US Government Gains Access to AI Models from Microsoft and Google

US Government Secures Access to Tech Giants’ AI Models for National Security Testing

The US Department of Commerce’s Center for AI Standards and Innovation (CAISI) has announced an agreement with tech giants Microsoft, Google, and xAI to provide access to their artificial intelligence models for national security testing. The agreement comes days after the Pentagon signed a similar pact with seven major tech companies, including Amazon Web Services, Nvidia, OpenAI, Reflection, and SpaceX.

Under the terms of the agreement, the US government will be allowed to evaluate the AI models before deployment and conduct research to assess their capabilities and security risks. Microsoft has stated that it will work with US government scientists to test its AI systems “in ways that probe unexpected behaviors”, developing shared data sets and workflows for testing its models.

The CAISI Director, Chris Fall, emphasized the importance of independent measurement science in understanding frontier AI and its national security implications. The move builds on previous agreements made by the administration under President Joe Biden’s leadership, which focused on developing AI tests, definitions, and voluntary safety standards.

CAISI has already completed over 40 evaluations of AI models, including cutting-edge ones not yet available to the public. Developers typically provide versions of their models with safety guardrails stripped back for CAISI to probe for national security risks. The announcements follow a broader agreement between the Pentagon and seven major tech companies to use their AI systems across classified computer networks.

The agreements have sparked reactions on Wall Street, with Microsoft’s stock down 0.6 percent and Alphabet (Google’s parent company) up 1.3 percent in midday trading. xAI is not publicly traded, and its response to the announcement has not been immediately available.

Source: Al Jazeera