Jailbreaking

Attempts to bypass an AI model's safety guidelines through clever prompting. A major security concern that motivates robust safety training and guardrails.