Press Kit

Project-AI documents an execution-governed architecture in which required governance precedes governed effects.

Project-AI is a documented execution-governed AI architecture. Its contract places a nine-gate sequence before governed effects and defines ALLOW, DENY, and SAFE_HALT as terminal execution verdicts. The current source includes a versioned receipt contract, normal server signing path, local verifier, and anonymous redacted Witness; complete gate coverage, deployed key pairing, ledger provenance, and deterministic replay remain unverified.

Project-AI documents a deny-by-default effect-governance target with nine named gates and three terminal execution verdicts: ALLOW, DENY, and SAFE_HALT. The TSCG-B codec exists, but compiler integration, signing, and Code Store loading are separate planned or unverified capabilities. /verify locally checks strict receipt-v1 bundles and Witness exposes a privacy-redacted anonymous suffix; neither surface proves deployment-wide enforcement, signer custody, or ledger completeness. Comparisons with OPA, Constitutional AI, Guardrails AI, and NeMo Guardrails are architecture-level rather than claims of benchmark or security superiority.