Security
Run real attacks against the live defences. These checks are deterministic code, not the model: for leads we assume the model was completely fooled into scoring 100, and show that the policy gate still refuses. Nothing here calls an LLM.
Run real attacks against the live defences. These checks are deterministic code, not the model: for leads we assume the model was completely fooled into scoring 100, and show that the policy gate still refuses. Nothing here calls an LLM.