AI agents increasingly drive critical systems—from healthcare diagnostics to autonomous vehicles. Yet their autonomy introduces vulnerabilities: adversarial attacks, data leakage, and unintended harmful behaviors. This standard provides a systematic methodology to test agent resilience across interfaces, models, tools, and life-cycle stages, ensuring they operate within ethical and safety boundaries.

Aligned with WDTA’s 3S principles (Speed, Safety, Sharing), this document accelerates secure AI adoption while fostering international collaboration. We commend the AI STR Working Group and contributors for pioneering a framework that balances innovation with accountability. Their expertise delivers actionable guidance for developers, auditors, and policymakers to build AI systems that serve humanity securely.