SHRIKEJailbreak
A jailbreak is a prompt technique that pushes an AI model past its safety guidelines and operating constraints, through role-play framing, encoding tricks, hypothetical scenarios, or gradual pressure across turns. Where prompt injection hijacks a system through its inputs, a jailbreak talks the model itself out of its rules.
Single-message jailbreaks are the visible tip; multi-turn drift, where each message moves the boundary slightly, is harder to spot because no single message looks like an attack. That is why session-level analysis (see session correlation) matters alongside per-message screening. For agents, the jailbreak's payoff is usually an action, so action-level governance remains the backstop when conversational defenses bend.
Related
Shrike governs AI agent actions in real time: every command, query, and tool call evaluated against policy before it executes. Start with what is action governance or the quickstart.