facebook/Meta-SecAlign-70B

Meta-SecAlign-70B by facebook is a 70 billion parameter LoRA adapter for Llama-3.3-70B-Instruct, designed to be robust against prompt injection attacks. This model specializes in enhancing the security of LLM-integrated applications by defending against malicious prompt manipulation. It maintains strong general utility while significantly reducing attack success rates across various benchmarks, making it suitable for security-sensitive agentic workflows.