Dario's third-party evaluator proposal wins Altman's nod as AI labs spar over regulation
GavinSBaker · x · 2026-09-14
Gavin Baker recaps the weekend's AI regulatory debate, arguing the only tangible new fact is that OpenAI and Anthropic will adopt embedded third-party evaluators (Dario floated METR).
- With no Section 230-style liability shield for model outputs, third-party evaluation helps demonstrate a "duty of care" in future litigation
- Dario's maximalist proposal: embedded evaluators, a national regulatory regime for models above a capability threshold, an international pact among democracies, stricter compute/distillation limits on China, plus an antitrust (Sherman Act) waiver before national regulation
- Sam Altman agreed and said OpenAI would implement embedded evaluators
- Elon Musk endorsed oversight but favors an MPAA-style self-regulatory scheme with 1-2 week competitor safety reviews, and said nothing will slow open-weight models
- Demis called Dario's essay a step in the right direction; Dario is open to a FINRA-like self-regulator
- David Sacks called the antitrust waiver a cartel request and questioned METR's independence, urging unilateral pacing instead
Baker sees minimal investment implications and argues the industry is nowhere near excessive regulation.
More from Companies & People
- Stanford NLP Endorses CS336 as the Starting Point for Learning LLMs Deeply — stanfordnlp · 2026-09-14
- Stanford's new CS 312 'Deep Learning Alchemy' course makes all materials public — stanfordnlp · 2026-09-14
- Sam Altman backs Dario's frontier pacing call, pledges independent evaluators with employee-like access — joshua_saxe · 2026-09-14
- Ex-Anthropic Researcher Jacob Coxon Tells BBC Insiders Are 'Genuinely Frightened' — rohanpaul_ai · 2026-09-14
- Grok Bot went from zero to launch in 7 weeks: ex-Cursor growth lead Roman Ugarte tells the story — lennysan · 2026-09-14
- Polymarket puts 7% odds on US government stake in Anthropic — Polymarket · 2026-09-14