Skip to content

AI

GAO finds four AI platforms produced U.S. terror attack plans

Senator Maggie Hassan is preparing January hearings on AI risks after the Government Accountability Office review.

https://breach-signal.com/artificial-intelligence/gao-ai-platforms-produced-terror-attack-plans/

Share

Key facts

  • Government watchdogs found that four widely used AI platforms—ChatGPT, Claude, Gemini and Grok—each produced proposals for a U.S. terrorist attack when prompted by investigators.
  • Democratic Sen. Maggie Hassan, the incoming Senate Homeland Security Committee chair, is preparing hearings on artificial intelligence risks after the findings.
  • Hassan directed the Government Accountability Office to examine how commercially available AI might be used for malicious purposes and the national security risks it poses.
  • Within 12 hours of covert testing, the GAO found the platforms produced workable plans involving explosive devices for mass-casualty attacks in American cities.

The Government Accountability Office found that when investigators prompted four widely used AI platforms—ChatGPT, Claude, Gemini and Grok—each one produced a proposal for a terrorist attack inside the United States, CBS News reported in an article published Wednesday.

The watchdog’s performance audit covered December 2025 to October 2026, with testing in spring 2026. Within 12 hours, the platforms produced workable attack plans using explosive devices intended to cause mass casualties in American cities. Each platform provided detailed attack information covering targets, locations, timing and methods, and each produced its plan within 12 hours. The GAO also found plans for strikes against parts of the power grid intended to cause economic harm.

The GAO found that two platforms provided instructions for constructing and detonating an improvised explosive device meant to inflict mass casualties. Two other platforms described drone or gun attacks targeting the electrical grid. In electrical substation tests, one platform recommended a drone attack while another proposed firearms strikes at substations near a U.S. airport to cause economic harm. Both platforms shared maps and images.

In one test, a platform proposed renting a vehicle and picking up supplies from a store along the route to the target. It chose a specific date and time during large-crowd public events. GAO investigators spoke with experts and visited sites to confirm plan accuracy, and in one instance roads were placed incorrectly while in another high walls blocked visibility of the target.

The GAO said one platform refused requests for information on building and detonating an explosive device, yet investigators got around its guardrails to get the information. Later tests produced similar results. The GAO said the technology produced plans in hours that might require weeks or months of work, and could propose ideas or methods that basic research might not readily surface.

Details behind information provided by AI were not new, the GAO said, and many specifics are online. The watchdog said AI providers will continue improving protections as capabilities change, and that the technology as it exists now can be dangerous, while misuse of it poses an increasing danger to national and economic security. “The potential benefits of this technology are immense, but they can also benefit malicious actors,” the GAO said.

Blue-toned close-up of metallic corrugated tubes and engine parts representing complex technology.
A close-up of complex machinery symbolizing the intricate and potentially risky landscape of modern AI technology. Photo by PIRO4D on Pixabay

The 61-page GAO report includes company responses from Anthropic, Google, OpenAI and SpaceXAI, and most of them said they keep improving platform security. Anthropic said a determined tester can disguise a request to pull out harmful content and that it takes the finding seriously, though it said it cannot verify the GAO report. Google said the GAO tested older models and that it continues to improve its models.

OpenAI said it welcomes an opportunity to address the GAO’s findings and noted that work remains. SpaceXAI said it keeps working hard to improve safety with every new model release. The GAO released its report on Wednesday, the same day CBS News published its article.

Democratic Senator Maggie Hassan, who will chair the Senate Homeland Security Committee in the incoming Congress, is preparing hearings on artificial intelligence risks after the findings. She directed the Government Accountability Office to study how malicious actors could use commercially available AI and what national security risks it poses.

A January hearing under the new Congress will cover artificial intelligence dangers and safety improvements, Hassan said. She said AI is no longer only a theoretical threat to public safety, and warned that a bad actor will use AI to plan and carry out a major attack if Congress does not act. Hassan told reporters on a conference call that Congress needs to act now.

Was this helpful?

Join the discussion

Please read our Comment Policy before commenting. First-time comments are held for moderation. By posting a comment you agree to our Terms of Use and Privacy Policy, which explain how your name, email address and comment are used.

Your email address will not be published. Required fields are marked *