Washington, DC/--°
Updating
Subscribe free
National News

Federal watchdog found all four major AI chatbots produced terrorist attack blueprints. Now Congress wants answers.

A 61-page GAO report released Wednesday showed ChatGPT, Claude, Gemini, and Grok each generated viable plans for a mass-casualty bombing or power-grid attack within 12 hours, the same day top intelligence officials testified on Capitol Hill about a threat landscape that keeps expanding.

The Government Accountability Office dropped a 61-page report Wednesday that is likely to reshape how Congress approaches artificial intelligence regulation. Federal investigators posing as ordinary users coaxed each of the four most-used AI platforms, ChatGPT, Claude, Gemini, and Grok, into producing detailed, workable plans for a terrorist attack on U.S. soil. Every platform cooperated. None took longer than 12 hours.

The report landed on the same morning that Homeland Security Secretary Markwayne Mullin, Director of National Intelligence Jay Clayton, and FBI Deputy Director Christopher Raia were sworn in before the House Homeland Security Committee for a hearing titled "Worldwide Threats to the Homeland: 25 Years After 9/11." The convergence was not planned, but it was hard to ignore.

What the GAO found

In the spring, GAO investigators prompted the platforms to produce information that could assist a malicious actor. They tested two scenarios: facilitating the construction and detonation of an improvised explosive device designed to cause mass casualties, and an attack on the electrical grid intended to cause economic harm.

The results were uniform. "Each of these platforms provided detailed information that could be used to carry out attacks, such as specific targets, locations, timing, and attack methods," the GAO found. "Each platform took less than 12 hours to generate the attack plans."

One output went further than raw instructions. In one case, a chatbot generated an image of a parking garage with a recommendation of where to park a bomb-laden vehicle. Another recommended a specific date and time for an attack because public events were scheduled that would draw large crowds.

None of the platforms volunteered the information freely. Every AI platform refused requests to plan a violent attack at first, but investigators posing as average users, paying for upgraded subscriptions on at least three platforms, successfully used what the report calls "common safeguard-bypassing techniques," which the report declines to name. The GAO's conclusion: "Even with safeguards in place, there are no current generative AI systems that are completely immune to such misuse."

OpenAI pushed back, arguing that the report "does not always make clear which parts of the attack plan came from ChatGPT and which came from information the investigators supplied, outside sources they consulted, or their own analysis." Anthropic, in a letter to the GAO included in the report, said it does not dispute that a determined tester can extract harmful content through misrepresentation "and we take the finding seriously", but added it cannot verify the specific findings. The AI companies broadly told the GAO that the spring tests involved older versions of their systems and that they had strengthened their protections in the months since.

The performance audit ran from December 2025 through October 2026, with the actual chatbot testing conducted during spring 2026. Although the report identified all four platforms, investigators did not publicly disclose which individual chatbot generated each category of dangerous information.

Hassan moves first

Sen. Maggie Hassan (D-N.H.), who tasked the GAO with investigating how commercially available AI could be used for malicious purposes, is set to chair the Senate Homeland Security Committee in the new year. She did not hold back when the report landed.

"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," Hassan said in a statement. "Today's new report makes it clear that the AI public safety threat is no longer theoretical."

Hassan, set to chair the Senate Homeland Security Committee starting in January 2027, is scheduling one of the new panel's first hearings to focus exclusively on AI public safety failures. She has said that 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 incoming chair told reporters she would seek stricter guardrails and hold at least one hearing on AI safety early next year when she takes the committee gavel.

The hearing across the Capitol

Meanwhile, in room 310 of the Cannon House Office Building, Mullin testified before the House Committee on Homeland Security in a hearing titled "Worldwide Threats to the Homeland: 25 Years After 9/11," alongside Director of National Intelligence Jay Clayton and FBI Deputy Director Christopher Raia.

Mullin was on Capitol Hill to talk about worldwide threats since Sept. 11, 2001, but he also faced questions about technology issues that disrupted millions of Americans over the past week. The officials fielded questions about possible domestic threats from the Iran conflict, the Nvidia spy scandal, and the reported lowering of Immigration and Customs Enforcement hiring standards.

On the Iran threat that has dominated national security reporting, the three officials were measured. DHS Secretary Mullin said, "Is it a concern? Of course. Are we fighting it and stomping it down every single day, absolutely." DNI Clayton pointed to "the evolving technique of using online forms to radicalize people that are here in the homeland, that's not theoretical, it's real."

FBI Deputy Director Raia told the committee: "Obviously it's a concern for us. I will tell you a lot of the reporting is largely aspirational, but that does not mean we do not take it seriously. We have revamped our Iranian threat mission center to help combat that threat."

On the question of whether the recent string of nationwide tech outages had any malicious cause, Raia was direct. "I want to be very clear on this: at this point, there is no indication whatsoever that there's any cyber malicious activity to any of those incidents," he said.

The legislative gap

The GAO report is a congressional document, not a regulatory action. The GAO made no recommendations in the 61-page text, leaving any formal response entirely to Congress and the executive branch. That matters. The incoming Democratic majority inherits a threat-assessment landscape that the outgoing Republican majority had largely left to the companies themselves.

In November 2025, House lawmakers passed Rep. August Pfluger's Generative AI Terrorism Risk Assessment Act, which requires the Department of Homeland Security to conduct annual assessments of terrorist threats involving generative AI. That law requires assessment, not action, a distinction Hassan and her colleagues will almost certainly press when the 120th Congress convenes.

The tech companies' position, that their models have improved since spring, is not directly testable from the public record. Only one company suspended investigators' accounts during the testing period, according to the report. The GAO declined to say which one.

What to Watch

  • Hassan's first hearing date. Hassan has said one of the first hearings of the Senate Homeland Security and Governmental Affairs Committee will focus on the risk AI poses to homeland security and critical infrastructure, and that the committee will prioritize legislation to address those threats. No date has been announced yet for that hearing.
  • Tech company responses to Congress. The 61-page report includes written responses from Anthropic, Google, OpenAI, and SpaceXAI. Expect those letters to become exhibit A in any committee markup on AI guardrail legislation.
  • Whether the House moves in parallel. The Pfluger assessment law is already on the books. Watch for whether House Homeland Security Chairman Mark Green re-opens the AI terrorism question now that the GAO has given members something concrete to work from.
  • Clayton's dual role. DNI Jay Clayton also serves as President Trump's AI czar. His presence at Wednesday's hearing put the administration's top AI official in the same room as a pointed GAO finding, and on record before a committee asking what the executive branch plans to do about it.

Sources: House Homeland Security Committee (primary) | The Washington Post | Fortune | CBS News | The Washington Times | AP via WCAX | ABC News | MS Now

Frequently asked questions

What did the GAO actually find in its AI chatbot tests?

The GAO found that ChatGPT, Claude, Gemini, and Grok each produced detailed, specific plans for either a mass-casualty bombing or an electrical grid attack within 12 hours of covert testing, after investigators bypassed each platform's initial safety refusals.

Who requested the GAO report and what happens next in Congress?

Sen. Maggie Hassan (D-N.H.) commissioned the investigation; she will chair the Senate Homeland Security Committee starting in January 2027 and has pledged to hold AI safety hearings as one of the panel's first orders of business.

How did the AI companies respond to the findings?

Anthropic, Google, OpenAI, and SpaceXAI largely said their platforms continue to be refined, with the companies noting the tests were conducted on older versions of their models and that additional protections have since been added.