About AgentAuditLabs
An independent, signal-based trust protocol for AI agents. Open source. On-chain attested. Reproducible from event replay.
Why AAL exists
AAL aggregates signals from security testing, reliability monitoring, third-party attestations, and on-chain registries to produce a Universal Trust Score on a 0–100 scale.
AAL is not a scoring formula. It is a protocol. Signals flow in from diverse, independent sources. The engine weights them by source trust and recency. The result is a transparent, reproducible trust assessment backed by Ed25519 cryptographic signatures.
Protocol facts
Architecture
The AAL protocol follows a linear pipeline: signals are collected from sandbox testing and external crawlers, stored with open-vocabulary tags in an append-only event store, processed by the scoring engine as a weighted mean of measured applicable signals, with Pedigree (source-diversity completeness) reported alongside, and output as scores, badges, on-chain attestations, and W3C Verifiable Credentials.
Trust, Pedigree, Passport
No sandbox test is on record for this agent. No score has been computed, and no Passport is issued.
Benign probes only — adversarial testing has not been authorized by the owner. The tier is capped at Gold and no Passport is issued. The Trust number reads against the Gold ceiling as N / 69 (Partial).
The full protocol ran with written owner authorization. A [TIER] Passport is issued: Trust is the weighted-mean score and Pedigree the signal completeness reported alongside it.