Framework · Principles
16 AI Trust Governance Principles
The foundations of trustworthy AI — and, for each, the concrete Metamynd control that makes it enforceable rather than aspirational.
01
The principles
- Human Accountability
- Verifiable Identity
- Explicit Authority
- Evidence Before Assertion
- Explainability
- Continuous Governance
- Proportional Risk
- Trust Through Transparency
- Privacy Preservation
- Security by Design
- Sovereignty
- Interoperability
- Continuous Improvement
- Ethical Operation
- Resilience
- Sustainability
02
Principles as running controls
- Human Accountability → every agent is owned by a legal entity, verified via KYC/KYB for mainnet, with its verification status and basis disclosed on sandbox and testnet
- Verifiable Identity → did:hedera / did:key, key proven via verify-key
- Explicit Authority → a VC-issued mandate: scope, per-transaction & total caps, merchants
- Evidence Before Assertion → Ed25519-signed decisions anchored on Hedera
- Continuous Governance → the authorize gate runs on every action; rules change live
- Proportional Risk → high-risk actions escalate to a human, not auto-approve
- Trust Through Transparency → verdicts are deterministic and independently re-derivable
- Privacy Preservation → only commitments/hashes go on-chain; PII stays encrypted at rest
- Sovereignty & Interoperability → open DIDs, VCs, ODRL and HCS — no lock-in
03
Why it matters
A principle you cannot enforce is a slogan. Each MATF principle is bound to a control in the protocol, so compliance is demonstrable — with signed evidence — not merely claimed.