Updated
Updated · The New York Times · Aug 4
White House Limits AI Security Reviews to Closed Models as Open-Source Systems Escape New Rules
Updated
Updated · The New York Times · Aug 4

White House Limits AI Security Reviews to Closed Models as Open-Source Systems Escape New Rules

3 articles · Updated · The New York Times · Aug 4

Summary

  • White House officials told major AI companies on Tuesday that the government will review only “closed” AI models for security risks, excluding open-source systems from the initial framework.
  • Closed models from companies such as OpenAI and Anthropic are set to face the new scrutiny, while open-source models from Meta, Google, Nvidia and Chinese firms remain outside the review for now.
  • Representatives from OpenAI, Anthropic, Microsoft, Meta, Google and Nvidia attended the meeting, where officials sought feedback but said the review rules were nearly complete.
  • The approach formalizes Trump administration oversight of leading AI labs, but postpones regulation of fast-improving open-source models and could aid rivals using open-source software to challenge the current leaders.

Insights

If federal security reviews only target closed AI systems, are we leaving the backdoor wide open for advanced foreign open-source models?
How can a security framework prevent catastrophic cyber risks if it completely ignores the rapidly evolving open-weight models flooding the market?