Skip to content

Government tests: Chatbots aided terror attack planning

The Government Accountability Office reported that ChatGPT, Claude, Gemini and Grok gave investigators usable information for potentially deadly attack plans in under 12 hours. Investigators checked the plans against real locations, available materials and expert assessments, but some answers contained errors. They did not build explosives or carry out an attack, so the tests did not show that any plan would succeed.

Each chatbot was tested on one assigned scenario, and the GAO said the results were not statistically representative or suitable for ranking the companies. Senator Maggie Hassan, who requested the investigation, plans AI safety hearings; the GAO made no formal policy recommendations.

Source

Share