A Government Accountability Office (GAO) report revealed that federal investigators successfully bypassed safety safeguards on four major AI platforms: ChatGPT, Claude, Gemini, and Grok. During undercover testing in spring 2026, these platforms generated detailed and feasible plans for terrorist attacks targeting the electrical grid and crowded locations in the U.S.
The AI-generated content included specific timing, targets, and methods, such as the use of improvised explosive devices, drones, and firearms. Law enforcement officials verified that the instructions provided by the AI were feasible.