US AI Safety Institute wins pre-release access to Google, Microsoft, xAI models
The US Center for AI Standards and Innovation (CAISI), the successor body to the US AI Safety Institute, announced on 5 May 2026 that it had signed agreements with Google DeepMind, Microsoft, and Elon Musk’s xAI to evaluate their frontier AI models before public release.
The deals expand a framework that already covered OpenAI and Anthropic under voluntary commitments dating back to 2024. With the three new signatories, every US frontier-lab now provides CAISI with some form of pre-deployment access for safety testing — covering capability evaluations, dangerous-content red-teaming, and biosecurity and cyber-risk assessments.
The announcement comes as the Trump administration leans further into AI oversight via voluntary partnerships rather than formal regulation. Critics note CAISI’s evaluations remain pre-deployment and non-binding; supporters argue that pre-release access is the single most useful lever for catching dangerous capabilities before they are shipped to millions of users.
Sources: CNBC · Claims Journal
