Microsoft, Google, and xAI have agreed to provide the US government with early access to advanced artificial intelligence models for national security testing. The arrangement will allow federal researchers to evaluate emerging AI systems before public deployment as concerns grow over the cybersecurity and military implications of increasingly capable models.
The agreements were announced by the Center for AI Standards and Innovation, or CAISI, within the US Department of Commerce. The agency said it will use the access to study model capabilities, test for security risks, and assess potential misuse scenarios ranging from cyberattacks to military applications. The move follows a pledge made by the Trump administration in 2025 to establish partnerships with technology companies for AI security evaluations.
Microsoft said it will collaborate with government researchers to test its models for unexpected behaviors and help develop shared datasets and evaluation workflows. The company previously signed a similar arrangement with the United Kingdom’s AI Security Institute. Google declined to comment on the agreement, while xAI did not immediately respond to requests for comment.
The push for earlier government access comes amid rising concern in Washington over the capabilities of frontier AI systems, particularly after Anthropic previewed its Mythos model. Anthropic recently disclosed that Mythos uncovered tens of thousands of software vulnerabilities, raising fears among policymakers and corporate security teams that advanced AI could dramatically accelerate cyberattacks and large-scale hacking operations.
CAISI, formerly known as the US Artificial Intelligence Safety Institute under the Biden administration, has become the government’s primary hub for evaluating advanced AI systems. The agency said it has already completed more than 40 model evaluations, including tests on unreleased systems. In some cases, developers provide versions of models with safety protections partially removed so researchers can more effectively probe for national security risks.
Governments Push For Earlier AI Oversight
The agreements reflect a broader shift toward proactive oversight of advanced AI systems before they are deployed commercially. Policymakers are increasingly concerned that the pace of AI development is outstripping existing regulatory and security frameworks, particularly as models become more capable in coding, reasoning, and autonomous decision-making.
The arrangements also suggest closer coordination between AI developers and national security agencies. As AI becomes strategically important, governments are treating frontier models less like conventional software products and more like critical infrastructure technologies with geopolitical implications.
AI Security Becomes A Strategic Battleground
The agreements build on earlier partnerships established with AI companies during the Biden administration, but they also arrive as competition intensifies among AI developers and cloud providers. The Pentagon recently signed agreements with seven AI companies to deploy advanced systems on classified military networks, signaling growing adoption of AI within defense operations.
Notably, Anthropic was absent from those Pentagon agreements amid reported disagreements over military guardrails for its AI systems. The situation highlights emerging tensions between commercial AI development, government oversight, and defense applications.
As AI capabilities continue to advance, security testing is becoming a core part of the deployment process. Companies are increasingly expected to demonstrate not only model performance, but also resilience against misuse, cyber threats, and unintended behavior before releasing systems at scale.
