The OWASP Agentic Security Initiative Top: A Practical Developer Guide
8 hours ago
- #AI Security
- #OWASP ASI
- #Agent Testing
- Tested Claude Haiku with 30 adversarial prompts across OWASP ASI categories, resulting in 20 passes and 10 uncertain returns.
- OWASP ASI Top 10 addresses agent-specific threats like prompt injection, scope violation, and memory manipulation, distinct from the LLM Top 10.
- Each ASI category includes testing guidance and a command-line tool (safelabs run) for evaluation, with examples of real-world vulnerabilities.
- AgentSafeLabs provides structured test cases covering all 10 categories, enabling reproducible security assessments for AI agents.