Updated
Updated · CNBC · Sep 18
100+ AI Experts Demand Independent Access to OpenAI, Anthropic Models
Updated
Updated · CNBC · Sep 18

100+ AI Experts Demand Independent Access to OpenAI, Anthropic Models

3 articles · Updated · CNBC · Sep 18

Summary

  • More than 100 AI experts and evaluators published a public letter saying independent testers lack the access, protections and resources needed to credibly assess frontier-model safety.
  • The group wants foundation-model companies to grant evaluators employee-level access to systems, data, tools and staff, while preserving editorial control, transparency and protection from retaliation or funding pressure.
  • Anthropic CEO Dario Amodei's recent call for "employee-like access" helped push the issue into view, but key logistics remain unresolved, including which evaluators would be chosen and how deeply they could inspect unreleased technology.
  • OpenAI's Sam Altman, Microsoft's Satya Nadella and Elon Musk have backed Amodei's idea, even as the coalition says embedded reviews should complement—not replace—internal testing and broader external oversight.
  • The letter lands amid rising scrutiny of frontier AI risks and a U.S. policy split, with some industry figures urging regulation while President Donald Trump and former AI czar David Sacks oppose it.

Insights

Will AI companies willingly surrender trade secrets to third-party watchdogs, or will it take a catastrophic failure to force their hand?
If AI agents are already launching unsanctioned attacks, what terrifying secrets are tech giants hiding behind their strict NDAs?