AI Agent Card

Making agentic AI transparent, auditable and controllable — one card at a time.

A standardized, machine-readable and human-readable card for documenting agentic AI systems.
Designed for transparency, human agency, regulatory compliance (EU AI Act), security (OWASP Agentic Top 10), and governance frameworks (ISO 27001:2022 + ISO/IEC 42001).

Why AI Agent Card?

As agentic AI systems become more autonomous, opacity becomes a major risk.
This standard provides a practical, governance-first "passport" for every deployed agent — making its goal, capabilities, autonomy level, human oversight mechanisms, risk mitigations, and treatment plan explicit and verifiable.

It complements discovery-focused Agent Cards (e.g., A2A protocol) by adding deep governance, risk treatment, and accountability layers.

Features

• Full support for EU AI Act risk classifications and obligations
• Detailed OWASP Top 10 for Agentic Applications mitigations (per-risk)
• Structured risk treatment plan with direct mapping to ISO 27001:2022 and ISO/IEC 42001 Clause 8 (Operation)
• Explicit human oversight ("how" it works, not just that it exists)
• EBSI Verifiable Credential ready (decentralized, cryptographically verifiable)
• Lightweight validation script + examples
• Designed for both individual researchers and enterprise ISMS/AIMS

Current Version

Draft v0.3 (RFC stage) — Actively seeking feedback and real-world use cases.

Quick Start

1. Clone or download the schema: agent-card-v0.3.json
2. Fill your agent card (start from the Grok Research Assistant example)
3. Validate: python validate-agent-card.py your-agent-card.json
4. Publish as JSON or wrap as an EBSI Verifiable Credential

Repository Contents

• agent-card-v0.3.json — Full JSON Schema
• validate-agent-card.py — Lightweight validation script
• examples/ — Populated examples (including Grok-based)
• docs/ — Additional guidance (coming in future releases)

Documentation

• Schema Documentation (add later)
• Risk Treatment Plan Guide (add later)

Contributing

We welcome contributions from AI governance, cybersecurity, standards, and policy professionals.
See CONTRIBUTING.md for details.

License

MIT License — see LICENSE

Related Work

• Worldview Belief System Card (WBSC) — Companion standard for base model/worldview transparency
• EU AI Act, OWASP Agentic AI Top 10, ISO/IEC 42001, ISO 27001:2022

---

Maintained by Rui Soares — Focused on practical transparency and human agency in AI.

Feedback, issues, and pull requests are welcome!