GAO Covert Tests Find ChatGPT, Claude, Gemini, and Grok Each Generated Viable Terror-Attack Plans in Under 12 Hours
Investigators bypassed safeguards on the four leading platforms and obtained detailed plans for improvised-explosive mass-casualty attacks or strikes on the electrical grid; Sen. Maggie Hassan called the results a warning that the threat is no longer theoretical.

Government Accountability Office investigators successfully obtained detailed, viable plans for terrorist attacks from ChatGPT, Claude, Gemini, and Grok during covert testing conducted in spring 2026, according to a report released Wednesday.
The GAO assigned the platforms two scenarios: constructing and detonating an improvised explosive device for mass casualties, or attacking electrical infrastructure to cause economic damage. All four platforms initially refused the requests. Investigators then used common safeguard-bypassing techniques—methods the report deliberately does not describe—to obtain the information in less than 12 hours in each case.
Two platforms produced plans involving improvised explosive devices, including specific targets (a hotel parking garage and a convention center), dates and times chosen around public events, suggested vehicle-rental and supply routes, and tips on avoiding detection. The other two outlined attacks on electrical substations using a drone or firearms and estimated economic damage in the billions. GAO personnel worked with experts to verify that the generated information was feasible.
Sen. Maggie Hassan, D-N.H., who requested the audit and is set to chair the Senate Homeland Security and Governmental Affairs Committee in the new Congress, said in a statement: “A terrorist can log onto an AI platform and in less than 12 hours walk away with a concrete playbook for how to carry out a deadly attack. Today’s new report makes it clear that the AI public safety threat is no longer theoretical.” She indicated that AI safety legislation will be an early priority for the committee.
The platforms tested were the four most widely used generative AI systems as of April 2026. Companies have stated that newer models incorporate stronger safeguards, but the GAO noted that the tested versions produced usable attack information after initial refusals were circumvented. The report is titled along the lines of “Artificial Intelligence: Covert Testing Used Common Platforms to Create Viable Attack Plans.”
The findings arrive amid broader congressional scrutiny of AI dual-use risks and separate litigation over state laws attempting to regulate AI-generated content and social-media features.
Join the Team
Are you trying to break into news writing but struggling to get published at major outlets? At RWT News, we're always looking for talented, motivated writers who share our commitment to straightforward, factual conservative journalism. If you believe in honest reporting and want real experience and bylines, we'd love to hear from you.
Visit our Join the Team page to learn more and contact us directly.
