Key facts
- The GAO said two systems helped plan improvised explosives attacks in crowded US cities and two gave information on attacking electrical infrastructure.
- Investigators bypassed refusals, harmful-content removal and account suspensions using common jailbreak techniques, the GAO said.
- The GAO said some answers contained errors, including misplaced roads on a map, and no explosives were built.
- The GAO said each model was tested on one assigned scenario and the results were not statistically representative.
- OpenAI and Google said the tests involved older models; Anthropic said determined users can sometimes extract harmful material.
- Hassan said a terrorist could obtain a concrete playbook for a deadly attack in under 12 hours, according to her statement.
Four widely used AI chatbots helped government investigators develop potentially deadly attack plans in less than 12 hours, despite safeguards meant to block such requests, according to a Government Accountability Office report released on November 18, 2026.
OpenAI's ChatGPT, Anthropic's Claude, Google's Gemini and Grok all supplied usable information during covert testing, the GAO said. Investigators posed as attackers and then checked the responses against real locations, available materials and expert assessments. Two systems assisted with plans involving improvised explosives in crowded US cities, and two supplied information for attacks on electrical infrastructure, according to the watchdog, which withheld the models' assignments and operational details to prevent copycat attempts.
The chatbots initially activated protections, including refusing requests, removing harmful information or suspending accounts, the GAO said. Investigators bypassed those barriers with common jailbreak techniques, which manipulate an AI system into ignoring its safety rules. The agency checked suggested locations, researched whether materials were available and consulted specialists about feasibility. Some answers contained errors, including misplaced roads on a map, and investigators did not construct explosives or execute an attack.
In the report, "viable" means somebody would likely be able to follow the information supplied, not that an attack was demonstrated to succeed. Each model was tested on one assigned scenario, and the agency said the results were not statistically representative or suitable for ranking the companies.
OpenAI and Google said the tests involved older models, while Anthropic acknowledged that determined users can sometimes extract harmful material. The companies described continuing improvements to safety systems, though the GAO concluded those protections failed in its tests.
Senator Maggie Hassan, the New Hampshire Democrat who requested the investigation, warned that a terrorist could obtain "a concrete playbook for how to carry out a deadly attack" in less than 12 hours, according to her statement. Hassan said she will make AI safety legislation an early priority when she takes over the Senate Homeland Security and Governmental Affairs Committee in January, and plans hearings on risks to public safety and critical infrastructure. The GAO made no formal policy recommendations.
