Google, Microsoft, xAI, OpenAI, and Anthropic agree to pre-release AI model testing with US government

Major AI labs including Google, Microsoft, xAI, OpenAI, and Anthropic have agreed to allow the US government to test their frontier AI models before public release under revised security testing agreements with CAISI (Consortium for Advanced AI Safety & Innovation). The framework lets Washington assess national security risks and evaluate alignment, interpretability, and red-teaming protocols in controlled settings.