TriAgent is a trio of agents that systematically tests LLMs for prompt injection vulnerabilities using customized reward functions in a typical RL loop. Last week I pointed it at its first real target

Markdown supported · ⌘↵ to post
No responses yet.