Auditable AI Governance
AI drafts. A named human approves. Every action is hashed and reversible.
OrPaynter is an auditable AI governance layer for autonomous agents. It is not a black box. The rules are public, the hashes are checkable, and the human gate stays in control.
Default-deny
Hash-bound approval
Portable undo
Public constitution
The one-line proof
Run any model through the arena. It can propose. It cannot approve, edit after signing, invent provenance, or rewrite the rules. The gate catches it. The verifier reports it. The root hash binds it.
What is true
- Provisional patent filed: 64/144,590 (2026-08-31).
- Constitution is a public artifact with a hash; immutable to agents.
- DecisionPackages require a named human before irreversible action.
- TRAE attaches provenance-linked evidence; scores risk; never convicts.
- Arena + verifier make learning reconstructable, not sneaky.
How to verify
- Read this page. It is the public claim, not a private dashboard.
- Open /llms.txt for the machine-readable directory.
- Inspect the ClaimFlow source when it is public: github.com/orpaynter/claimflow.
- Check that every irreversible action stops at STATUS: UNSIGNED until a named human signs.
FAQ
Is it unbreakable?
No. It is auditable. Failures are visible, reversible, and attributable.
Can AI act alone?
No. Default-deny packages require a named human signature.
Is the governance hidden?
No. This page, the constitution, and the hashes are the public surface.