AI you don’t have to take on faith.
When a HITS product carries this mark, its AI is governed by construction — not by a policy document or a promise. The constraints below are enforced in the execution path itself. The AI cannot route around them, because they are not instructions to the AI; they are the rails it runs on.
Every governed decision emits a cryptographically signed receipt — a tamper-evident record of what was asked, what evidence was used, what the governance decision was, and who (if anyone) approved it. Anyone holding the receipt can verify it independently, without trusting us.
What the mark guarantees
Grounded answers cite the specific evidence they rest on, by id. A claim with no citation to supplied evidence does not ship — the system holds the answer rather than inventing one.
When the evidence cannot answer the question, the product says so plainly. “I can’t answer that from what I have” is a valid, logged outcome — not a gap to be filled with a confident guess.
Consequential actions pass through a gate. A human stays in the loop where it matters, and the approval — who decided, when, under which policy — is recorded as a verifiable artifact.
The AI will not flatter, pressure, or steer you away from your own judgment. Its posture is invariant to how you phrase a thing — it owes you the strongest honest counter-case, not the answer it thinks you want to hear.
Each governed decision is signed and hashed into a record you can check yourself. Tampering breaks the signature. The proof does not depend on trusting HITS.
Where it runs
The Governed AI capability is a shared spine. These HITS products inherit it:
See it work
Run a grounded query and watch it cite its evidence by id — or ask something it can’t answer and watch it refuse instead of guessing. Then recompute the signed receipt’s hash in your own browser. No login, about 30 seconds, live at raptor.harpyits.com/proof.
Watch the proof, live →