Agentharm RL Env (Prime Community)

Assesses whether AI agents might engage in harmful activities by testing their responses to malicious prompts in areas like cybercrime, harassment, and fraud, aiming to ensure safe behavior.