Open-Weights Mandate Proposed as Alternative to Frontier AI Regulation

A Silicon Valley technology veteran has issued a public challenge to Anthropic PBC Chief Executive Officer Dario Amodei, urging the artificial intelligence startup to champion a legal mandate requiring open weights for all publicly released AI models. The open letter proposes that forcing laboratories to publish model weights upon release is the single most effective mechanism to naturally slow frontier AI development without triggering regulatory capture.

The proposal follows Anthropic’s release of a policy paper titled "We Must Pace the Frontier," in which the San Francisco-based firm committed to integrating embedded third-party evaluators into its deployment pipeline. The paper urged government authorities to require competing frontier AI developers to match those commitments. Shortly after the document was published, OpenAI Chief Executive Officer Sam Altman publicly expressed agreement with Amodei's call for coordinated pacing across the industry.

In the open letter, first reported by Hacker News, software engineer and startup founder Jacob Gold argued that traditional regulatory frameworks ultimately entrench established tech giants. Gold, whose background includes engineering and founding roles at Google, Docker, Nuro, and Bluesky, asserted that policy ideas like compute thresholds, antitrust waivers, and evaluator mandates are consistently crafted with input from incumbent frontier labs, creating a costly legal infrastructure that insulates dominant players from emerging competitors.

To address safety risks and market concentration simultaneously, the letter advocates for legislation requiring companies to publish the underlying weights of any AI model made commercially or publicly accessible. Internal research runs and unreleased models would remain exempt, allowing labs to retain proprietary control over internal testing while still providing governments access to non-public systems. Under the proposed framework, publicly launched systems such as Anthropic’s hypothetical Fable model or OpenAI’s Project Astra would require immediate open-weights distribution.

The primary mechanism of the proposed policy relies on venture capital economics rather than state administration. Current private market valuations for frontier AI startups rely heavily on the assumption that model weights will remain proprietary, creating software moats that protect future revenue. Forcing public models to open their weights would systematically depress these valuations, significantly reducing the massive capital pools available to finance increasingly expensive training runs across the sector.

The letter emphasizes that Anthropic’s corporate structure as a Public Benefit Corporation (PBC) makes Amodei uniquely positioned to champion such an initiative. Anthropic’s corporate charter legally mandates that the executive team prioritize the responsible development and maintenance of advanced AI for the long-term benefit of humanity, explicitly allowing the leadership team to forgo short-term commercial gains in pursuit of its primary safety mission.

Amodei established Anthropic alongside his sister Daniela Amodei and several former OpenAI researchers in late 2020 after departing OpenAI over safety disagreements. The company has frequently taken non-traditional corporate stances, supporting hardware export controls and reaching a historic $1.5 billion copyright settlement in Bartz v. Anthropic. Proponents of the open-weights policy argue that utilizing Anthropic's corporate structure to alter venture financing incentives offers a far cleaner method for slowing artificial general intelligence development than expansive regulatory bureaucracies.