US Government Secures Early Access to Frontier AI Models via New Security Pact

The Centre for AI Standards and Innovation (CAISI) at the US Department of Commerce has announced an agreement granting Microsoft, Google and xAI access to their new artificial intelligence models for national security testing. The arrangement, revealed on Tuesday, permits US officials to evaluate these frontier systems prior to their public deployment to assess potential security vulnerabilities. This move fulfils a pledge made by President Donald Trump's administration in July to partner with technology firms to vet AI models for national security threats.

Under the terms of the deal, Microsoft has committed to working with US government scientists to probe unexpected behaviours within their systems. The two entities will develop shared data sets and workflows specifically designed for testing the company's models. This approach mirrors a similar agreement Microsoft signed with the United Kingdom's AI Security Institute, highlighting a coordinated international effort to manage the risks associated with advanced artificial intelligence.

CAISI Director Chris Fall stated that independent, rigorous measurement science is essential for understanding the national security implications of frontier AI. By securing early access to these models, US officials aim to identify threats ranging from cyberattacks to military misuse before the tools are widely available. The agency noted that developers frequently provide versions of their models with safety guardrails removed to allow the centre to probe for specific national security risks.

The agreement builds on previous collaborations established under the Biden administration, when CAISI was known as the US Artificial Intelligence Safety Institute. During that period, the institute focused on developing AI tests and voluntary safety standards under the leadership of Elizabeth Kelly, who has since joined Anthropic. CAISI reports that it has already completed more than 40 evaluations, including assessments of cutting-edge models that were not yet available to the public.

The announcement follows a separate, concurrent pact between the Pentagon and seven major technology firms to utilise AI systems in classified computer networks. This Department of Defense agreement aims to augment warfighter decision-making in complex operational environments. Notably absent from this Pentagon list is Anthropic, following a public dispute and legal fight with the Trump administration regarding the ethics and safety of AI usage in warfare.

Market reactions to the news varied across the sector. Microsoft shares fell by 0.6 per cent in midday trading on Wall Street following the announcement. Conversely, Alphabet, the parent company of Google, saw its shares rise by 1.3 per cent. xAI, which is not publicly traded, declined to respond immediately to requests for comment regarding the arrangement.