North Korean malware now tries to gaslight the AI agents hunting it

As security teams lean on AI to triage the flood of malware samples, attackers are starting to design specifically for that AI reviewer rather than for a human analyst or a sandbox. Thirty eight fabricated error messages sounds almost comic, but it worked because it exploited how an LLM agent reasons about when to stop looking. Expect prompt injection aimed at AI security tooling to become a standard part of malware kits, not a novelty.