Security at Phronesis

Phronesis is the decision-assurance layer for AI agents — so our own security posture is part of the product, not a footnote to it. We hold ourselves to the discipline we sell: say only what is true, and make it verifiable.

Reporting a vulnerability

Email [email protected]. We acknowledge reports and work them in good faith. Please give us reasonable time to remediate before public disclosure.

Machine-readable disclosure policy: security.txt (RFC 9116).

How the surface is built (what's true today)

• Least exposure by design. Our agent-facing surface answers only read-only queries at its open tier; every consequential action (anything that moves money, binds a commitment, or asserts an identity) is gated and, at open access, refused — not executed.
• Identity never confers capability. An agent can register, but registration grants identity only; it can never widen what that agent is allowed to do.
• Fail-closed. If our access controls can't be verified, the door refuses rather than opening.
• A tamper-evident record. Every decision we publish is cryptographically anchored and independently verifiable — anyone can check what we said and when, and no one, including us, can alter it after the fact.

What we don't claim

We attest; we do not guarantee outcomes, and we do not carry the risk of the actions we assure — liability stays with the principal and the executor. Assurance is a check on the action, not a warranty on the result.