Anthropic Compliance API: Auditing Security Posture and Controls
Learn how the 2026-11-05 Claude Platform update adds security posture audits, Activity Feed tracking, IP allowlists, and remote session revocation.
- Anthropic added Security posture data in beta to the Compliance API on 2026-11-05 for Claude Enterprise accounts.
- Changes to attestations on the Security posture page (Security Center) stream directly into the Compliance Activity Feed.
- A dedicated endpoint exposes point-in-time ratings across automatic, manual, and information-only controls.
- Complementary governance features introduced on the same date include Claude Console IP allowlisting and Admin API global sign-out.
Security Posture and Governance Updates in the 2026-11-05 Release
On 2026-11-05, Anthropic introduced administrative governance capabilities across the Claude Platform release notes, documented at the release notes overview. The primary update exposes Security posture data inside the Compliance API in beta for Claude Enterprise organizations. This feature allows programmatic ingestion of security evaluations directly into security information and event management (SIEM) platforms, complementing earlier exports reviewed in our guide to the Anthropic Compliance API unified chats and exports.
The release also launched two companion safeguards in beta: organization-level IP allowlisting inside the Claude Console to prevent stolen credentials from operating outside approved CIDR boundaries, and a member revocation endpoint in the Admin API that signs a specified user out of every active Claude session. Together, these updates shift Claude Enterprise governance from manual dashboard inspections to automated, audit-ready compliance workflows.
| Capability | Availability | Interface | Primary Use Case |
|---|---|---|---|
| Security Posture Results | Beta (Claude Enterprise) | Compliance API endpoint | Fetch active ratings across controls for compliance reporting |
| Security Posture Events | Beta (Claude Enterprise) | Compliance Activity Feed | Stream attestation alterations and expiration events to SIEM |
| IP Allowlisting | Beta (Console Organizations) | Claude Console settings | Restrict API key usage to authorized IPv4 and IPv6 CIDR blocks |
| Sign Member Out Everywhere | Beta (Claude Enterprise) | Admin API endpoint | Terminate all active user sessions during access reviews or offboarding |
Evaluating Controls, Attestations, and Ratings
The Security posture screen inside claude.ai Organization settings—also headed as the Security Center—benchmarks an enterprise organization’s active configurations against Anthropic’s baseline recommendations. As detailed in the Security posture overview, the engine measures compliance across three distinct control classifications:
- Automatic controls: Evaluated programmatically by Anthropic by contrasting current organization parameters against recommended target values, exemplified by sso_enabled.
- Manual controls: Assessed solely through human declaration because Anthropic cannot read configurations maintained in external vendor consoles, exemplified by oauth_conditional_access.
- Information-only controls: Reported without rating benchmarks or desired states, such as allowed_invite_domains.
Authorized administrators submit attestations directly within the user interface to state their operational standing. An attestation can register as completed, accepted_risk, or control_failed. An attestation remains active until modified, until an automatic check passes, until underlying values change, or until reaching an optional review_date duration of 90 days or none. Each evaluation ultimately yields one of four ratings:
| Rating Code | Status Description | Determining Factor |
|---|---|---|
| passed | Passed | Configuration satisfies desired state or carries active completed or accepted_risk attestation |
| failed | Failed | Configuration does not meet recommendation, or attestation records control_failed |
| not_attested | Not Attested | Manual control lacks an active attestation from an authorized administrator |
| descriptive | Descriptive | Information-only setting, unrated control, or a check that could not be evaluated |
Ingesting Security Posture Changes Through the Activity Feed
Every update to a Security posture attestation produces an event inside the unified Compliance Activity Feed. The feed structures operational history in reverse chronological order. Standard activities typically become queryable within 1 minute of occurrence, though security posture events can take longer to appear in feed queries. Data is retained for 6 years, but event collection begins only after an organization activates the Compliance API without backfilling prior administrative history.
Clients poll the activity log via HTTP GET requests using cursors. The cursor parameter sets the paging direction while endpoint sorting determines chronological ordering. Passing last_id into after_id steps backward to inspect older events, while before_id using first_id fetches newer occurrences. Responses cap results at limit values between the default 100 and a maximum of 5,000 items.
curl --fail-with-body -sS \
"https://api.anthropic.com/v1/compliance/activities?limit=1" \
-H "x-api-key: $ANTHROPIC_COMPLIANCE_ACCESS_KEY" \
-H "anthropic-version: 2026-10-21"
Each activity record contains actor union details including user_actor, api_actor, admin_api_key_actor, unauthenticated_user_actor, anthropic_actor, system_actor, or scim_directory_sync_actor. Attestation events logged by individuals include user_id, email_address, ip_address, and user_agent metadata. Feed events record attestation states but omit attestation comments, which can hold up to 2,000 characters in the web console.
Querying Current Results with the Security Posture Endpoint
While the Activity Feed captures delta changes over time, security teams requiring point-in-time posture assessments can query the dedicated endpoint: GET /v1/compliance/organizations/{organization_id}/security_posture. This resource delivers current control evaluations across linked organizations without historical series.
Each entry returned by the endpoint exposes the control id, display_name, and current evaluation rating. For administrators cross-checking configurations against corporate governance baselines, this call validates whether automatic checks or accepted risk exceptions currently satisfy internal standards.
curl --fail-with-body -sS \
"https://api.anthropic.com/v1/compliance/organizations/org_01Wv6QeBcDfGhJkLmNpQrSt8/security_posture" \
-H "x-api-key: $ANTHROPIC_COMPLIANCE_ACCESS_KEY" \
-H "anthropic-version: 2026-10-21"
Calling this endpoint returns a 404 response if the targeted organization is a standalone Claude Console account, if the organization operates outside the Claude Enterprise tier, or if the Security posture view has been disabled in organizational settings. When checking posture programmatically, systems should account for situations where automatic controls attested as completed still return failed if the underlying automatic verification fails.
Key Types, Access Scopes, and Rate Limits
Accessing governance data requires matching API keys and permission scopes. As outlined in the Compliance API reference, two primary credential types unlock administrative resources: Compliance Access Keys starting with sk-ant-api01- created within claude.ai, and Admin API keys starting with sk-ant-admin01- provisioned through the Claude Console.
- read:compliance_activities or read:org_audit: Required to read the Activity Feed, including posture attestation events. Available to both Compliance Access Keys and Admin API keys.
- read:compliance_org_data or read:org_audit: Required to query current posture state via the GET security_posture endpoint.
- read:compliance_user_data: Combined with read:compliance_org_data to reveal the user ID inside the attested_by field in current posture results.
All endpoints under /v1/compliance/* share a global rate limit of 600 requests per minute per parent organization, or per organization for standalone console setups. Integrations must include the mandatory anthropic-version: 2026-10-21 header. If rate limits are exceeded, the API returns a 429 Too Many Requests response.
Organizations managing large inference footprints alongside administrative calls can estimate overall platform expenses using our LLM API cost calculator or compare operational pricing structures across Claude models on our LLM API pricing tool.
Companion Controls: IP Allowlisting and Everywhere Sign-Out
The 2026-11-05 platform release also deployed network perimeter protections and identity remediation tools, detailed further in our IP allowlisting and session revocation setup guide. The Claude Console IP allowlist feature allows administrators to limit API traffic strictly to verified company networks, blocking compromised keys from operating from unauthorized IP addresses.
- Allowlist Capacity: Supports IPv4 and IPv6 individual addresses as well as CIDR notation ranges, accepting up to 1,000 entries per organization.
- Enforcement Scope: Unlisted IP addresses attempting requests against covered endpoints receive immediate rejections.
- Admin API Everywhere Sign-Out: Enables administrators to programmatically sign a member out everywhere across Claude Enterprise surfaces.
- Identity State: Revoked members remain in the organization directory and retain the ability to authenticate again through designated enterprise identity providers.
Organizations operating sensitive data workloads should note that customer-managed encryption keys (CMEK), zero data retention (ZDR) configurations, and HIPAA-ready Enterprise plans do not truncate attestation metadata retention or encrypt attestation records, as detailed in the documentation for API data retention and training policies.
Implementation and Audit Checklist
To integrate the 2026-11-05 governance capabilities into existing SIEM collectors and operations workflows, complete the following implementation steps:
- Verify Subscription Tier: Ensure the organization runs Claude Enterprise under a parent organization structure, as standalone console accounts receive 404 on posture endpoints.
- Provision Scoped Credentials: Create a Compliance Access Key with read:compliance_activities, read:compliance_org_data, and read:compliance_user_data scopes.
- Establish SIEM Polling: Implement an automated cursor loop storing last_id from the Compliance Activity Feed to track incoming attestation changes over time.
- Configure Allowlist Ranges: Register known build servers and gateway egress subnets in Claude Console IP allowlisting, keeping within the 1,000 range allowance.
- Automate Revocation Triggers: Connect security incident runbooks to the Admin API member sign-out endpoint to terminate sessions across claude.ai during credential triage.