Six steps, 0 to 5
Each score tier states the evidence it requires. A higher score never means a better story, only better evidence.
012345
Evaluate
How we examine an actor
We read what the actor published, record what it says in its own words, and score it against the rubric. The record is a ledger anyone can read.
Eight fields per entry
Every finding is one row, with the same fields each time.
ActorSpherePillarSourceDirect quoteAudit questionScore, 0 to 5Verification
Tech enterprises
System cards, responsible scaling policies, frontier model commitments and open-weights release summaries.
Regulators and sovereign bodies
EU AI Act codes of practice, the UK AI Security Institute framework and China's generative AI measures.
Academia and watchdogs
Benchmark results, independent shadow audits, red-team papers and vulnerability disclosures.
Moral, philosophical and religious bodies
The Rome Call for AI Ethics and the UNESCO Recommendation on the Ethics of AI.
First audit
Read the audit
Frontier AI security and provider accountability
The ten most common security failure modes across 2,311 research papers, mapped to seven providers.