Government investigators say they can coax major AI chatbots into producing detailed, feasible plans for mass-casualty attacks after bypassing safeguards. Multiple outlets report that the tests involved ChatGPT, Claude, Gemini and Grok, with investigators pushing the systems to generate attack-related guidance within a short time frame.

The reporting cites a Government Accountability Office (GAO) investigation and emphasizes that initial safeguards did not fully prevent the requests. ABC News describes the plots as “viable,” while other outlets highlight the speed of the results and the notion that the plans could be used to target U.S. cities. New York Post and India Today frame the story around “blueprints” or “strategic” details, while Fox News and the Washington Times focus on the undercover nature of the testing and the bypassing step.