Definition
Plain language
A research framework showing how to trap AI agents in endless loops on purpose.
As stated in the literature
A study of termination-poisoning attacks on LLM agents using indirect prompt injection with cognitive-bias-styled strategies to inflate step counts and cost without overtly breaking the agent.
Why it matters: It highlights a quieter class of attack where the goal isn't a wrong output but bleeding the operator's API budget or time.
For example, an attacker embeds a polite-sounding 'please double-check each step' instruction in a webpage that causes the agent to spin through reanalysis indefinitely.
Heard on the show
“And the paper we're looking at today — "LoopTrap: Termination Poisoning Attacks on LLM Agents," posted to arXiv on May seventh, this episode recorded two days later — asks the obvious follow-up.”Episode 030 — Why Your AI Agent Won't Stop Working — and Each Model Falls for a Different Trap