CyberSecEval
Meta's benchmark suite for LLM cybersecurity risk and defensive skill: attack compliance, insecure code, prompt injection and SOC tasks.
Tags
Description
CyberSecEval 4 builds on versions 1-3 and adds defensive benchmarks such as CyberSOCEval. Tests include MITRE ATT&CK compliance and false-refusal rates, instruct and autocomplete secure-code tests, and prompt-injection tests. Code and benchmarks are MIT, in PurpleLlama/CybersecurityBenchmarks.
Related
- LlamaFirewall: Meta's open-source guardrail framework for agents that chains scanners for prompt injection, goal hijacking and insecure generated code.
- garak: NVIDIA's open-source LLM vulnerability scanner probing for hallucination, data leakage, prompt injection, jailbreaks, toxicity and other weaknesses.
- PyRIT: Microsoft's open-source Python Risk Identification Tool for proactively red teaming generative AI systems.
- AgentDojo: ETH Zurich benchmark environment for prompt-injection attacks and defenses on tool-using LLM agents, with suites like workspace, banking, travel and slack.
Sources
- https://github.com/meta-llama/PurpleLlama/tree/main/CybersecurityBenchmarks
- https://github.com/meta-llama/PurpleLlama/blob/main/CybersecurityBenchmarks/LICENSE
- https://meta-llama.github.io/PurpleLlama/CyberSecEval/
Machine-readable
- JSON
- Markdown
- Source file on GitHub (edit via pull request)