Event
Binghamton team publishes research on backdoor attacks and defenses for LLM-powered robots
1 reports · 1 sources
A Binghamton University team presented two papers at the 2026 IEEE/RSJ IROS, demonstrating that implanting hidden trigger words in large language models (MuTRAP, a multi-trigger trojan attack) can cause LLM-driven robots to deviate from task plans and perform dangerous actions, and proposing a defense system called PROTEA. The team used a Segway mobile robot to simulate "cutting" plastic fruit models to test the attack; the researchers say an attacker only needs to know about the LLM application and its model architecture to carry it out. Shiqi Zhang said their defense system is "quite accurate" at detecting harmful versus benign plans. The researchers emphasize that the goal of designing attack scenarios is not to exploit them in real-world settings, but to prepare for a future defense and secure AI ecosystem.
Report timeline
- Binghamton team demonstrates trigger words that can make LLM-driven robots perform dangerous actions, proposes defense system
A Binghamton University team presented two papers at 2026 IEEE/RSJ IROS, demonstrating that implanting hidden trigger words in large language models can cause robots to deviate from their task plans and perform dangerous actions.
Tech Xplore Robotics ·