US to safety test new AI models from Google, Microsoft, xAI (www.bbc.com)

The US Department of Commerce is set to implement safety testing for new AI models developed by Google, Microsoft, and xAI. This initiative, conducted through the Center for AI Standards and Innovation (CAISI), requires these companies to voluntarily submit their AI tools for evaluation prior to public release. This move builds on earlier agreements made with companies like OpenAI and Anthropic during the Biden Administration and aims to ensure that commercial AI systems are secure and beneficial for public use. CAISI director Chris Fall emphasized the importance of these collaborations in scaling work that serves the public interest. This expansion of industry collaboration signifies a substantial shift in US policy towards AI regulation, contrasting with the previous administration's more lenient approach. The evaluations will encompass testing, collaborative research, and developing best practices for AI applications. Notably, prominent tools such as Google's Gemini and Microsoft's CoPilot are under scrutiny, reflecting heightened concern as the military adopts advanced AI technologies. The shift in oversight comes in light of growing apprehensions about AI capabilities and safety, particularly following claims by companies like Anthropic about powerful unreleased models.