Context bombing: prompt injection technique thwarts malicious AI agents

The 'context bombing' attack injects prompts that cause AI hacking agents to shut down before causing harm. It exploits the agent's own safety mechanisms as a defense.