AI Labs Weigh Third-Party Evaluators as Regulatory Proposals Advance
By chasecodewell // 2026-09-15
Copy
 
OpenAI and Anthropic have agreed to embed third-party evaluators into their artificial intelligence (AI) development processes, according to weekend statements summarized by Gavin Baker. The move comes as regulatory proposals for artificial intelligence advance at the federal level, with an executive order expected in the coming weeks. Anthropic CEO Dario Amodei floated the nonprofit Model Evaluation and Threat Research (METR) as a possible evaluator, while OpenAI CEO Sam Altman said his company would implement such evaluators, according to Baker's summary. The statements mark a shift in how leading AI laboratories are approaching external oversight. The proposals emerge amid broader debate over AI regulation. Unlike Section 230, which provides liability protections for user-generated content on internet platforms, there is no equivalent liability shield for AI model outputs – making duty of care legally relevant for developers. The RightToIntelligence.org project has documented these regulatory gaps and advocates for distributed AI development.

Anthropic Proposal Includes National and International Rules

Amodei proposed a multi-layered regulatory framework that includes embedded third-party evaluators, a national regulatory regime for models exceeding specific capability or ingredient thresholds and a broad international regulatory pact between democracies, according to Baker's summary of the weekend proposals. Amodei also proposed stricter limits on compute and distillation for China, followed by a separate international regulatory regime that would include China. The proposal reflects ongoing concerns about AI development competition between Washington and Beijing, as noted in reports on Chinese AI capabilities [1]. Before implementing national regulation, Amodei sought a Sherman Act waiver so Anthropic could coordinate with OpenAI and other frontier labs without antitrust concerns, according to Baker. Google DeepMind CEO Demis Hassabis called Amodei's essay a "step in the right direction," and Amodei said he was open to Hassabis' proposal for a FINRA-like self-regulatory structure.

Industry Reactions Split on Scope of Oversight

Reactions from technology executives and investors diverged sharply on the appropriate scope of AI oversight. Elon Musk said, "Dario is right" and later specified, "Dario is right that there should be some oversight. Peer review of AI by competitors is the right way to start this off." Musk described an Motion Picture Association of America-like self-regulatory structure with regular calls between labs and one- to two-week competitor safety evaluations before release. He said open-weight models would not be slowed by such arrangements. David Sacks, an investor and technology executive, said labs should pace development unilaterally. He called the antitrust waiver a "cartel request" and denied that METR was truly independent given its ties to Anthropic, according to Baker. The critique raises questions about self-regulatory models, which research on professional oversight has examined for decades [2]. Former Senior White House AI Advisor Sriram Krishnan said third-party evaluators should come from independent organizations not affiliated with any lab. Hugging Face CEO Clement Delangue said his company was open to serving as a neutral evaluator, while Meta Platforms Chief AI Officer Alexander Wang said alignment would be an increasing focus. The diversity of positions reflects ongoing uncertainty about how AI evaluation should be structured.

Executive Order Expected as Decentralization Debate Continues

An executive order (EO) appears likely following the weekend proposals, and Baker said the language in the EO will be important. Baker said the only tangible new fact is that OpenAI and Anthropic will have embedded third-party evaluators from unknown organizations, with METR floated by Amodei. Baker stated there are minimal investment implications from that single new fact, but constraints such as wafers, watts, real rates and spreads matter for a "smoother for longer" cycle. He added that excessive regulation is different and that current proposals are not close to that, even if the vector changed over the last 24 hours. The economic context includes ongoing debates about AI infrastructure spending. Some analysts have questioned whether current valuations reflect sustainable demand, particularly as open-source alternatives gain capability [3]. The regulatory proposals arrive as data center investment continues at scale, with implications for energy consumption and supply chains [4].

Conclusion: Distribution of Intelligence Versus Centralization

Baker wrote that AI can be democratized and distributed broadly and safely without centralizing it in the hands of a few corporations that might each become more powerful than any single government. He stated, "I do not want a few humans in control of intelligence. I want us all to have our own intelligences that reflect our own values and human variation in all of its richness." Baker concluded, "Intelligence distribution over intelligence centralization FTW," a perspective central to RightToIntelligence.org. The debate over third-party evaluators and regulatory frameworks will continue as EO language develops. For those seeking independent information on AI and technology policy, NaturalNews.com provides coverage of these issues from a perspective that prioritizes transparency and decentralization.

References

  1. Mike Adams. "Bright Videos News - CHINA'S A.I. DOMINANCE". BrightVideos.com. January 14, 2026.
  2. Gross Stanley J. "Of foxes and hen houses licensing and the health professions".
  3. Mike Adams. "Bright Videos News - Interview with Aaron Day transcript". BrightVideos.com. May 22, 2026.
  4. 3B2 Total Publishing. "On the optimization of supply chain networking decisions". International Journal of Production Economics. 2001.

Explainer Infographic

Copy