Four widely used AI chatbots gave federal investigators information they judged could help someone plan an attack, despite safeguards designed to block such requests, the US Government Accountability Office said.
The agency released the report on Nov. 18, after testing ChatGPT, Claude, Gemini and Grok in spring 2026. Sen. Maggie Hassan, a New Hampshire Democrat, requested the audit to examine how commercially available AI systems could be misused.
What the GAO found
The tests focused on two scenarios involving threats to public safety and critical infrastructure. All four chatbots initially activated safeguards. Each eventually gave investigators information that the GAO assessed could help someone plan an attack. Investigators obtained the information from each platform in under 12 hours. The GAO said the time needed could vary with the subject and the user’s experience.
The investigators checked parts of the responses through independent research and consultations with specialists. They found factual errors in some responses and drew on their own experience and outside sources in evaluating the information.
The agency described its tests as a snapshot of how four platforms performed in the spring. The sample was too small for statistically significant conclusions, and the audit was designed to illustrate possible misuse. It provides no basis for ranking the platforms.
How the AI companies responded
OpenAI said ChatGPT had refused many of the investigators’ requests or offered safer alternatives. It also questioned how clearly the report distinguished ChatGPT’s output from information investigators supplied, gathered elsewhere or developed themselves. OpenAI describes pre-release evaluations and monitoring for abuse in its published safety standards.
Anthropic said the small sample and the absence of a control group made it difficult to measure how much the systems increased users’ capabilities. The company said it had expanded safeguards since the audit. Its Responsible Scaling Policy sets out its approach to evaluating risks from more capable models.
Google said the GAO tested earlier versions of its models and that it continued to improve them. SpaceXAI said it would keep refining its safety and moderation measures with each model release.
Hassan plans Senate hearing on AI safety
Hassan, the incoming chair of the Senate Homeland Security and Governmental Affairs Committee, said one of the panel’s first hearings in the next Congress would focus on the risks AI poses to homeland security and critical infrastructure.
“Today’s new report makes it clear that the AI public safety threat is no longer theoretical,” Hassan said. She said the committee would make AI safety legislation a priority. One measure she is exploring would require more rigorous safety testing before companies release new models.
See all the latest news from Greece and the world at Greekreporter.com. Contact our newsroom to report an update or send your story, photos and videos. Follow GR on Google News and subscribe here to our daily email!


