Agentifact assessment — independently scored, not sponsored. Last verified Mar 6, 2026.
Garak
Open-source LLM vulnerability scanner from NVIDIA for automated red-teaming of AI agents and models. Probes for jailbreaks, prompt injection, hallucination, data leakage, misinformation, and toxicity across 18+ attack methods. Supports OpenAI, Hugging Face, LiteLLM, REST APIs, and local GGUF models. Free to use; outputs JSONL vulnerability reports.
Viable option — review the tradeoffs
You need to systematically red-team your LLM endpoints for jailbreaks, prompt injections, hallucinations, and data leaks before deployment.
Comprehensive scans in minutes with detailed hit logs; high detection rates (e.g., 90%+ on encoding attacks) but false positives possible; modular for custom probes.
You want ongoing security monitoring for production AI agents across multiple providers without custom scripting.
Reliable for pre/post-deployment checks; excels on known attacks like DAN/snowball but may miss novel zero-days; free and extensible.
Probe-focused, not full pentest
Tests predefined 50+ attack modules; lacks dynamic fuzzing or custom agent workflows beyond basic LLM probing.
API costs during scans
Probes generate 100s of calls to target LLM; monitor provider quotas/billing—use --max_successes to limit.
Trust Breakdown
What It Actually Does
Garak tests AI models and chatbots for security flaws by running attack scenarios that probe for jailbreaks, prompt injection, and harmful outputs. It works with OpenAI, open-source models, and custom APIs, generating detailed vulnerability reports.
Open-source LLM vulnerability scanner from NVIDIA for automated red-teaming of AI agents and models. Probes for jailbreaks, prompt injection, hallucination, data leakage, misinformation, and toxicity across 18+ attack methods. Supports OpenAI, Hugging Face, LiteLLM, REST APIs, and local GGUF models.
Free to use; outputs JSONL vulnerability reports.
Fit Assessment
Best for
- ✓llm-security-testing
- ✓vulnerability-scanning
Score Breakdown
Protocol Support
Capabilities
Governance
- audit-log