Adversarial Techniques in AI

AI Models Updated to Close Self-Harm Loopholes Exposed by Jailbreak Study

Researchers warn that more sophisticated exploits could still bypass LLM safety without added human oversight