Open letter to Amodei: force publicly released AI models to ship open weights

Responding to Dario Amodei’s “We Must Pace the Frontier” essay — which pledged Anthropic to embedded third-party evaluators and urged governments to require the same of all frontier labs — a blogger argues that every regulatory lever Amodei has championed inevitably leads to regulatory capture. Compute thresholds, embedded evaluators, and antitrust-waived industry coordination all get drafted with input from the incumbent labs, since no one else grasps the technical detail, and each new incident only adds rules that entrench the largest players who can afford the compliance overhead.

The author’s counter-proposal is a single law: any model a company makes available to the public must be released as open weights. Internal and research models would be exempt, and governments could still access unreleased systems, but public launches — Anthropic’s Fable, OpenAI’s Astra — would have to publish weights on day one. The mechanism for slowing progress is financial rather than regulatory: frontier funding rests on valuations that assume weights stay proprietary, so mandating openness would shrink the capital available for future training runs across all labs simultaneously, with no regulator making case-by-case calls.

The appeal is aimed squarely at Amodei personally, citing his departure from OpenAI, his advocacy for export controls that cut into his own market, Anthropic’s $1.5B author copyright settlement, and the company’s Public Benefit Corporation structure as evidence he is uniquely positioned — and uniquely credible — to call for a rule that leaves nothing to game or dodge. The piece is more a rhetorical challenge than a policy blueprint, but it reframes the open-weights debate as a test of whether “pacing the frontier” means genuine sacrifice or self-serving moat-building.