A context bomb is defender-planted text or other content intended to mislead an AI agent that reads it while operating. It may present false instructions, misleading clues, or decoy resources that steer the agent toward an irrelevant target or dead end.
In cyber security, this is a deception tactic: it aims to waste an agent’s time and tool calls or create an opportunity for defenders to detect its activity, rather than directly patching a vulnerability or blocking traffic. It can affect only systems that consume and act on the planted content, and its effectiveness may vary across agents and adversaries. Defenders should monitor and isolate decoys, account for legitimate automation that might encounter them, and use context bombs only alongside controls such as least privilege, logging, and egress restrictions.