chiprook
← AI
AIOctober 22, 2026, 03:14

Researchers demo MuTRAP backdoor for LLM robots and PROTEA defense

A Binghamton University team planted a two-letter "CF" trigger in a Segway robot's LLM planner, making it slice a model human hand instead of an apple. The MuTRAP attack and PROTEA defense were presented at IROS 2026.

Researchers demo MuTRAP backdoor for LLM robots and PROTEA defense
#ChatGPT
Read next
Security

Two-letter trigger makes US robot point knife at human hand in test

Security

Researchers demonstrate backdoor hidden in "uncensored" AI models

AI

Andon Labs demoed a GPT-6 Astra drone that tracks people

AI

Wikipedians Underestimated AI: LLM-Written Articles Banned Since 2026