US to safety test new AI models from Google, Microsoft, xAI

The US government has announced a new initiative to safety test artificial intelligence (AI) models developed by major tech companies Google, Microsoft, and xAI. These firms have voluntarily agreed to submit their AI systems for evaluation through the Commerce Department’s Center for AI Standards and Innovation (CAISI). The expanded collaboration aims to assess the capabilities and security of commercial AI tools, including Google’s Gemini, Microsoft’s CoPilot, and xAI’s Grok chatbot. This move reflects growing concerns about the safe deployment of AI technologies, especially as some models are increasingly integrated into sensitive sectors such as defense. CAISI’s evaluations will focus on testing, collaborative research, and the development of best practices for commercial AI systems. The center has already conducted 40 prior assessments, including reviews of advanced models that have not been publicly released, though specific details about these models remain undisclosed. Microsoft emphasized the importance of joint efforts between governments and industry to address national security and public safety risks associated with AI. Google’s DeepMind and SpaceX, which controls xAI, have not publicly commented on the new testing arrangements. This initiative marks a significant shift from the previous US administration’s approach to AI oversight. The Trump administration favored deregulation and rapid technological advancement, signing executive orders aimed at minimizing regulatory barriers for AI development. However, with the increasing military use of AI and concerns over potentially dangerous models like Mythos, the current government appears to be adopting a more cautious stance. Recent interactions between former Trump officials and AI company leaders highlight ongoing tensions around safety and control, underscoring the broader debate about balancing innovation with responsible governance in the AI sector.