AI Compliance
Evidence for every AI decision
Ferentin records each model and tool call with the person behind it, the policy that decided it and who approved it. Export the evidence or stream it to the SIEM you already run.
The problem
Auditors ask what the AI did and why
An auditor wants to know which agent touched customer data, on whose authority and under which rule. Model and tool vendors log their own side, with no record of your identities or your policies. A log is also only evidence if the auditor can trust nobody edited it.
How it works
How it works
- 1
Record every call with its decision
Every LLM and MCP call is recorded with the identity behind it, the policy that matched, what it decided and OpenTelemetry trace context. Filter Activity by policy or decision to pull exactly the calls one control governed. Filters live in the URL, so an investigation is a link you can paste into a ticket.
- 2
Keep every approval on the record
When a Step-up policy holds a call, the request records the tool, its arguments, who asked and who decided. A denial requires written reasoning and nobody can approve their own request. A closed request can carry a signed receipt of who decided what and when.
- 3
Show which policy was live
Activations records every policy version your edges and managed devices enforced, who published it and what changed. Export it as CSV or JSON for the audit period. The Admin source in Activity shows who changed each policy and when.
- 4
Stream it to your SIEM
Register an OTEL sink and a telemetry policy sends traces, metrics and logs from LLM and MCP activity to it, at the sampling ratio you set. Datadog, Grafana, Splunk and any other OTLP collector work as destinations, as do Kafka and S3. Your own system keeps that copy under its own retention.
Works with your stack
Works with your stack
Identities come from your identity provider. Evidence goes to the observability tools you already run.
Okta
Microsoft Entra ID
Google Workspace
Datadog
Grafana
Anthropic
OpenAI
Slack
Route agents through a control point that records each call with identity and policy. In Ferentin, every LLM and MCP call lands in Activity with the person or agent behind it, the policy that matched and its decision. Filter by policy, decision, server, tool or actor, then open any row for the full record. An empty policy field means no policy matched, which is itself a finding.
Go deeper
Learn more
- ISO/IEC 42001 control mappingAll 38 Annex A controls mapped to the evidence Ferentin produces, what it supports and what stays with you.
- OWASP Agentic Top 10 coveragePer-risk controls and evidence pointers for agentic applications, with honest gaps.
- Ferentin is SOC 2 Type II certifiedWhat the SOC 2 Type II report covers and what it means for your organization.
See it on your own stack
Walk through it with our team on your identity provider, your models and your MCP servers. Or start free today.