OWASP Agentic AI Top 10
The OWASP Agentic AI Top 10 catalogs the top risks for autonomous / agentic AI systems (tools, planners, multi-agent workflows). Gadriel Code renders a per-control report; today every control is manual review.
Related: Compliance overview · OWASP Agentic Skills Top 10 · OWASP LLM Top 10.
At a glance
| Property | Value |
|---|---|
| Slug | owasp-agentic |
| Aliases | (none) |
| Controls | 10 |
| Manual-only | 10 (fully manual) |
| Machine-assessable | 0 |
| Direct mapping key | (none — structural gap) |
Fully manual today. ComplianceMapping has no owasp_agentic key, so
direct (Tier-1) rule-to-control mapping is not yet possible. Every
control renders as NOT ASSESSABLE and carries a narrative for the human
reviewer. The framework rollup will PASS on the manual-neutral rule.
Generating the report
bashgadriel code report --compliance owasp-agentic
Outputs: .security/compliance/owasp-agentic.{md,typ,pdf}.
Why it still ships
The report is a structured review workbook: each control renders with its official description, guiding questions, and space for reviewer notes. Use it as your agentic-AI threat-modeling artifact even before automated coverage lands.
For source-level scanning of agentic skill bundles (SKILL.md, .cursor/rules,
.windsurf/rules, AGENTS.md), see the companion
OWASP Agentic Skills Top 10 scanner — that pack
is fully automated.