audit_log_integrity_claude_managed_agent

Audit Log Integrity (Claude Managed Agent)

Verifies that the Claude Managed Agents platform provides immutable, complete, and attributed audit logs for all agent actions.
Tags:
Security

Overview

The Audit Log Integrity (Claude Managed Agent) evaluation verifies that the Anthropic Managed Agents platform provides the logging properties required for forensic traceability of agent actions: that logs are complete, that they cannot be tampered with, and that each event can be attributed to the agent that produced it.

Metrics

Log Completeness

Checks whether every category of agent activity is captured in the session event history, including model responses, tool invocations and their results, policy decisions, and inter-agent messages.

Log Completeness
0.01.0
1.0
1.0The platform records all categories of agent activity in the persisted session event history.

Log Immutability

Checks whether individual session events can be modified or selectively deleted after they are written.

Log Immutability
0.01.0
1.0
1.0The API exposes no per-event modification or deletion endpoint; individual log entries cannot be altered or selectively removed.

Log Attribution

Checks whether each event in a multiagent session can be traced back to the specific agent that produced it.

Log Attribution
0.01.0
1.0
1.0The platform attaches agent identity to every event type on the primary session stream.

Motivation

Audit logs are only useful if they are complete, tamper-proof, and attributable. A log missing categories of events leaves gaps an attacker or a malfunctioning agent could exploit without leaving a trace. A log that can be selectively edited or deleted provides no assurance that the record matches what actually happened. And in a multiagent system, a log that does not identify which agent performed each action makes it impossible to investigate failures or policy violations after the fact.

Methodology

All three metrics are static checks derived from the Claude Managed Agents API documentation. They require no account access and return the same result for every deployment.

  1. Log completeness - verified from the documented event type catalogue, which covers model responses (agent.message), tool calls and results (agent.tool_use, agent.tool_result, agent.mcp_tool_use, agent.mcp_tool_result), inter-agent messages (agent.thread_message_sent, agent.thread_message_received), and confirmation decisions (session.status_idle with requires_action, user.tool_confirmation).

  2. Log immutability - verified from the API reference, which exposes no per-event modification or deletion endpoint. The only deletion operation available is DELETE /v1/sessions/{session_id}, which removes an entire session. Note that this means a session's complete history can be erased; operators who need long-term audit retention must export events before deletion or use archiving instead.

  3. Log attribution - verified from the documented event schemas, which carry session_thread_id and agent_name on thread lifecycle events, from_agent_name and from_session_thread_id on inter-agent message events, and session_thread_id on cross-posted tool use events.

Scoring

Log Completeness

Log Completeness
Score valueExplanation
1The platform records all categories of agent activity in the persisted session event history.

Log Immutability

Log Immutability
Score valueExplanation
1The API exposes no per-event modification or deletion endpoint; individual log entries cannot be altered or selectively removed.

Log Attribution

Log Attribution
Score valueExplanation
1The platform attaches agent identity to every event type on the primary session stream.

Run Evaluation in LatticeFlow AI Platform

Use the following CLI command to initialize and run the evaluation in LatticeFlow AI Platform.
Requires LatticeFlow AI Platform CLI
lf init --atlas audit_log_integrity_claude_managed_agent

Metrics

Log CompletenessLog ImmutabilityLog Attribution

Don't have the LatticeFlow AI Platform?

Contact us to see this evaluation in action:
Contact Us