OSS LLM vulnerability scanner from NVIDIA. Apache-2.0. Probes target LLMs for jailbreaks, prompt injection, hallucinations, toxicity, leakage. Garak is to LLMs what nmap is to networks.
Garak is NVIDIA's OSS LLM vulnerability scanner — Apache-2.0 license. Probes target LLMs for jailbreaks, prompt injection, hallucinations, toxicity, leakage. Garak is to LLMs what nmap is to networks. Pick Garak for offensive LLM red-teaming + vulnerability assessment.
Garak's offensive scanning model fills a critical L5 gap: testing LLMs for vulnerabilities BEFORE production. From a Trust Before Intelligence lens, this is the canonical adversarial-red-team primitive. Pair with NeMo Guardrails for defensive runtime + Garak for offensive scanning = comprehensive L5 LLM security.
Scan campaigns are batch.
Probe DSL.
Scanner — deployment-driven.
Provider-agnostic.
Probe metadata + attack catalog.
Detailed scan reports per probe.
Governance/security is its purpose. 2/6 -> 5 lenient.
Reports as observability output. 1/6 -> 4 lenient.
Batch scanner. 3/6 -> 3.
1/6 -> 3.
5/6 -> 4.
Best suited for
Compliance certifications
Library — N/A.
Use with caution for
Guardrails for defensive runtime; Garak for offensive scanning. Use both.
View analysis →Promptfoo for evaluation; Garak for security scanning.
View analysis →Role: L5 LLM vulnerability scanner.
Upstream: Target LLM endpoints.
Downstream: Scan reports + vulnerability catalog.
Mitigation: Combine with manual red-teaming + bug bounty + adversarial datasets. Garak is one input, not comprehensive.
Garak's purpose.
Run on every model release.
Combine with manual + other tools.
This analysis is AI-generated using the INPACT and GOALS frameworks from "Trust Before Intelligence." Scores and assessments are algorithmic and may not reflect the vendor's complete capabilities. Always validate with your own evaluation.