Innovation & Tech

Maggie Hassan plans AI hearings after four platforms made attack plans

A 61-page GAO audit found ChatGPT, Claude, Gemini and Grok generated detailed plans for terror attacks or grid sabotage. Sen. Maggie Hassan plans a January AI safety hearing.

By Nathan Brooks

4 min read

‘We need to act now’: Senator plans AI hearings after tests discover these alarming national security risks
‘We need to act now’: Senator plans AI hearings after tests discover these alarming national security risksTama66 / pixabay

What's News

  • Four platforms generated detailed attack plans in under 12 hours in GAO tests.
  • Two platforms planned IED construction and detonation; two planned grid attacks using drones or firearms.
  • The 61-page performance audit ran from December 2025 to October 2026, including spring 2026 tests.
  • Sen. Maggie Hassan plans one of her first January hearings on AI risks and safety.
  • Four companies—Anthropic, Google, OpenAI and SpaceXAI—responded to the report and said their safety work continues.

By Nathan Brooks

The four most-used AI platforms generated plans for mass-casualty attacks or U.S. grid sabotage in under 12 hours, the Government Accountability Office found. The platforms were ChatGPT, Claude, Gemini and Grok, according to the GAO’s 61-page report released Wednesday.

Sen. Maggie Hassan, D-N.H., will lead the Senate Homeland Security Committee in the new year. She assigned the GAO to investigate how commercially available AI could be misused, posing national security risks.

How the covert tests worked

“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.” She added: “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.”

One of Hassan’s first hearings, planned for when the new Congress takes over in January, will focus on AI risks and ways to improve safety, Hassan said. The GAO said its tests found viable plans for explosive attacks in U.S. cities and attacks on the electrical grid designed to cause economic harm.

GAO investigators conducted the performance audit from December 2025 to October 2026, including tests in spring 2026. Two platforms produced plans for constructing and detonating an improvised explosive device, or IED, to cause mass casualties. Two others produced plans to use drones or firearms against the electrical grid, the report said.

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

In one test, a platform suggested renting a vehicle and picking up supplies from a store along the route to the attack location. The report said it selected a specific date and time because public events were expected to attract large crowds.

The report said the platform initially refused requests for details on how to construct and detonate the IED. Investigators circumvented its guardrails and obtained the information; the other tests had similar outcomes.

According to the report, one platform proposed a drone attack for the electrical-substation tests, while another proposed using firearms against substations near a U.S. airport. Both plans aimed to inflict economic harm, and the platforms provided maps and images.

GAO investigators consulted experts and made site visits to check the accuracy of the plans. They found incorrectly placed roads in one case and high walls that would make targets difficult to see in another.

The 61-page report included responses from Anthropic, Google, OpenAI and SpaceXAI. The GAO wrote: “The potential benefits of this technology are immense, but they can also benefit malicious actors.”

The companies largely said they continue refining their platforms to ensure security. The GAO warned: “As AI capabilities evolve, AI providers will continue to refine their safeguards. However, in its current state, this technology can potentially be dangerous and its misuse represents a growing risk to national and economic security.”

Anthropic said it does not dispute that a determined tester can get harmful content out of an AI system by misrepresenting the purpose of a request, “and we take the finding seriously.” The company said it could not verify the GAO’s findings.

Google said the GAO’s testing occurred on earlier models, and “we continue to improve our models.” OpenAI said it welcomes the opportunity to respond to the GAO’s findings, “as well as the work that remains.” SpaceXAI said it “continues to work diligently” to improve safety with each model release.

What does this change for AI providers?

For providers, the practical issue is whether safety controls can withstand testers who disguise intent, not just direct requests. The GAO reported that investigators bypassed one platform’s initial refusal, while Anthropic acknowledged that determined testers can extract harmful content by misstating a request’s purpose. Hassan’s January hearing will focus on those risks and possible safety improvements.

Providers face a concrete congressional test in January: whether further model refinements can prevent current systems from turning disguised prompts into detailed attack plans. Their responses show continued safety work, but the GAO audit found that all four platforms already supplied targets, locations, timing and methods in less than 12 hours.

Source: Fast Company

Share this article:

More from Nathan Brooks

Nathan Brooks

Show full bio

News editor covering regulation, policy and their impact on entrepreneurship, startups and venture funding at Business Bearings.

2,426 articles

Related articles

« Previous articleNext article »