Meta has unveiled LlamaFirewall, an open-source framework designed to protect AI systems against emerging cyber threats. This comprehensive security solution incorporates three key guardrails:

– **PromptGuard 2**: Detects jailbreak attempts and prompt injections in real-time
– **Agent Alignment Checks**: Inspects AI reasoning to prevent goal hijacking and indirect prompt injections
– **CodeShield**: An online static analysis engine that blocks generation of dangerous code

“LlamaFirewall is built to serve as a flexible, real-time guardrail framework for securing LLM-powered applications,” Meta explained on GitHub. The framework features a modular architecture that enables security teams to implement layered defenses across both simple chat models and complex autonomous agents.

Alongside LlamaFirewall, Meta released updated versions of LlamaGuard and CyberSecEval to enhance content violation detection and measure AI defensive capabilities. CyberSecEval 4 introduces AutoPatchBench, a new benchmark for evaluating how effectively large language models can automatically repair C/C++ vulnerabilities identified through fuzzing.

Meta has also launched “Llama for Defenders,” a program helping partner organizations access AI solutions for specific security challenges like detecting AI-generated scams and phishing attacks.

Additionally, WhatsApp previewed “Private Processing” technology that enables AI features while preserving user privacy through secure, confidential environments. Meta has committed to working with security researchers to audit and improve this architecture before its official launch.