Runtime EU AI Act compliance enforcement. Deterministic. Not probabilistic.

EU AI Act Article 5 and Article 50 runtime deny for AI agents — on every input and output, in production. Classifiers (eu-ai-act-*) score the system; ComplyEdge denies this prompt or output. Article 50 here is unlabeled or deceptive use, not C2PA.

pip install complyedge
pip install trustlint
pip install 'complyedge[mcp]'
npx -y @complyedge/mcp
claude mcp add complyedge -- npx -y @complyedge/mcp
pip install 'complyedge[agents]'

CI: uses: complyedge/trustlint-action@v1

from complyedge import compliance_check

@compliance_check(jurisdiction="EU")
def your_ai_function(text: str) -> str:
    return call_llm(text)  # compliance enforced automatically

64 YAML rules, compiled to 71 OPA/Rego policies (64 leaf policies and 7 package aggregators), of which 51 leaves are the EU AI Act:

• EU AI Act Article 5 — 9 prohibited practices (social scoring, biometric ID, predictive policing, emotion recognition, ...)
• EU AI Act Article 50 — 5 transparency obligations (AI disclosure, deepfakes, chatbot identity, watermarking)
• GPAI Articles 51–55 — 5 obligations (model classification, copyright, documentation, systemic risk, downstream)
• Other EU AI Act articles — 11 covering Art 4, 6, 9, 10, 12, 13, 14, 15, 16, 26 and 27
• GDPR — 6 (consent, DPIA, erasure, minimisation, breach notification, cross-border transfer)
• US + global — 28 across SOX, HIPAA, TCPA, COPPA, PCI DSS, sanctions and prompt security

Every EU AI Act rule is open source, and every blocked request carries a verbatim article citation.

• 🌐 complyedge.io — Website
• 📖 Quick Start Guide
• 📚 API Reference
• 💰 Pricing

Apache 2.0 · EU AI Act: GPAI obligations carry fines from 2 August 2026