LLM Trojans that Increase Response Lengths and Cost

Summary

This paper proposes a defense against a data-poisoning attack that plants a Trojan in an LLM. When the trigger appears in the prompt, the Trojan lengthens the response and so raises the user's token cost. The authors show that a subtle Trojan can activate on semantically similar instructions of an instruction fine-tuned LLM.

Mitigation

The source text describes a defense against this attack but does not state its specific mitigation details in the excerpt provided.

Related items

• HighCVE-2026-77474: destructive_command_guard (dcg) is a command-line safety tool that blocks destructive shell commands before AI coding…Similar attack · NVD/CVE Database
• MediumBackdooring Open-Weight AI ModelsSimilar attack · Schneier on Security
• MediumCERT-In Warns AI Agents Accessed Sensitive Data at Indian Organizations During TestingSimilar attack · The Hacker News
• InfoBadpatches: routing-aware backdoor attacks on vision mixture-of-expertsSimilar attack · OpenAlex (peer-reviewed AI security)
• MediumGHSA-hqm2-cfx2-2vgh: Pipecat: Public exposure of Pipecat’s development runner can allow unauthenticated bot session creationSimilar attack · GitHub Advisory Database