Mark it — and keep it verifiable.
Marking is now an industry standard. The unsolved half is keeping content verifiable months later, after CDNs and re-uploads strip the credential. AI Act 50 is the registry + public verification layer that closes that gap.
From 2 August 2026, EU AI Act Art. 50 applies to all AI outputs; machine-readable marking becomes enforceable 2 December 2026. As autonomous agents publish at scale, marking is converging across the industry — what no tool guarantees is that a regulator can still verify the content months later. AIACT50 is the registry + public verification layer that closes that gap.
Every AI-generated output must carry machine-readable marking. No exceptions for agencies, platforms, or autonomous agents. Transparency breaches: fines up to €15M or 3% of global turnover.
Google SynthID and Adobe CAI each address a single layer. The EU AI Office's Code of Practice explicitly requires a multi-layer approach. Single-layer deployments leave organizations exposed.
AI agents generating content autonomously — with no human in the loop — require a certification infrastructure that works without API keys, pre-registration, or human approval. Traditional tools weren't designed for this.
Every company that has already adopted C2PA could use our solution to be AI compliant. We don't replace the standard — we complete it for the agentic economy.
Marking at generation is solved. The gap is keeping content verifiable months later — that's what L3/L4 rescue.
Every layer maps to a specific requirement in the EU AI Office Code of Practice. The first three exist. Only Vottun connects all four.
Integrate in 20 minutes. Certify every AI output. Hand clients a verification link they can forward to auditors — or regulators.
Content hash + metadata → immutable Base L2 record + C2PA 2.1 Credential. Returns certificate ID, public verify URL, and transaction hash.
Public endpoint. No API key. No account. Anyone — regulators, auditors, clients — can verify any certificate independently. No trust in Vottun required.
Enterprise dashboard. All certifications, compliance reports, content type breakdown. Exportable PDF/CSV. GRC webhook integration.
import vottun_ai_compliance client = vottun_ai_compliance.Client("vt_live_...") # 4-layer certification: watermark + C2PA + blockchain + detection cert = client.certify( content=ai_output, content_type="text/html", ai_system="claude-sonnet-4-5", classification="fully_ai_generated", # EU AI Act Art. 50 metadata={"client": "Regulated Corp", "deployer": "agency-name"} ) print(cert.verify_url) # → https://verify.vottun.com/c/0x7f3a... print(cert.tx_hash) # → Base L2 transaction hash print(cert.c2pa_manifest) # → C2PA 2.1 Content Credential JSON
AIACT50 is built from the ground up for the autonomous AI economy. Using the x402 protocol (Cloudflare + Coinbase), any agent — on any model, any infrastructure — calls the certification endpoint directly, pays from its own Base L2 wallet via machine-to-machine HTTP payment, and receives an immutable EU AI Act Art. 50 compliance record in under 4 seconds. The wallet is the identity. No prior registration, no human in the loop, no API key required at any step.
Via Claude, GPT, DeepSeek, or any model. Content hash computed locally before any network call. No data leaves the agent until it decides to certify.
The endpoint is open. The agent calls it directly and receives HTTP 402 with USDC payment instructions — zero prior registration required.
The agent sends USDC from its Base L2 wallet. Confirms on-chain in ~2 seconds. The wallet address is the agent's cryptographic identity — no X.509 certificate needed.
All 4 layers applied. Certificate anchored on-chain. Verifiable by any human or machine via GET /verify. No intermediary, no trust in Vottun required.
# No API key. No registration. No human approval. # The agent calls the open endpoint directly. response = requests.post( "https://api.vottun.com/v1/certify", headers={ "X-402-Payment": usdc_proof, # USDC from agent wallet "X-Agent-Wallet": "0x..." # Wallet = identity }, json={ "content_hash": sha256, "ai_system": "agent-v2", "classification": "fully_ai_generated" } ) # cert.verify_url → immutable Art. 50 compliance record # cert.tx_hash → Base L2 on-chain anchor # Total time: <4 seconds. Zero humans involved.
Each layer runs in public or private mode. Marking can be standard (wrap C2PA + SynthID) or proprietary. The verification layer is the same in all four — public and private registries are two trust models, not two levels of compliance.
Composite content — a foundation model plus your own RAG or transformations — is certified as a single output via a composition_record assertion carrying a Merkle root of ingredient declarations. Neither OpenAI nor Google certifies the final composed output; AI Act 50 does.
The SDK, MCP server, and C2PA extension spec are fully open source under MIT licence — so any developer, organization, or regulator anywhere in the world can inspect, implement, and build on it without restriction. A compliance standard only becomes a real standard when anyone can adopt it, audit it, and trust it independently. That's why we open-sourced everything.
The open-source approach plus blockchain-anchored proof is exactly what our regulated clients need. This is the infrastructure layer we've been waiting for — not another compliance tool. — Compliance Partner, Big 4 Consulting Firm
Free on testnet. 20-minute integration. EU AI Act Art. 50 applies from 2 August 2026 (machine-readable marking enforceable 2 December 2026) — and the standard that gets adopted first is the one that becomes the standard.
Already have an account? Login