An open letter to Dario: if you mean it, open the weights

Software developer Jacob Gold published an open letter to Anthropic chief executive Dario Amodei on 12 September 2026, arguing that if Amodei genuinely wants to slow the pace of frontier AI development he should back a different law: one that would require any model a company offers to the public to be released as open weights.

What prompted the letter

Amodei had published an essay, “We Must Pace the Frontier”, the same day, committing Anthropic to embedded third-party evaluators and asking governments to require every other frontier company to match; Sam Altman agreed publicly within hours and OpenAI matched the evaluator commitment, according to the Politico coverage cited in the letter. The stated goal of that approach is to slow AI progress.

The proposal

Gold’s counter-proposal is narrow: it would cover models made available to the public, not every model a company trains. Internal and research models would be unaffected, and governments could still obtain access to unreleased systems. Anthropic would have to publish the weights of its next model, Fable, on release day, and OpenAI would have to do the same with Astra. The mechanism is financial rather than regulatory: frontier training depends on valuations that assume the weights stay proprietary, and removing that assumption reduces the money available for future training runs, slowing every lab at once without a regulator deciding anything.

Why he doubts the regulatory route

Gold argues that the rules Amodei has asked for — embedded evaluators, compute thresholds, industry coordination with antitrust waivers — would end in capture, because the labs that understand the technical details well enough to help write them are the labs being regulated. Once such rules are law they tend to accumulate, each incident adding a requirement that is never removed, while compliance teams and lawyers favour the largest companies and make it harder for anyone else to catch up.

An appeal to the mission

The letter lists sacrifices Amodei has already made: leaving OpenAI at the end of 2020 to found a company with safety built in from the start, years of asking for regulation of his own industry, support for export controls that shrink his own market, and a $1.5 billion settlement with authors instead of a fight to the end. It points to Anthropic’s status as a public benefit corporation, whose stated mission is the responsible development and maintenance of advanced AI for the long-term benefit of humanity, and concludes that Amodei is the only person who could propose such a law and be taken seriously — so if he means it, he should.