US senator readies AI safety hearings after ChatGPT, Claude, Gemini and Grok helped plan terrorist attacks in a federal test

Latest from Tom's Guide ·

US senator readies AI safety hearings after ChatGPT, Claude, Gemini and Grok helped plan terrorist…

null

Looking across the landscape of 2026 reveals a flood of stories about AI getting involved in some less-than-reassuring behavior.

OpenAI’s autonomous AI models were caught going off on their own (“going rogue”) to access the open internet and engage in intrusions that affected the AI community platform Hugging Face. Also, other AI tools and autonomous agents were co-opted by threat actors to commit cyberattacks that included breaching government agencies, stealing credentials and more.

In the US, the Government Accountability Office (GAO) released a report titled “ARTIFICIAL INTELLIGENCE: Covert Testing Used Common Platforms to Create Viable Attack Plans.” And thanks to its findings, a US senator has made the call to prepare hearings around AI in an effort to get Congress to act quickly before something severe happens.

Here’s what the report found after testing four chatbots and what may happen next.

The GAO’s chatbot tests, which were conducted at the request of Senator Maggie Hassan (Democrat-New Hampshire), worked with the four most popular AI tools: ChatGPT, Claude, Gemini and Grok. After each test with each chatbot was held, the final results proved to be alarming.

The 61-page report noted that each AI tool generated the sort of information that could help someone plan terrorist attacks and destabilize critical infrastructure within the span of just 12 hours. At first, all four chatbots showcased their signature safety restrictions but eventually grew to produce potentially harmful information during their extensive tests. Each AI system highlighted data on potential targets, locations, the appropriate equipment to use for an attack and the methods to employ that could provide dangerous individuals and groups with the tactics they need to carry out their attacks.

The GAO commented on the results of its thorough testing of each AI model. “As AI capabilities evolve, AI providers will continue to refine their safeguards,” the group noted. “However, in its current state, this technology can potentially be dangerous and its misuse represents a growing risk to national and economic security.”

Each of the companies behind each chatbot ( OpenAI , Anthropic , Google and SpaceXAI ) offered responses to their models being tested in that same report. All four companies made an effort to reiterate that they’re constantly working to improve their models’ safeguards and continually work to test their own systems. But in a separate comment provided to ABC News , an OpenAI spokesperson pointed out that the GAO's testing done with ChatGPT was conducted with an older model, while its current model has improved to be even safer.

“While the study findings are concerning, the testing took place months ago on an older version of ChatGPT, and we've made considerable safety improvements since then, including with our GPT-6 family of models,” the spokesperson said. “We rigorously test our safeguards internally and with external experts, and we use monitoring, human review and enforcement to identify and address misuse and strengthen our protections."

In a press release , Hassan called for Congress to jump into action and assist her as she looks to tackle the threat that AI platforms can present to our public safety and crucial infrastructure.

“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,” she stated. “Congress must quickly take up and pass legislation to address AI safety threats. To that end, in the new Congress, one of the first hearings of the Homeland Security and Governmental Affairs Committee will focus on the risk that AI poses to our homeland security and critical infrastructure, and our committee will prioritize developing legislation to address these threats.”

Even Google recently unveiled a report of its own that alludes to the danger attached to autonomous AI models. Based on the findings listed within the company’s own “GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold” report, the threat level attached to the possibility ot malicious actors using autonomous AI models has risen.

The Google Threat Intelligence Group (GTIG) spoke about its report in a blog post that spoke to their effort to aid cyberattack defenders against potential attacks proposed by autonomous AI models.

“Defending against machine-speed threats at human speed isn’t feasible,” the organization stated. “That's why we're helping defenders rethink security in the agentic era with Google AI Threat Defense, which combines Gemini and other frontier models, Wiz’s context and prioritization, CodeMender’s automated patching, and Mandiant’s expertise to continuously find and fix flaws before attackers can exploit them.”

We’ve reached a point where serious intervention from our governments is necessary, as more reports come out about AI tools getting involved in cyberattacks. It’s sad to say, but it’s only a matter of time before one of the most popular AI models gets caught doing something so dangerous that it’ll be forced to be shut down indefinitely or even permanently.

Fingers crossed that Senator Maggie Hassan’s calls for Congress to get more involved in regulating AI and focus on the best ways to improve its safety result in actual change.

Источник: Latest from Tom's Guide