← Vetro Attest · Free Art. 12/14 checker · Free EU AI Act tools
Audit log for Amazon Bedrock Agents
Amazon Bedrock Agents orchestrate a foundation model against tools ("action groups") and knowledge bases behind a managed InvokeAgent API. Unlike a framework you host yourself, the orchestration loop runs inside AWS — so the audit trail starts with whatever the service is willing to tell you about a given invocation.
What Bedrock Agents already gives you
Calling InvokeAgent with tracing enabled returns a stream of trace events alongside the response: the model's rationale at each step, the action group or knowledge base it decided to call, the input it sent, and the observation it got back. This is close to a full record of one invocation, produced by AWS rather than by your own code.
What it doesn't give you
- Retention and location. The trace is returned to your caller at invocation time; nothing stores it for you afterward. If you don't capture and keep it, it is gone.
- Tamper evidence. Whatever you do store — a log line, a database row — is editable by anyone with access to it, the same problem as any other application log (see tamper-evident logs).
- Attribution. Nothing in the raw trace proves later that it wasn't altered before being handed to a reviewer.
Closing the gap
Capture each trace event as your caller receives it and append it to a signed, hash-chained ledger instead of an ordinary log line:
import { Vetro } from '@vetro/core';
const vetro = new Vetro();
const run = vetro.startRun({ name: 'bedrock-agent-invocation' });
// for each trace chunk in the InvokeAgent response stream:
run.event('annotation', { kind: 'bedrock.trace', trace });
run.end('completed');
Map the trace's own structure onto typed events where it matches — a model invocation step to run.llmRequest / run.llmResponse, an action group call to run.toolCall / run.toolResult — and fall back to run.event for anything specific to Bedrock's trace format. Enforcing policy *before* an action group runs isn't possible from outside the managed loop; if you need hard blocking rather than after-the-fact recording, put the check inside the Lambda function backing the action group itself.
Why keep your own copy
CloudTrail records that InvokeAgent was called; it does not capture the reasoning trace or the action group's arguments and results. For Article 12 logging and Article 26 deployer log retention, the trace content is what an auditor actually wants, and it needs to outlive whatever default retention your logging setup has.
Check your own logs
Paste a Vetro ledger or an OpenTelemetry GenAI trace export into the free checker: it verifies integrity and shows which record-keeping obligations your evidence covers, in your browser. When you need something to hand to an auditor or a customer, Vetro Attest issues a countersigned EU AI Act report for 29 USD, one-off.
See a sample ledger verified in one click, check your own logs free or get the countersigned EU AI Act report.