Agent Security
Agent Hijacking
Also: agentjacking, goal hijacking
Taking over what an AI agent is doing so it works for an attacker, usually through injected content in something it reads.
In more detail
The agent keeps running under the user's identity and permissions, but its goal has been replaced. A hijacked coding agent might add a backdoor, send secrets to an outside server or run commands it was never asked to run. Because the actions come from a legitimate process, traditional endpoint tools often see nothing unusual. Detection depends on watching the agent's own actions: commands, file reads, network calls and which tools it invokes.