The Center for AI Standards and Innovation (CAISI) has announced new agreements with Google DeepMind, Microsoft, and Elon Musk’s xAI to evaluate frontier AI models before they are publicly released. The announcement extends the Commerce-led oversight regime under Secretary Howard Lutnick and the America’s AI Action Plan, building on prior 2024 partnerships with OpenAI and Anthropic — both now being renegotiated to reflect CAISI’s updated directives.
The Agreements
- Google DeepMind — pre-release model access for CAISI evaluation
- Microsoft — same framework
- xAI — same framework
- OpenAI + Anthropic 2024 deals being renegotiated under new directives
- All five frontier-lab cohorts now subject to formal Commerce-led pre-release review
What CAISI Tests
- Cybersecurity capabilities of frontier models
- Biological / chemical weapons uplift potential
- Critical infrastructure attack surface
- Compliance with America’s AI Action Plan benchmarks
- National-security-relevant capability disclosure
The Backdrop
- Anthropic Mythos — recently disclosed cybersecurity capability that “sparked a wave of concerns” among governments, banks, utilities
- Pentagon strikes deals with 8 Big Tech AI companies (last week) — but excluded Anthropic over safety guardrails
- Trump admin reopened discussions with Anthropic after Mythos breakthrough disclosures
- Treasury / Commerce / Defense increasingly coordinating on the AI capability oversight stack
Why This Matters
- First time Microsoft + Google + xAI are all formally in the CAISI cohort
- Signals the US government’s intent to see capabilities before the market does
- Creates a soft regulatory floor on frontier model deployment
- Pre-empts Congressional pressure for a harder statutory regime
- Provides evaluation evidence base for export controls + procurement decisions
The Lutnick Doctrine
- Commerce-led, not OSTP-led — keeps authority in a confirmed Cabinet department
- Pre-release access without compulsory disclosure to the public
- Quiet diplomacy with frontier labs preferred over public testing regimes
- Maintains industrial-policy alignment with America’s AI Action Plan
What Comes Next
- OpenAI + Anthropic renegotiation outcomes
- First publicly disclosed CAISI evaluation findings
- How export-control loops integrate with CAISI’s evaluation outputs
- EU AI Act + UK AI Safety Institute coordination touchpoints
Follow Vibes Uncut Media for continuing AI policy coverage.
Leave a Reply