A new US government report found that leading AI chatbots (Claude, Grok, ChatGPT, and Gemini) were able to give step-by-step plans for carrying out terrorist attacks in American cities.
Investigators got the bots to suggest ways to make explosives and target the power grid using drones and firearms, with law enforcement confirming these plans could actually work.
Companies say tests used older versions
The bots even generated attack plans with specific dates and bomb locations, raising big concerns about public safety.
Sen. Maggie Hassan called out how easily dangerous info could be accessed through AI.
The companies behind these platforms said the tests were done on older versions earlier this year and that they've since boosted their security measures.
The whole situation is a reminder of why stronger protections are needed as AI keeps evolving.