For the complete documentation index, see llms.txt. This page is also available as Markdown.

AI Security Detectors

The AI Security detectors surface OWASP LLM Top 10 (2025) risks against the AI assets in a repository. Every finding carries:

Facet
Meaning

OWASP LLM Top 10 (2025)

Primary taxonomy bucket (OWASP-LLM:2025:LLM01LLM10) — the grouping axis used in reports and in the UI.

OWASP ASI Top 10 (2026)

Secondary agentic-taxonomy tag (OWASP-ASI:2026:ASI01ASI10), for teams mapping findings to the OWASP Top 10 for Agentic Applications.

OWASP AST Top 10 (2026)

Behavior-layer tag (OWASP-AST:2026:AST01AST10), present only on detections that touch the skill surface (skills, MCP tool bundles, coding-tool skills).

Red-team vectors

Operational attack-vector tags (Prompt Injection, Indirect Injection, Jailbreaks, RAG Poisoning, Multi-Modal Injection, Data Exfiltration, …) for filtering and hunting.

AI asset kind

The inventory asset the detection applies to (ai_prompt, ai_agent, ai_skill, …), cross-linked to the AI inventory.

The full detector catalog — with descriptions, severities, standards mappings, and remediation guidance — is published at detectors.xygeni.io and updated on every release.

Secrets in prompts are detected by the Secrets Scanner, which inspects prompt files alongside the rest of the source tree. Secret findings located inside an AI prompt artefact are enriched with the LLM07 OWASP tag, so they also surface under the AI Security view — without duplicate findings.

Detectors can be enabled, disabled, or tuned per project — see AI Security Scanner Configuration.

Last updated