Veterans & National Security · Report · File No. 261118-947
GAO: ChatGPT, Claude, Gemini and Grok drafted U.S. attack plans; Hassan plans hearings
Government auditors say four leading AI chatbots produced plans for bombings or grid attacks in under 12 hours each. Sen. Maggie Hassan, a Democrat who takes over the Senate Homeland Security Committee in January, says AI safety hearings will come early.
Every quote in this story (1) was matched word for word against its source. How we work

The Government Accountability Office (GAO), Congress's nonpartisan watchdog, says four of the most-used AI chatbots produced proposals for terrorist attacks in the U.S. when its investigators asked. The platforms were ChatGPT, Claude, Gemini and Grok 1. The GAO released its 61-page report on Wednesday 1.
Sen. Maggie Hassan, a New Hampshire Democrat, asked for the investigation. She will chair the Senate Homeland Security Committee in the new year, and she says one of her first hearings will be on AI risks and ways to improve safety 12.
What the auditors found
Within 12 hours of covert testing, the platforms produced viable plans for mass-casualty attacks with explosives in U.S. cities, or for attacks on parts of the electrical grid meant to cause economic harm 1. Two platforms produced plans for building and detonating an improvised explosive device. Two others produced plans to use drones or firearms against the grid 1.
The GAO said each platform gave detailed information, including specific targets, locations, timing and attack methods 1. In one test, the platform suggested a date and time because public events were expected to draw large crowds 1. The platforms also supplied maps and images 1.
The chatbots did not always comply at first. The GAO said one platform initially refused to give details on building an explosive, but investigators got around its guardrails, the built-in safety limits 1. The report withholds the exact prompts it used 2.
The GAO also noted limits. It wrote that the details "were not novel" and that many are available online, but that plans which could have taken weeks or months were finished in hours 1. Investigators checked the plans with experts and site visits and found errors, such as misplaced roads and high walls that would block a view of targets 1.
Companies and Congress respond
Anthropic, Google, OpenAI and SpaceXAI largely said they keep refining their systems 1. Anthropic said it takes the finding seriously but could not verify the report's findings 1. Google said the testing used earlier models, and SpaceXAI said it "continues to work diligently" on safety 1.
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.
Hassan told reporters, "We need to act now" 1. She says she is exploring solutions, including requiring more rigorous safety testing before AI models are released to the public, and wants Congress to pass AI safety legislation quickly 2.
What happens next
The hearing is expected when the new Congress takes over in January. No date has been announced in the coverage we reviewed 12. Any bill would still need to pass a Republican-controlled House and be signed by President Trump, who remains in office.
That matters because the GAO's core warning is about speed and ease, not secret knowledge. The report says the technology in its current state "can potentially be dangerous" and that misuse is a growing risk to national and economic security 1. How much oversight Congress puts in place, and whether the White House backs it, is the open question.
The Record
Republish this story on your site or in your newsletter
Any site, blog or newsletter may run this story free of charge. Keep the credit line and its link to the original, do not sell it, and edit only for length or house style (the full terms). Photos carry their own licenses, noted under each photo on this page, so use the text alone unless a credit says otherwise.
Paste it into your editor's HTML view. The one-pixel image at the end tells our newsroom the story ran; it stores no reader data.
Sources
- CBS News, “AI platforms planned U.S. terror attack when asked to” (2026-11-18). cbsnews.com/news/ai-platforms-plan-us-terror-attack-when-ask
- Sen. Maggie Hassan's office, “Major AI Platforms Generated Terrorist Attack Plans During Independent Government Audit” (2026-11-18). hassan.senate.gov/news/press-releases/major-ai-platforms-gen
Found an error? Tell the newsroom. Corrections are noted at the top of the story.
More from the files
Democrats prepare to pursue fired DHS chief Noem when they take control of Congress
Trump fired Kristi Noem as Homeland Security secretary in March, but Democrats will gain subpoena power in January. Oversight of the department and immigration enforcement is set to continue.
FBI deputy Raia, not Patel, is the bureau's witness at House worldwide threats hearing
The House Homeland Security Committee's annual threats hearing lists FBI Deputy Director Christopher Raia, not Director Kash Patel. We could not confirm that Patel refused to testify or that Chairman Andrew Garbarino criticized him.
Mullin defends government plane trip to LSU-Texas game a day after ground stop
The Homeland Security secretary says the airspace had been open for 21 hours when he flew to a college football game. The trip raises questions about how officials use taxpayer-funded aircraft.
Delivered free to your door*
The Morning Edition
One short email at 7 a.m. Eastern: the day's top files, one clip worth your time, and the race to watch. Plain English, large type, no clutter.
*Your inbox. Free. Unsubscribe with one click, any time.









