Solutions  ·  2026-09-10

NVIDIA garak v0.17.0 adds EU AI Act risk-category mapping to LLM red-teaming scanner

SolutionsMedium impactGlobal
Garak v0.17.0 (released Sept 9, 2026) adds an EU AI Act Mapping feature providing reference tags that group probe results by EU AI Act risk categories, plus improved plugins, in the widely-used open-source LLM vulnerability scanner.
Ties automated LLM red-teaming output directly to EU AI Act compliance categories, letting organizations map technical red-team findings to regulatory risk classes — notable OSS capability as EU AI Act enforcement ramps up.
Teams running LLM red-teaming/compliance testing in EU-regulated contexts should upgrade to leverage the new mapping for audit and compliance reporting.
GitHub — NVIDIA/garak releases
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →