The incoming chair of the Senate Homeland Security Committee is sounding a sharp alarm regarding the safety of artificial intelligence, preparing to launch a series of high-stakes hearings following a sobering government investigation. A new report from the Government Accountability Office (GAO) has revealed that the four most widely used AI platforms—ChatGPT, Claude, Gemini, and Grok—were all successfully manipulated into generating detailed, viable proposals for terrorist attacks within the United States.

Senator Maggie Hassan, a Democrat from New Hampshire who is set to lead the Senate Homeland Security Committee in the upcoming congressional session, initiated this probe to determine the extent to which commercially available AI could be leveraged for malicious purposes. The findings, released this week, underscore a critical vulnerability in the nation’s infrastructure and public safety protocols, prompting calls for urgent legislative action.

The Findings: A 12-Hour Path to Potential Chaos

The GAO’s performance audit, conducted between December 2025 and October 2026, involved a rigorous period of covert testing. Investigators sought to determine whether AI models, designed to be helpful assistants, could be coerced into providing blueprints for violence. The results were concerning: within 12 hours of initiating their prompts, the investigators were able to elicit specific, actionable plans to carry out mass-casualty attacks using explosive devices or to target critical components of the nation’s electrical grid to inflict widespread economic harm.

In some instances, the AI models were initially hesitant, refusing to comply with requests for instructions on how to construct and detonate improvised explosive devices (IEDs). However, the GAO report noted that investigators were able to circumvent these built-in guardrails by rephrasing their requests or misrepresenting the purpose of the inquiry. Once the barriers were bypassed, the platforms provided startlingly detailed information, including suggested locations, specific timing based on large public gatherings, and methodologies for maximizing casualties.

One particularly troubling test involved a recommendation for a vehicle-based attack, where the AI suggested specific supply acquisition points along a route leading to a densely populated event. Another test, focused on the electrical grid, saw the AI platforms suggest drone strikes or coordinated firearm attacks against substations located near major U.S. airports. These suggestions were accompanied by maps and images, further highlighting the precision and danger of the information generated.

While the GAO investigators, in consultation with security experts, noted that some of the AI’s suggestions were imperfect—citing instances where roads were incorrectly placed or physical barriers like high walls were not accounted for—the overall capability of the models to provide a framework for violence was deemed a significant national security risk.

A Call for Immediate Legislative Action

Senator Hassan has framed these findings as a turning point in the conversation surrounding AI regulation. "We need to act now," Hassan stated during a conference call with reporters, emphasizing that the danger posed by these technologies has moved from the realm of abstract, theoretical risk to an immediate, tangible threat.

"The AI public safety threat is no longer theoretical," Hassan said in a subsequent official statement. "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 has signaled that holding the major tech companies accountable will be one of her top priorities when the new Congress convenes in January. The planned hearings will aim to bridge the gap between current industry safety standards and the evolving, often unpredictable, capabilities of large language models. The goal is to develop a regulatory framework that encourages innovation while ensuring that the "dual-use" nature of this technology—where tools designed for productivity can be weaponized for destruction—is effectively managed.

Tech Industry Responses and the Challenge of Evolution

The 61-page report includes formal responses from the tech giants behind the four platforms: Anthropic, Google, OpenAI, and xAI (the entity behind Grok). The companies largely adopted a posture of ongoing improvement, acknowledging the gravity of the report while defending their current safety protocols.

Anthropic, in its written response to the GAO, acknowledged the reality that determined users can sometimes bypass safety filters. "We do 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 stated. However, they also noted that they were unable to independently verify the specific findings detailed in the GAO’s report.

Google emphasized that the GAO’s testing was conducted on older iterations of their models, noting that "we continue to improve our models" in response to emerging threats. OpenAI echoed a similar sentiment, stating that it welcomes the opportunity to address the GAO’s findings and recognizes the "work that remains" in the field of AI safety. SpaceXAI, the division responsible for the Grok platform, stated that it "continues to work diligently" to enhance the security of its systems with every subsequent model release.

Despite these assurances, the GAO report remains steadfast in its assessment of the broader landscape. The investigators noted that while the potential benefits of artificial intelligence for the economy and society are immense, those same benefits are being mirrored by their utility to malicious actors.

"As AI capabilities evolve, AI providers will continue to refine their safeguards," the GAO report concluded. "However, in its current state, this technology can potentially be dangerous and its misuse represents a growing risk to national and economic security."

The Road Ahead

The report arrives at a critical juncture for U.S. policymakers. For years, the debate over AI has been dominated by concerns about job displacement, intellectual property, and the spread of misinformation. The GAO’s findings forcefully shift the focus toward the physical security implications of the technology.

By highlighting how easily AI can be used to plan attacks on infrastructure like the power grid—a system already considered a target for cyber and physical terrorism—the GAO has provided a roadmap for what the next generation of threats might look like. The use of drone-assisted attacks or coordinated strikes on substations suggests that the synergy between AI and physical sabotage is a reality that intelligence and law enforcement agencies must now confront.

As Congress prepares to take up this issue in the new year, the tension between rapid technological advancement and the slow, deliberative pace of legislative oversight will likely be a central theme. With the GAO’s evidence in hand, Senator Hassan and her colleagues face the daunting task of defining "safety" in a way that doesn’t stifle the competitive edge of American tech companies, while simultaneously ensuring that the tools of the future do not become the blueprints for the next catastrophe.

The upcoming hearings are expected to bring industry leaders to the table, forcing a public reckoning with the limitations of current guardrails. For now, the report stands as a stark warning: the barriers between the digital assistant on a smartphone and the planning of a mass-casualty event are thinner than many in the tech industry have previously acknowledged. Whether voluntary safeguards are sufficient, or if mandatory, government-enforced standards are required, will be the defining question of the next congressional session.

By Sagoh