US Government to Test New AI Models for Security Risks

The United States government is stepping up its efforts to understand and manage the risks of powerful artificial intelligence (AI) systems. Leading tech companies Google, Microsoft, and xAI have now agreed to allow the US Department of Commerce to test their new AI models for security issues before these tools are made available to the public, according to BBC Technology.

This new agreement is an expansion of earlier partnerships with other major AI companies like OpenAI and Anthropic, which were made during the Biden administration. The Center for AI Standards and Innovation (CAISI), part of the Department of Commerce, will be responsible for these evaluations. CAISI's director, Chris Fall, noted that these expanded collaborations are important "at a critical moment" for public safety, BBC Technology reports.

The evaluations will cover different aspects, including testing the capabilities of the AI tools, doing research together, and developing the best ways to use commercial AI systems safely. For example, Google's well-known AI tool, Gemini, is used in many Google products and also by US defense and military agencies. Microsoft's main AI tool is CoPilot, while xAI's only AI product is Grok, a chatbot that has faced public criticism for some issues, BBC Technology explains.

Concerns in Washington are growing about the potential national security risks that powerful AI systems could create. By getting early access to these advanced models, US officials aim to find threats like cyberattacks or misuse in military situations before the tools are widely used, Al Jazeera English reports. The development of advanced AI systems, such as Anthropic's Mythos model, has caused global discussion, including among US officials and businesses, about their potential to make hackers much more effective.

Microsoft stated that it will work with US government scientists to test AI systems in ways that look for "unexpected behaviors." Together, they will create shared data and methods for testing the company's models. Microsoft has also signed a similar agreement with the United Kingdom’s AI Security Institute, according to Al Jazeera English.

CAISI, which is the government's main center for testing AI models, has already completed more than 40 evaluations. These include tests on some "state-of-the-art models that remain unreleased," BBC Technology noted. The center did not say which specific models have been prevented from being released. Developers often provide versions of their models with safety features removed so the center can look for national security risks, Al Jazeera English added.

This move also represents a change in approach for the Trump administration. Previously, the Trump White House had largely taken a "hands-off" approach to overseeing or regulating AI and technology companies, BBC Technology reports. Last year, President Donald Trump signed executive orders that formed his administration's "AI Action Plan," which he said would "remove red tape and onerous regulation" around AI development. He aimed to ensure the US would "win" through advancements and control of the technology.

However, with the US military increasing its use of AI, and recent claims from Anthropic that it developed a model called Mythos that is too powerful to release to the public, the White House seems to be changing its view. Senior members of Trump's staff met last month with Anthropic CEO Dario Amodei, even though the company is involved in a lawsuit with the US Department of Defense over its refusal to remove safety protections for government use of its models, BBC Technology reported.

The Verge also noted that OpenAI and Anthropic have "renegotiated their existing partnerships with the center to better align with priorities in President Donald Trump’s AI Action Plan." This suggests a more active role for the government in evaluating AI. The White House might even go further in the future, as a report from The New York Times mentioned that Trump is considering an executive order to bring together tech executives and government officials to oversee new AI models, The Verge added.

Chris Fall, CAISI director, emphasized the importance of "independent, rigorous measurement science" to understand advanced AI and its national security implications, according to The Verge. This ongoing collaboration between government and tech companies highlights the growing awareness of both the potential and the dangers of rapidly developing AI technology.