Every AI-governance vendor claims enforcement. GAUNTLET doesn't ask — it fires a battery of real hostile intents at the policy engine and publishes what got through. Proof, not promises.
Press FIRE — the server runs the fixed public battery
(/sovereign-api/gauntlet): every verdict is decided by the
live Sovereign policy engine, right now, and the run is sealed into its hash-chained ledger.
The catalog lives server-side — nobody can inject their own intents or tailor the test.
Nothing on this page is staged.
Testing your own enforcement stack against the battery is the paid product — a GAUNTLET assessment. Live against our Sovereign is free to watch; aimed at your engine, it needs a license.
The probe is fired server-side against the target. Licenses are issued per assessment — see plans.
Most AI-governance products record what the agent did. Very few decide whether it may act at all. GAUNTLET makes the difference measurable in public: if your platform only writes a prettier audit log, the scorecard says so — in red.
GUARDIAN Sovereign is the reference implementation: every verdict above is also the live production ledger on this host — watch it enforce →