Frontier AI models reach 10–54% legal compliance when deployed as agents (Aithos LARA, 12 models, published May 2026). The Agent Authorization Envelope (AAE) makes authorization constraints structural — enforced before the tool call.
Built for compliance officers, AI platform leads, and conformity assessment bodies. This surface is for technical evaluation and integration planning — not self-service signup.
MolTrust's compliance surface is built for the parties who need to evaluate, document, or report on agent behaviour against legal requirements — not just the developers integrating the SDK.
draft-kroehl-agentic-trust-aae-00 — published as an IETF Internet-Draft.Each Article 12 obligation mapped to a concrete MolTrust feature, with current implementation status.
did:moltrust:* — registered, signed, resolvable. Live.POST /compliance/assess, POST /compliance/declaration, GET /compliance/report/{did}. Live.POST /compliance/incident — staggered reporting deadline (2 / 10 / 15 days) tracked per incident. Live.Singapore's IMDA Model AI Governance Framework for Agentic AI (January 2026) recommends that each agent carry a unique, cryptographically verifiable identity with scoped, recorded authorization — the properties the AAE implements. The AAE aligns with that guidance.
Each endpoint runs on the live API at api.moltrust.ch. A classification returns the risk tier, the obligations that attach to it, and the EUR-Lex article each rests on. The assessment is a protocol-layer determination pinned to the text; it does not itself constitute compliance, which remains a determination for the deploying organisation.
POST /compliance/assessPOST /compliance/declarationGET /compliance/report/{did}POST /compliance/incidentPOST /delegation/create · /verifyPOST /reputation/batch-sync · /anchors/batchA use case listed in Annex III is high-risk under Article 6(2). The Article 6(3) derogation removes that status for a narrow procedural, preparatory, or human-improving task; a system that performs profiling of natural persons stays high-risk in every case (Article 6(3), final subparagraph). Article 7(2) supplies the criteria — intended purpose, autonomy, data sensitivity, reversibility of harm — that weigh a borderline case.
Issued by POST /compliance/declaration, verifiable by any JOSE library against api.moltrust.ch/.well-known/jwks.json. Trimmed for display.
{
"@context": ["https://www.w3.org/ns/credentials/v2",
"https://api.moltrust.ch/contexts/trust/v1"],
"type": ["VerifiableCredential", "MolTrustConformityDeclaration"],
"issuer": "did:web:api.moltrust.ch",
"validFrom": "2026-07-11T19:58:35Z",
"credentialSubject": {
"declarationStandard": "EU AI Act Annex V (Reg (EU) 2024/1689, Art 47)",
"aiSystem": { "name": "Acme Triage Assistant", "reference": "ATA-2026.1" },
"provider": { "name": "Acme Health GmbH", "address": "Berlin, DE" },
"soleResponsibilityStatement": "This EU declaration of conformity is issued under the sole responsibility of the provider.",
"conformityStatement": "The AI system is in conformity with Regulation (EU) 2024/1689.",
"dataProtectionStatement": "The AI system complies with Regulation (EU) 2016/679, Regulation (EU) 2018/1725 and Directive (EU) 2016/680.",
"harmonisedStandards": ["EN ISO/IEC 42001:2023"],
"issuance": { "place": "Berlin", "date": "2026-07-11",
"signatory": { "name": "J. Muster", "function": "Head of Compliance" },
"onBehalfOf": "Acme Health GmbH" },
"annexVComplete": true
},
"proof": {
"type": "Ed25519Signature2020",
"verificationMethod": "did:web:api.moltrust.ch#key-ed25519",
"canonicalizationAlgorithm": "JCS",
"proofValue": "84cadfba8f4fd2a535a09292529c3593977ab792d96372632bc9b43b7bee2b68..."
}
}
Regulation (EU) 2024/1689 applies generally from 2 August 2026, with the exceptions set out in Article 113(a)–(c). Prohibited practices (Chapters I–II) have applied since 2 February 2025; governance and general-purpose-model rules since 2 August 2025; high-risk classification under Article 6(1) applies from 2 August 2027. Dates verbatim from Article 113.
MolTrust is a deterministic cryptographic protocol layer, comparable in role to TLS or PKI. It carries agent authorization metadata; it does not process customer payload, personal data of end users, or business secrets in the intended configuration.
Each step is a single command or call. Together they produce a signed, on-chain-anchored Article 12 logging envelope per agent action — the same envelope your conformity assessor will later inspect.
One command from your project root. The SDK is available for TypeScript, Python, and Go.
npm install @moltrust/sdk
For each agent action, issue an AAE that records what the agent is authorized to do and under what constraints. The envelope is signed, anchored on Base L2, and immediately auditor-ready.
import { moltrust } from '@moltrust/sdk'; const aae = await moltrust.issue({ did: 'did:moltrust:your-agent-id', mandate: 'process invoice approvals up to USD 5,000', constraints: { maxAmount: 5000, validUntil: '2026-12-31' }, validity: { from: 'now', until: '+30d' } }); // → AAE issued, signed, anchored on Base L2. // → aae.hash gives you the anchor reference.
At audit time, export a signed PDF audit bundle covering any period. Your auditor can verify it independently without contacting MolTrust.
const bundle = await moltrust.exportAuditBundle({ did: 'did:moltrust:your-agent-id', period: { from: '2026-01-01', to: '2026-03-31' } }); // → Signed PDF (PAdES-B-LT), hash anchored on Base L2. // → Verifier URL included in bundle for independent verification.
Signed audit bundles are included in every paid plan — there is no separate compliance SKU. See pricing → · See a sample audit bundle PDF →
The Agent Authorization Envelope (AAE) is the cryptographic primitive behind MolTrust's compliance surface. The full specification — including the formal model for delegation, expiry, and revocation — is published as a technical paper (arXiv preprint).
draft-kroehl-agentic-trust-aae-00). Aligns with W3C DID and VC working groups.Ordered by audience arrival on the page. No marketing claims — just what holds under regulatory scrutiny.
verify.moltrust.ch/bundle/<hash> URL. Your auditor visits that URL — a public page that shows the on-chain anchor transaction on Base L2 (any block explorer confirms it independently).moltrust verify bundle.pdf from the open-source CLI (or any W3C DID-conformant tool). It re-checks the signature against the on-chain anchor without contacting any MolTrust service.Building identity-and-access infrastructure for AI agents? The compliance and IAM surfaces are intentionally separate (NIS2 / ISO 27001 role separation). For machine identity, agent authentication, and IAM-focused integration patterns, see the API documentation or contact us via /contact.