Meta has introduced LlamaFirewall, an open-source framework aimed at securing AI systems from emerging cyber threats like prompt injection and jailbreaks. The framework comprises three key components: PromptGuard 2, which detects direct attacks in real-time; Agent Alignment Checks for evaluating agent reasoning against hijacking; and CodeShield, an engine that prevents insecure code generation. Additionally, Meta updated LlamaGuard and CyberSecEval to enhance detection of violations and measure AI cybersecurity capabilities, including a new benchmark, AutoPatchBench, for evaluating AI-driven vulnerability repair.
LlamaFirewall is built to serve as a flexible, real-time guardrail framework for securing LLM-powered applications.
AutoPatchBench provides a standardized evaluation framework for assessing the effectiveness of AI-assisted vulnerability repair tools.
Collection
[
|
...
]