Definition
Checkable AI: claims you can verify without trusting anyone.
Checkable AI is AI whose governance claims can be verified by an outsider with no inside access. Four things make a system checkable: rules published in plain language, a cryptographic hash that locks the rules, receipts for what the system actually did, and labels that separate fact from inference. If verification requires believing the vendor, it is not checkable. It is marketing.
The three-step check, two minutes, no account
Step one: read the rules. Ours is a 42-article Constitution in the public domain, written to be read in one sitting. Step two: verify the lock. The canonical text publishes a SHA-256 hash, so any copy anywhere can be proven identical or proven tampered. Step three: read a receipt. The public ledger records actions, denials, and corrections in append-only, hash-linked entries. Run all three steps against any AI vendor you are evaluating. Most fail at step one.
Checkable versus trusted
Trusted AI asks you to believe a policy PDF and a press release. Checkable AI hands you the evidence and gets out of the way. The difference matters most for federal contractors and regulated teams, where an auditor will not accept a vendor's word, and it is the foundation of the audit trail standard we build to. Even our scoreboard is checkable: the live stats page counts events without cookies, and the raw JSON is public.
Try it on us right now
Run any claim through The Gate, our free GREEN AMBER RED guardrail. Compare the definition here against sovereign AI, its infrastructure sibling. Then check the services built on both. You do not have to trust us. That is the whole point.