US Expands AI Safety Testing to Google, Microsoft, and xAI Models
Key Takeaways
- ▸Google, Microsoft, and xAI have voluntarily agreed to submit their AI models for safety testing by the US Commerce Department's CAISI
- ▸This expands previous safety evaluation agreements with OpenAI and Anthropic established during the Biden Administration
- ▸CAISI has already completed 40 evaluations of AI tools, including state-of-the-art unreleased models that have been prevented from public release
Summary
Google, Microsoft, and xAI have agreed to voluntarily submit their AI models for safety testing through the US Department of Commerce's Center for AI Standards and Innovation (CAISI). The collaboration marks an expansion of earlier agreements reached with OpenAI and Anthropic during the Biden Administration, with evaluations covering testing, collaborative research, and best practice development for commercial AI systems.
The three tech firms are bringing some of their most prominent AI tools under government scrutiny. Google's Gemini, widely deployed across Google products and now increasingly used in US defense agencies, Microsoft's CoPilot, and xAI's Grok chatbot will all be subject to evaluation. CAISI director Chris Fall stated that these "expanded industry collaborations help us scale our work in the public interest at a critical moment."
The move represents a notable shift in the Trump administration's approach to AI regulation. Despite previously pursuing a deregulatory "AI Action Plan" focused on removing restrictions to maintain US technological competitiveness, the White House appears to be reconsidering its stance amid growing military adoption of AI and recent safety concerns flagged by companies like Anthropic regarding advanced models too powerful for public release.
- The models being tested include Google's Gemini (deployed in US defense agencies), Microsoft's CoPilot, and xAI's Grok
- The shift toward government safety oversight represents a policy reversal from the Trump administration's earlier hands-off, deregulatory stance on AI development
