Open-source attestation toolkit for the three regulations breathing down your AI roadmap.
AttestProto is the cryptographic per-decision attestation layer that auto-maps to the regulations you actually have to comply with. Self-hosted, MIT-licensed, no telemetry. Free for SMBs, free forever for self-hosted use.
Pick your compliance stack
EU AI Act — Aug 2 deadline
Article 12 (logging) + 13 (transparency) + 14 (oversight) + 17 (QMS) + 19 (retention) auto-mapped from a single attestation. €15M penalty if you miss the deadline. Free for EU SMBs.
AI lending — FCRA, ECOA, CFPB
FCRA § 1681m + Reg B § 1002.9 + CFPB Circular 2022-03 — specific-reasons disclosure auto-flagged when missing. GDPR Article 22 + Colorado AI Act bundled. Built for AI lending fintech compliance teams.
NYC LL144 bias-audit infrastructure
Per-tool-call attestation in the exact shape § 20-871 selection-rate + impact-ratio computation needs. Built for the audit firms. Skip the data-plumbing, bill more analysis hours.
Why now
Three regulations, all with different deadlines, all needing the same underlying primitive: a per-decision cryptographic record of what the AI agent did, what data it saw, and what it concluded.
- EU AI Act — high-risk obligations enforceable August 2, 2026, with €15M / 3% global revenue penalty for non-compliance.
- CFPB Circular 2022-03 + 2023-03 — explicit "specific reasons" disclosure for AI/ML credit decisions in the US, enforced today.
- NYC Local Law 144 § 20-871 — annual bias audit for every NYC AEDT, with enforcement ramping up after the December 2025 NYC Comptroller report.
Today every platform — OpenAI, Anthropic, Google, Stripe — ships its own logging. Cross-vendor compliance is impossible without a neutral attestation layer.
What you get
Portable on-the-wire format
Single JSON object signed with Ed25519 over RFC 8785 canonical bytes. Verify with 30 LOC in any language.
Multi-hop delegation
Cryptographic chain-of-proof reconstruction across independently-operated agents. Each link is independently verifiable.
Confidential mode
Zero-knowledge proof option: prove conformance without revealing inputs or outputs. Healthcare, finance, M&A ready.
Reputation engine
Time-decayed agent quality scores. Underwrite AI liability insurance. Route work to higher-trust agents.
Compliance mapping
Auto-translate raw attestations into EU AI Act, NIST AI RMF, SOC 2, HIPAA, ISO 42001 citations. No bespoke consulting.
MCP-native adapter
Drop-in for any MCP-compatible agent. Single line of code: attestproto.attach().
30-second example
{
"version": "0.1.0",
"attestation_id": "att_x7QrJ8k2pL4mN9vBfH3wXa",
"agent": {
"id": "asst_kn4XJZ8K2pLm9N",
"platform": "openai",
"model": "gpt-5o-2026-04-15",
"operator": "https://example-fintech.com"
},
"task": {
"type": "fintech.kyc.verify-identity",
"spec_hash": "sha256:9f86d081884c7d659a2feaa0c55ad015a3bf4f1b2b0b822cd15d6c15b0f00a08"
},
"input": { "hash": "sha256:2cf24dba…", "redaction_policy": "full-redacted" },
"output": { "hash": "sha256:486ea462…", "verdict": "success" },
"signature": {
"alg": "ed25519",
"key_id": "https://example-fintech.com/.well-known/agent-keys.json#ops-2026",
"value": "MEUCIQDxyz…"
}
}Status
| Spec section | Status |
|---|---|
| §3 Attestation object structure | v0.1 stable |
| §4 Canonicalization (RFC 8785) | v0.1 stable |
| §5 SHA-256 hashes | v0.1 stable |
| §6 Multi-modal identity resolution | v0.1 stable (DNS / DID / HTTPS / fingerprint) |
| §7 Multi-hop delegation chain-of-proof | v0.1 stable |
| §8 Zero-knowledge confidential mode | v0.2 (Q3 2026) |
| §9 Co-signatures | v0.1 stable |
| §11 Reference ledger + reputation engine | skeleton in repo |
Reference implementations
- Python (
@attestproto/core) — sign + verify + canonicalize + CLI. Docs - Node.js (
@attestproto/core) — cross-language interoperable with Python. Docs - FastAPI ledger — POST /attest, GET /verify, /chain, /agent/score, /epoch. Docs
- MCP server — exposes the ledger to Claude / GPT / Gemini agents. Docs