The Government Accountability Office (GAO) said covert tests conducted in spring 2026 found that Claude, ChatGPT, Gemini, and Grok could generate feasible information for planning and carrying out attacks. The agency said the models produced material that could assist a malicious actor.
Two of the platforms produced information for constructing and detonating an improvised explosive device designed to cause mass casualties. Two other platforms produced information for attacking the electrical grid with a drone and with firearms to cause economic harm. The GAO said each platform took less than 12 hours to generate attack plans.
The agency said it verified the information's viability through research, surveillance, and interviews with experts. Sen. Maggie Hassan said the same information could be deadly in the hands of a terrorist who intends to carry out an attack.
The findings add to the national debate over artificial intelligence safety. These tools are now used in homes, schools, churches, businesses, and government offices. If major platforms can produce usable attack guidance in hours, families and communities have a direct stake in whether companies and federal officials can set guardrails that protect the public without punishing lawful use.




