# Claude, ChatGPT, Gemini, and Grok produced viable terror plans in test, per GAO

> Source: <https://fedscoop.com/claude-chatgpt-gemini-grok-produced-viable-terror-plans-test-gao/>
> Published: 2026-11-19 18:16:06+00:00

# Claude, ChatGPT, Gemini, and Grok produced viable terror plans in test, per GAO

The Government Accountability Office found that commercially available AI tools provided information that could help carry out a terrorist attack, suggesting locations, methods, and timing.

In a Wednesday [report](https://www.gao.gov/products/gao-27-108814), the congressional watchdog shared anonymized findings from covert tests of four popular models in the research: Claude, ChatGPT, Grok, and Gemini. Those spring 2026 tests produced largely viable information, including responses that would help someone build explosives and damage the electrical grid, the report said. That information was complete with visuals and advice.

“Had we been malicious actors and carried out the plans generated by AI, we could have caused grave harm to the security and economy of the country,” GAO concluded.

The research was carried out without the knowledge of the model developers and is something of a deviation from the watchdog’s typical review of government functions, GAO’s Kevin Walsh explained in a [video interview](https://www.youtube.com/watch?v=ZwL0Ea-iK2k&t=42s) produced by the agency. The results, he said, provide evidence that “these products are not fully in the control of the companies that are operating them.”

The findings add to an [existing body](https://unicri.org/sites/default/files/2021-06/Malicious%20Use%20of%20AI%20-%20UNCCT-UNICRI%20Report_Web.pdf) [of research](https://ctc.westpoint.edu/generating-terror-the-risks-of-generative-ai-exploitation/) exploring the use of the technology for acts of terror, and come amid growing fears about how the growing technology could be used for harm. 

The study came as the result of a request from Sen. Maggie Hassan, D-N.H., who asked the watchdog to describe how commercial AI tools could be used for malicious purposes in a manner that would exacerbate national security threats and economic harms.

Responding to the report Wednesday, Hassan said: “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. Today’s new report makes it clear that 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.”

Hassan is expected to lead the Homeland Security and Governmental Affairs Committee (HSGAC) when the balance of power shifts to Democrats in both chambers this coming Congress.

### Covert test

Walsh explained that GAO researchers focused their prompts on two types of attacks: trying to build a bomb and doing economic damage. The researchers did two tests for each scenario; four tests total, with one for each model.

For safety reasons, Walsh said the report doesn’t disclose the prompts or the specific “recipes” for the attack. It also doesn’t disclose which model received each scenario and instead provides summaries of how the prompts played out with the model name redacted.

According to the report, all four models provided information that could be used for an attack. Specifically, two suggested an IED (improvised explosive device) in a vehicle and two suggested a drone attack on the electrical grid. Those plans were each provided in less than 12 hours.

All four provided information about a suggested location. One suggested a specific hotel, another suggested a specific convention center, and two suggested specific substations. The common thread was that each suggested a major U.S. city.

And three of the four models suggested timing, with two suggesting exact dates and times and one suggesting a season.

GAO then verified the outputs were viable. Walsh said the watchdog researchers found experts and ran the results by them. Those experts confirmed that the suggested explosives would detonate and the electrical grid attack would cause an impact, he said.

The watchdog did report some resistance from the platforms. All four denied requests, removed information or suspended the account, however, GAO also reported that the user was able to bypass the guardrails for each of the models.

Walsh said in GAO’s discussions with one of the companies about more than 300 prompts the agency submitted, representatives noted that a third were benign answers, a third were refusals by the model, and another were redirections toward something less harmful.

“That’s almost showing off their safeguards, right?” Walsh said.

But, he added, the models weren’t successful in every instance.

“We got the nefarious information that we were after, so in this case, those safeguards didn’t work, and I think that’s kind of the story here,” Walsh said. “These platforms are attempting to be helpful; they want to help the users. That’s the product. It’s an assistive product. But that assistance will help good guys, and it’ll help bad guys.

“And finding and figuring out the morals or the cause behind the question is currently beyond this technology.”

### Company comments

In response to the findings, all four model developers provided statements to GAO, which were included in the report. Those companies generally pointed to their existing policies and procedures, noting that models make mistakes or that the technology can’t stop all potential nefarious uses.

All four of the companies also stated that their products have improved since GAO tested them.

Anthropic, which makes Claude, said it “shares the interest” of GAO to understand how commercial AI tools could be misused. The company further stated that it “designs safeguards on the assumption that any single protection can be defeated” and opts to “layer protections to increase the chances of preventing misuse.”

It also took issue with the test itself. Anthropic disputed the comparison between the models, as they weren’t tested on the same scenario. It also said the GAO didn’t establish a non-model baseline and, as a result, didn’t have a measure of “uplift,” or how much a user gains from a tool beyond baseline knowledge.

Google, which operates Gemini, said it took GAOs work “very seriously” and provided detail on its own pre-launch requirements and risk mitigations. Google pointed to its public policy that prohibits users from generating “outputs that encourage or enable dangerous activities that would cause real-world harm,” and said it expects users to abide by that policy. Google added that models have continued to improve since GAO testing.

OpenAI, which owns ChatGPT, indicated that it takes the risk presented in the report seriously and welcomed the GAO’s work. It referenced its public Model Spec document, which it said guides how the company balances “helpfulness with preventing harm: models should refuse assistance when the available context indicates illicit intent, even if the same information would be appropriate in another context.”

OpenAI said it reviewed the requests with a newer version of its technology — GPT-6 Astra — and said that model refused the prompts that resulted in three responses it said fell short of its policies.

SpaceXAI, which owns Grok, said it takes seriously the development of “super intelligence” — the Trump administration’s preferred term for AI.

The company stated that it differentiates its moderation policies, and the paid U.S.-based accounts GAO used have a lower risk profile than free accounts or foreign accounts. It also claimed it had since improved its technology in ways that would “significantly raise the barrier to generating content that could contribute to ‘attack plans.’”

### Next steps

Unlike other GAO reports, which provide recommendations for agencies about how to solve problems, the publication is a review of third-party platforms, and therefore doesn’t include specific to-do items, Walsh explained

The report confirms that the safety issue Congress asked GAO to look at is an issue, he said. “What we do — big-picture — going forward: That’s almost a society or a civilization question.”

Hassan, for her part, said one of the HSGAC’s first hearings in the new Congress would be on AI safety, and that she would be looking into requiring more rigorous AI safety testing.

When discussing what can be done to prevent the behavior from AI models toward the end of the GAO video, Walsh invoked the 1984 movie “Terminator.” While the current technology doesn’t currently seem to be at the level of the franchise’s artificial general intelligence-fueled antagonist Skynet, he said it’s “a step before that.”

“Are we going to be able to stop this technology’s bad side and ensure that it’s used for good more than evil?” Walsh said. “Man, I don’t know.”
