New Agentic AI Red Team Checklist Adds 222 Tests Across 20 Attack Categories
Researcher Ravi Rajput released 222 agentic-AI red-team tests across 20 categories, mapped to OWASP and MITRE ATLAS.
Security researcher Ravi Rajput released a free spreadsheet with 222 tests in 20 categories for assessing autonomous AI agents beyond prompt injection. The tests cover attack-surface mapping, inputs and unsafe outputs, tools, memory, agent networks, and Model Context Protocol servers, plus deployment pipelines, privilege abuse, persistence, and data theft. Proposed ratings are 75 Critical, 108 High, 30 Medium, and nine Low; these are test severities, not confirmed flaws in a product. The independent resource, which is not an official OWASP checklist, maps tests to OWASP and MITRE ATLAS and omits fixed CVE identifiers.
- 222 tests cover infrastructure, cloud access, tools, memory, and agent links.
- Phases run from discovery through inputs, tools, and impact scenarios.
- Examples include SSRF credential theft, pickle loading, and cross-tenant access.
- Ratings propose 75 Critical and 108 High tests, not confirmed flaws.
- Maps to OWASP and MITRE ATLAS and omits fixed CVE identifiers.
From the original article
3 min read at cybersecuritynews.com
A free agentic AI red-team checklist offers 222 tests across 20 attack categories to assess autonomous AI systems beyond prompt injection, covering infrastructure, cloud access, tools, memory, and agent communications. The checklist targets a common testing gap. Teams may spend days trying to manipulate a model while overlooking an exposed MLflow server, a reachable cloud […]
By Abinaya · cybersecuritynews.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.