The Metamynd AI Trust Framework
MATF is the governance framework for trustworthy AI — the principles, domains, lifecycle and deliverables required to govern AI at scale, made enforceable by the Metamynd Agentic Governance Protocol (MAGP).
Most 'AI governance' stops at documents. MATF goes further: every principle maps to a control that actually runs — an identity you can verify, an authority you can bound, a policy that returns allow / observe / block / escalate before an action happens, and evidence you can check without trusting us. The framework is vendor-neutral; Metamynd's platform is one implementation of it.
The seven domains
- Identity & Ownership
- Authority & Delegation
- Policy & Governance
- Evidence & Audit
- Risk & Assurance
- Certification & Trust Marks
- Federation & Interoperability
How the framework is enforced (MAGP)
Before a governed agent acts, it presents an Ed25519-signed request to the gate. The gate checks the agent's mandate, the enforced Standards, and the owner's SOPs with a deterministic policy engine and returns allow / observe / block / escalate — no LLM in the decision. Every verdict is anchored as tamper-proof evidence. The rules live in Metamynd and a compliance owner edits them live, with no redeploy of the agent.
The AI Trust Lifecycle
- Establish identity — did:hedera or a caller-controlled did:key
- Delegate authority — a VC-issued mandate bounded by scope, caps and merchants
- Govern with policy — Standards & SOPs → allow / observe / block / escalate
- Capture evidence — signed, Merkle-batched, anchored on Hedera
- Assess risk — escalate high-risk actions to a human
- Certify — KYC/KYB assurance + verifiable credentials
- Operate continuously — TrustOps + the Trust Index
Framework deliverables
- Principles
- Governance operating model
- Maturity model
- Certification framework
- TrustOps discipline
- Trust Body of Knowledge
- Global trust ecosystem
Vendor neutrality & interoperability
MATF is built on open standards — W3C DIDs & Verifiable Credentials, ODRL policy, and Hedera Consensus Service (HCS) — and interoperates with the Hashgraph Online (HOL) standards and the HCS-28 trust ecosystem. An agent trusted here can be recognised elsewhere, with no lock-in.
See it in code
import { createGuardFromConfig } from "@metamynd/agentsafe-guard";
const guard = await createGuardFromConfig("./agent.metamynd.json");
// Every governed tool call passes the mandate + Standards + SOP gate first.
const bookFlight = guard.guardTool("flight-purchase", rawBookFlight, mapArgs);
// $400 → allow · $600 → block (SOP_SPEND_CAP) · high-risk → escalate (RISK_REVIEW)