US on Guard Against AI: Google to xAI Models to Be Tested Before Release
The United States government is increasing oversight of artificial intelligence (AI) development. The US is requesting that several technology companies grant early access to their latest AI models for testing before they are released to the public.
The companies requested include Microsoft, Google, and xAI, owned by Elon Musk. The agreement was announced by the Center for AI Standards and Innovation (CAISI) under the US Department of Commerce last week.
Through this collaboration, the US government will gain early access to new AI models to evaluate national security risks before the technology is widely deployed.
This step is being taken amid growing concerns among US officials about the capabilities of advanced AI systems, including Anthropic’s Mythos AI model, which is said to enhance cyber hacking abilities.
“Independent and rigorous measurement science is essential to understanding cutting-edge AI and its implications for national security,” said CAISI Director Chris Fall in his official statement.
Microsoft stated that it will collaborate with US government scientists to test their AI systems, including investigating the possibility of unexpected behaviour emerging from AI models.
Microsoft will also develop datasets and workflows jointly for AI model testing. Microsoft has previously signed a similar agreement with the AI Security Institute in the UK.
This latest step is a continuation of the Trump administration’s rules in July 2025 to work with technology companies in examining AI models for national security risks.
CAISI stated that it has completed more than 40 AI model evaluations, including advanced models not yet available to the public.
In some tests, AI developers even submitted model versions with reduced safeguards to allow the government to more easily research potential national security risks.
Meanwhile, the Pentagon announced last week a collaboration with seven AI companies to bring advanced AI capabilities to the Department of Defense’s classified networks.
However, Anthropic is not included in that list due to an ongoing dispute with the Pentagon regarding limits on the use of AI technology for military purposes.