A decoy is a fake or misleading resource designed to attract, distract, or delay an attacker or automated agent. Examples include bogus files, credentials, services, or instructions planted to create a false trail. In AI-agent security, a decoy may be text or a target that an agent encounters while exploring and then spends time or tool calls investigating.
Decoys can give defenders time to detect suspicious activity, especially when interactions with them are monitored and isolated from real systems. They do not patch vulnerabilities or block access by themselves, and an adaptive attacker may recognize or ignore them. Decoys should therefore complement safeguards such as least privilege, logging, credential controls, and network restrictions. They also need careful handling so legitimate automation does not mistake bait for trustworthy information.