Open Letter to Anthropic: Open Weights as the Key to Slowdown

On September 12, 2026, an open letter addressed to Anthropic CEO Dario Amodei was published. Its title is “An open letter to Dario: if you mean it, open the weights.” Reporting by routelastresort on Hacker News (Best) shows that the letter drew a major response. The thread received 260 points and 90 comments. The trigger was a policy document released by Anthropic on the same day, “We Must Pace the Frontier.” The document called on the government to embed third-party evaluators and to mandate equivalent measures for other companies. That OpenAI CEO Sam Altman expressed support within hours also drew attention. The author of the letter states that it is precisely because he believes this intent to slow down that he calls for a different law.

The law sought by the letter is simple. Any AI model offered to the public must be released with open weights. Models for research or internal use are excluded. Government access to unreleased models is maintained. Even so, Anthropic would publish the weights of Fable on its release date, and OpenAI would bear the same obligation for Astra. No other frontier company would be exempt. The proposal aims to change the flow of development funding without going through regulators’ judgment. By breaking the current structure that depends on valuations premised on weight monopoly, the idea is to slow all labs at the same time.

Any AI model a company offers to the public has to be released as open weights.

The above is the central sentence of the letter. Its distinctive feature is directly linking public offering with weight disclosure. Exceptions are limited to internal research and government use. This simplicity is emphasized in contrast with existing regulatory proposals.

Core Claim Presented by the Open Letter

The letter proceeds on the premise of Amodei’s sincerity. It touches on his history of leaving OpenAI to found a company built on safety. It also notes that he did not change his stance despite criticism for calling for regulation of his own industry. It mentions his support for export controls that shrink his own market. It cites as an example his decision to pay $1.5 billion to settle litigation with authors. It also stresses his choice of the public benefit corporation form, PBC, enshrining humanity’s long-term interests in its charter. On that basis, it presses him that if he truly wants to slow the pace, he should call for a law that entails his own sacrifice. Weight disclosure could damage Anthropic’s own business value. Even so, he is positioned as being able to prioritize the mission. It is a strong claim that no one else is better suited as the proposer.

Embedding third-party evaluators is attracting attention as a measure to ensure transparency. The debate over building evaluation infrastructure overlaps in part with attempts at public testing infrastructure shown in Inspired by GNOME OS Test Center and Apple TestFlight. The idea of increasing verifiability by opening testing environments to the outside is shared. However, the letter regards such measures as not leading directly to a slowdown. Evaluation helps make safety visible. Still, it does not touch the funding source of the development race. That distance is the starting point of the criticism.

Structural Problem of Existing Regulation

Leading to Capture

The letter asserts that all existing regulatory proposals will end in capture. Examples cited are embedded evaluators, compute-based thresholds, and industry coordination with antitrust exemptions. None of them can be drafted without advice from the current frontier labs. That is because the actors who understand the technical details are limited. After enactment, rules are added with each accident. Once created, rules are never reduced. Large companies can adapt with legal and compliance staff. For new entrants, only the barriers grow higher. As a result, incumbents’ positions and revenues are protected. It points to the paradox in which slowdown measures turn into protection measures.

This logic is not distrust of regulation in general. It points to the need to separate slowdown from restriction of competition. Thresholds create disputes over the definition of measurement criteria. Coordination blurs the boundary with collusion. Embedded evaluation also risks becoming a formality depending on how it is operated. The letter holds that the more complex the rules become, the more large organizations benefit. Its position is that a simple obligation is less susceptible to capture. Weight disclosure can minimize regulators’ discretion. The only criterion is whether disclosure occurred. This clarity is presented as an advantage.

Impact of Weight Disclosure on the Funding Cycle

Development funding for frontier models is supported by valuations premised on monopoly of weights. The letter makes this premise explicit. If weights are disclosed, replication and redistribution become easy. The prospect of recovering costs through usage fees weakens. Investors’ expected returns decline. Funding for future training runs thins out. The key point is that it acts on all labs simultaneously. It requires no review or permission for specific companies. It is designed to affect speed solely by changing market conditions.

The evolution of on-device execution environments is not unrelated to trends in dedicated compute infrastructure shown in Snapdragon Reality Elite, New Chip for AR/MR. Disclosed weights could be widely used on such infrastructure. The pattern in which the cost of maintaining an exclusive technology stack is passed on to user companies is illustrated by the infrastructure migration case seen in T-Mobile to Leave VMware Amid Broadcom Lawsuit, Migrating 300,000 Cores. Weight disclosure broadens options for users. At the same time, it changes the revenue structure for developers. This duality lies at the heart of the debate.

Meaning of the Sacrifice Demanded of Amodei

The letter repeatedly expresses trust in his character. It praises his decision to abandon a safe career to found the company. It also touches on his history of enduring criticism that accompanied calls for regulation. It cites the choice of the PBC form as a legal basis. Its understanding is that prioritizing the mission is justified under the charter as well. The current proposal conflicts with short-term profit. It is positioned as precisely the kind of judgment envisioned by the PBC system. If proposed by others, it would not be taken seriously. The argument is that it is meaningful precisely because it comes from him as an insider.

The design excluding unreleased models includes practical considerations. It preserves freedom of research while regulating only public provision. It also maintains the government’s access path. It does not impair national-security visibility. Simultaneous publication on the release date equalizes competitive conditions. Both Fable and Astra would follow the same rule. It functions as a simple rule with few loopholes. This clear-cut approach characterizes the proposal.

Editorial Opinion

On short-term impact: over the next three to six months, embedding third-party evaluation is likely to become the industry-standard debate. The authority of evaluators and the burden of costs will be the focus. The weight-disclosure proposal itself has no prospect of legislation. However, we assess that it will affect disclosure items at the time of fundraising.

On the long-term view: in one to three years, a two-tier structure of open weights and proprietary weights is likely to take hold. Lower-cost inference infrastructure will broaden choices for users. Developers will be forced to redesign how to recover training investment. We assess that changes in revenue structure will shift the axis of competition.

Questions from the editors: it is being asked whose benefit a slowdown serves. User convenience and developer revenue do not coincide. It also needs verification whether a disclosure obligation increases the openness of research. How far a mission-first corporate form remains effective will also become an issue.

References

• “An open letter to Dario: if you mean it, open the weights”, by routelastresort — Hacker News (Best), 2026-09-12T19:15:32.000Z (ARR)
• Source URL: https://jacob.gold/posts/open-letter-to-dario-amodei-about-open-weights/