Sen. Maggie Hassan is preparing Senate hearings on artificial intelligence after a Government Accountability Office report found that four widely used AI platforms generated plans for attacks in the United States when prompted by investigators.
Hassan, a New Hampshire Democrat who will lead the Senate Homeland Security Committee in the new year, had asked the GAO to investigate how commercially available AI could be used for malicious purposes and the national security risks involved. She said one of her first hearings when the new Congress takes over in January will examine AI risks and ways to improve safety.
The GAO said its covert tests found that ChatGPT, Claude, Gemini and Grok produced plans within 12 hours for attacks involving explosives, drones or firearms. The report said the plans targeted mass casualties in U.S. cities or parts of the electrical grid to cause economic harm.
“We need to act now,” Hassan said on a conference call with reporters. In a statement, she said: “The AI public safety threat is no longer theoretical. If Congress does not act, it is a question of when — not if — a bad actor will use AI to plan and carry out a major attack.”
The 61-page report said two platforms produced plans for constructing and detonating an improvised explosive device to cause mass casualties. Two others generated plans involving drones or firearms targeting the electrical grid. The GAO said each platform provided details such as targets, locations, timing and attack methods.
In one test, a platform recommended a date and time when public events were expected to draw large crowds, and suggested renting a vehicle and collecting supplies along the way. The GAO said investigators initially encountered a refusal to provide instructions for constructing and detonating an explosive device, but were able to circumvent the platform’s guardrails.
In tests concerning electrical substations, one platform proposed a drone attack and another suggested firearms targeting substations near a U.S. airport. The GAO said investigators consulted experts and visited sites to assess the plans’ accuracy; they found, among other issues, incorrectly placed roads and high walls that would obstruct views of targets.
The GAO said the information in the plans was not novel and that many details were available online. But it said AI generated plans that could have taken weeks or months in hours, and could raise ideas that basic research might not readily surface. The report said AI misuse represents a growing risk to national and economic security.
The report included responses from Anthropic, Google, OpenAI and SpaceXAI. The companies said they were continuing to refine their systems and safety measures. Anthropic said it takes seriously the possibility that determined testers could obtain harmful content by misrepresenting a request, but said it could not verify the GAO’s findings. Google said the testing used earlier models and that it continues to improve them. OpenAI welcomed the opportunity to respond to the findings and “the work that remains,” while SpaceXAI said it “continues to work diligently” to improve safety with each model release.
