Tags on Permission Sets.**\
  Permission sets display P0 metadata tags for better visibility into managed resources.

#### API & CLI Enhancements

* [p0 ls](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-ls.md) **Filtering.**\
  Filter results by resource type or name with the `p0 ls` command.
* **CLI Version Check Messaging.**\
  The CLI notifies you about available updates and can display custom messages from P0 (security advisories, deprecation notices).
* **`--request=false` for Role Assumption.**\
  New flag for [p0 aws role assume](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-aws-role-assume.md) to bypass the access request check, useful for debugging federated SSH.
* **`--debug` Flag for AWS/SSM Validation.**\
  Validates AWS CLI and SSM plugin installations to help troubleshoot SSH connection issues.
* **Sudo Access by Default (Environment Variable).**\
  New environment variable to control whether [p0 ssh](https://docs.p0.dev/p0-cli/p0-commands-and-usage/p0-ssh.md) requests sudo access by default.
* **Two-Character SSH Group Search.**\
  Lowered the minimum search length to two characters when searching SSH group targets.
* **Date Column in Request History.**\
  Request history table now shows a visible date column.

### Enhancements

#### User Experience

* **Activity Page Search and Filtering.**\
  Search and filter controls on the activity and audit log pages in the web UI.
* **Configurable Page Sizes.**\
  Page-size selectors on pre-approvals and activity log pages for viewing more entries at once.
* **AWS Account Filtering in Access Policies.**\
  Filter access policies by specific AWS accounts. See [AWS filtering](https://docs.p0.dev/policy-studio/policy-studio/aws-filtering.md).

#### Notifications & Integrations

* [Slack](https://docs.p0.dev/integrations/notifier-integrations/slack.md) **Re-Request Duration Selector.**\
  Choose a different time window when re-requesting access from a Slack notification.

#### Reliability & Performance

* **AWS SSH Node Refresh Interval.**\
  Configurable refresh interval for AWS node discovery, improving SSH target visibility timing.
* **Concurrent SSH Grant/Revoke.**\
  Fixes reliability issues when granting and revoking SSH access simultaneously on the same host.
* **AWS IDC Credential Retries.**\
  Increases retries for AWS credential fetching with a user-friendly transient error message.

#### Stability & Fixes

* **EKS ARN Fix for us-east-1.**\
  Corrects ARN formatting for EKS clusters in `us-east-1` that prevented access requests. See [Kubernetes](https://docs.p0.dev/integrations/resource-integrations/kubernetes.md).
* **EKS Auto-Mode Installer Warning.**\
  Warns during Kubernetes setup if the cluster uses Auto-Mode, which requires extra configuration.
* **Slack Modal Timeout.**\
  Slack modals no longer time out during access requests; heavy processing runs in the background.
* **CLI Login Page Text.**\
  Fixes login success page to not show a premature success message before login completes.
* [Snowflake](https://docs.p0.dev/integrations/resource-integrations/snowflake.md) **Installer.**\
  Resolves an error in the Snowflake integration installer that blocked initial setup.
* [Okta](https://docs.p0.dev/integrations/directory-integrations/okta.md) **Session Expiration.**\
  Improves behavior when an Okta session expires, prompting re-authentication instead of showing an error.
* **Kubernetes Permission Display.**\
  Fixes incorrect permission display for certain Kubernetes RBAC configurations.
* **Web Modal Validation.**\
  Surfaces field-level validation errors in access request modals instead of silently failing.
