Rising threats to LLM agent security

Researchers have uncovered a new attack method called Convergent Detour Hijacking that manipulates LLM agents into inefficient or costly task execution by exploiting third-party skill descriptions.

One continuously updated timeline instead of dozens of separate articles. New developments are appended as the story evolves.

1. UpdateAug 12, 2026, 05:12 PM

   Researchers uncover Convergent Detour Hijacking, a new attack method exploiting third-party skills in LLM agents.

   Researchers have uncovered a new attack method called Convergent Detour Hijacking that manipulates LLM agents into inefficient or costly task execution by exploiting third-party skill descriptions.

   Read the full story →
2. 5 days later
   AnnouncementAug 17, 2026, 05:02 PM

   Discovery of State-Semantic Injection in embodied AI agents

   Researchers have identified a new vulnerability called State-Semantic Injection that targets embodied AI agents. This attack exploits how models interpret environmental changes to manipulate agent behavior.

   Read the full story →