US, Google, Microsoft and xAI to review AI models for security

The U.S. Department of Commerce’s Center for AI Standards and Innovation (CAISI) has signed agreements with Google DeepMind, Microsoft and xAI to assess early versions of their advanced AI models before public release. The collaborations are aimed at identifying cybersecurity, biosecurity and chemical-weapons-related risks and improving measurement science around frontier AI.

Under the agreements, CAISI will conduct pre-deployment evaluations and targeted research to probe unexpected behaviours, misuse pathways and potential failure modes in the models. Microsoft said it will also work with U.S. government scientists to test systems in adversarial settings, and separately announced parallel cooperation with the UK’s AI Security Institute to broaden testing approaches.

Markets may react to the news as investors reassess regulatory and compliance costs for major AI vendors, while cybersecurity vendors could see increased demand for tools and services. Short-term volatility is possible if companies delay releases or disclose new safeguards; longer term, transparent evaluation frameworks could reduce systemic risks and support broader adoption of AI in regulated sectors.

The agreements build on earlier CAISI arrangements with OpenAI and Anthropic and come amid heightened scrutiny following concerns about recently released models that may expand hacking or misuse capabilities. CAISI’s mandate under the Commerce Department and NIST is to advance evaluation methodologies and coordinate unclassified assessments of frontier AI that may pose national security risks.

Analysts expect that while pre-release government testing may slow some product rollouts and raise short-term costs, it could bolster market confidence over time if evaluations reduce catastrophic risk. Investors will monitor CAISI’s findings, company remediation plans and any further regulatory guidance as key indicators for sector risk and valuation adjustments in the coming months.