Portable evidence for high-risk AI actions.

AAP is a portable evidence standard and verifier for AI system activity. It packages manifests, hash-chained events, file hashes, timestamp evidence, and profiles for independent review.

Talk to us View AAP spec on GitHub

Problem

Reviews cannot rely on screenshots.

When AI systems operate in regulated or high-risk workflows, teams need more than logs scattered across dashboards. They need portable records that can show what happened without forcing everyone to trust a hosted UI.

Where it fits

Use AAP when evidence needs to leave the originating system: audits, incident review, regulated workflows, release control, runtime records, or third-party verification.

How it works

Evidence is not a dashboard view. It is a record that can travel beyond the system that created it.

Observe

Collect AI action semantics through SDKs, collectors, and sidecars outside the critical path.

Normalize

Turn traces, tool calls, configuration, and content references into a consistent evidence structure.

Anchor

Hash and timestamp evidence so later modification or deletion becomes detectable.

Verify

Export artifacts that reviewers can inspect without relying on screenshots or a single vendor dashboard.

What this can support

Example artifact

Abstract evidence-layer artwork with translucent paths and verification marks

Capture action-level context with OpenTelemetry-native observer architecture.

Record prompts, responses, tool envelopes, errors, durations, and config fingerprints.

Use identifiers and content hashes to minimize sensitive data storage by default.

Package tamper-evident evidence artifacts for independent review.

Who it is for

Regulated AI teamsSecurity and governance leadersAudit and compliance reviewersTeams building agentic systems

Want to use this in a workflow?

We can decide whether this building block fits your workflow, deployment boundary, and review requirements.

Talk to us