"Open Weights and American AI Leadership" is a joint open letter published July 24, 2026 by a coalition of AI developers, chipmakers, cloud and application companies, investors, and open-source foundations. It argues that open-weight models — models "that anyone can download, inspect, modify, and run on their own infrastructure" — are a necessary part of American AI leadership, and urges policymakers to expand compute access, invest in shared training assets, and avoid "premature restrictions on open models that stifle competition or drive innovation overseas." It was published as the Trump administration was weighing executive action on open-weight AI in response to a wave of Chinese open-weight releases (Open-Weight Frontier Models).
The letter has no single originating entity. The fullest primary copy located is hosted on Microsoft's corporate-responsibility site, which carries both the complete text and a live signatory roster (Source: microsoft.com).
Argument
The letter opens with an analogy to 1980s open-source software, which it describes as having "challenged the prevailing belief that software would advance only if companies kept tight control over their code." It states that open-source software now supports most of the internet and underlies systems used by the world's largest technology companies as well as the U.S. military and federal agencies, and that beyond lowering cost it "created a shared foundation of knowledge on which generations of American engineers and entrepreneurs built their institutional sovereignty."
The United States is described as facing "a similar choice with artificial intelligence." The letter's framing of the criterion for leadership is that it "will be judged not by one frontier AI model, but by whether the United States builds a strong, open ecosystem that diffuses into every sector."
Four affirmative claims follow.
Access to the AI economy. Startups, businesses, universities, and public institutions can build on advanced models "without training one from scratch or paying frontier-model prices for every task." Open weights let organizations "match the right model to the right job at the right cost, reserving frontier-scale capability for genuine frontier problems and running efficient, specialized models everywhere else," which the letter presents as the condition for AI to remain "economically sustainable as its use scales into the billions of everyday tasks."
Competition. Open weights are said to create rivalry "not only among model developers but across cloud chips, applications, and services," which the letter ties to lower costs and broader distribution of gains rather than concentration "in a few hands."
Customer control. Open weights are offered as assurance against provider lock-in, allowing organizations "to control their own data, evaluate and adapt models to their own needs, and deploy them wherever their business requirements demand," and to own accumulated value "through self-improving models, specialized capabilities, and accumulated knowledge."
Safety and security. The letter concedes that "open weights carry real and distinct risks," specifically that "once released, the weights are beyond the original developer's control, and modified versions are difficult to trace or reverse." It argues the response should not be prohibition: in a world where cybersecurity attackers use advanced AI, "defenders need access to models with comparable capabilities so they can detect, simulate, and respond to emerging threats." It goes further, stating that "openness may be one of the most important paths to AI safety and security," on the reasoning that closed models "can be breached, misused, or fail in ways that outsiders cannot detect," and that concentrating capability behind a few closed models "results in a small number of single points of failure, weakens competition, and leaves critical technology in the hands of a few providers." The analogy is again to open-source software, which the letter says "demonstrated that transparency can be more secure than obscurity," and it calls for "protections tied to real and demonstrated harms rather than assuming that closed systems are safer by default."
Policy asks
Two sets of asks are directed at policymakers.
The first is affirmative: expanding compute access for startups and researchers, investing in shared training assets — named as datasets, tools, and evaluation frameworks — and "keeping the frontier plural by avoiding premature restrictions on open models." The letter adds that these measures "must also look at how strong application layers can expand sovereign use of AI across the economy."
The second concerns distillation. The letter states that policymakers "should be careful not to conflate legitimate model-development techniques with misappropriation," describing distillation — "the practice of using one model's outputs to help train or improve another" — as "a widely used technique for model improvement, evaluation, and validation" reflecting "a long tradition of learning from, building upon, and improving existing technologies." It distinguishes this from "unlawful efforts to extract value from closed models," which it says "raise legitimate concerns" that "should be addressed through targeted legal and commercial frameworks rather than sweeping restrictions on techniques that play an important role in AI innovation."
This passage is the direct counter to the distillation-as-appropriation framing advanced the preceding week by Michael Kratsios, who on July 23, 2026 accused Moonshot AI of having "developed a sophisticated internal platform to conduct large scale distillation against U.S. models," and by Anthropic in briefings accusing Chinese developers of appropriating its intellectual property (Source: lawfaremedia.org; wired.com). See Adversarial Distillation.
Signatories
The roster is live and grew continuously after publication, so counts are meaningful only with an observation date. Coverage reported 25 signatories on launch day, then 35 and 50 on July 25 (Source: tomshardware.com; implicator.ai). The letter's own page listed 77 names when observed on July 27, 2026.
Signatories at that observation included frontier and applied AI developers (OpenAI, Google, Meta, Microsoft, Mistral, Cohere, AI21, Sakana AI, Nous Research, Reflection, Arcee AI, Black Forest Labs, Fastino Labs, Prime Intellect, Periodic Labs, Perplexity); hardware and infrastructure firms (NVIDIA, AMD, Dell Technologies, Cisco, IBM, Nebius, Modal, Baseten, Fireworks AI, Telnyx); security companies (CrowdStrike, Palo Alto Networks, Cloudflare); open-source foundations and tooling (The Linux Foundation, Mozilla, Hugging Face, Ollama, LM Studio, LMSYS, Unsloth, LangChain, vLLM-adjacent projects such as GPU MODE, Interconnects AI); platform and enterprise software (GitHub, Replit, Vercel, Box, Block, ServiceNow, Glean, DoorDash, Palantir, Scale, SpaceX); and investors and networks (Andreessen Horowitz, Y Combinator, Unusual Ventures, Emergence Capital, Atreides Management, American Innovators Network).
Anthropic and Amazon were absent at every observation. OpenAI was not among the launch-day signatories and added its name on the evening of July 24; Sundar Pichai signed for Google on July 25 (Source: implicator.ai).
Reception
Publication drew immediate attribution to individual executives. Jensen Huang marked the letter with his first post on X, warning the industry against repeating a mistake he says software narrowly avoided in the 1980s; Satya Nadella also backed the effort (Source: fortune.com).
Coverage placed the letter within an industry split that tracked company size. The New York Times and Wired reported that OpenAI and Anthropic were separately holding private briefings with lawmakers framing Chinese AI as an urgent national-security concern, while smaller startup founders described open-weight competition as a market reality favoring acceleration over restriction (Source: washingtonpost.com; wired.com). The letter followed by two days a separate appeal in which nearly 200 startups and investors, including Y Combinator, urged the administration not to ban Chinese open-weight models (Source: politico.com).
The letter's safety argument runs against the position taken by the UK AI Security Institute, which holds that open release "precludes many of the safety measures that closed model developers can use to detect and disrupt misuse" and creates "a persistent and irreversible risk of misuse" for models with dangerous capabilities, while measuring the open-weight cyber lag at four to seven months (How Far Behind the Frontier are Leading Open Weight Models on Cyber? (UK AISI, July 2026)). It aligns more closely with Nathan Lambert's argument that open weights slightly behind the closed frontier are "our natural buffer to mitigate the risks" (Kimi K3: The open-weights escalation (Nathan Lambert, July 2026)), and with the marginal-risk evaluation method set out in Wallace et al. (2025).
Provenance
The letter is undated in its body beyond the "July 24, 2026" dateline and carries no named author or secretariat. Two signatory-hosted copies are known: the Microsoft corporate-responsibility page transcribed here, and an NVIDIA-hosted PDF at images.nvidia.com/pdf/Open-Weights-and-American-AI-Leadership.pdf, which is the copy Dario Amodei linked when responding to the letter (Our position on open-weights models (Amodei, July 2026)). The Microsoft-hosted copy is the fullest primary version located, being the only one carrying the live signatory roster; its canonical and og:url metadata match the hosting URL, its og:title reads "Open Weights and American AI Leadership," and its article:modified_time was 2026-07-26T22:31:20Z when observed. Hosting by signatories rather than a neutral coalition body is a provenance limitation: no independent copy of record exists, and both known copies belong to companies that signed. The text and launch date were independently corroborated by Tom's Hardware (July 24), The New York Times (July 25), and Forbes (July 25). Because the roster is live, any signatory count drawn from secondary reporting is stale on arrival and should be cited with its observation date.
Relationships
- supports: Open-Weight Frontier Models — the industry case against restriction.
- contradicts: Our position on open-weights models (Amodei, July 2026) — Amodei rejects the safeguard-development and defender-advantage premises while agreeing with much of the rest.
- contradicts: How Far Behind the Frontier are Leading Open Weight Models on Cyber? (UK AISI, July 2026) — on whether irreversibility of release makes open weights net-riskier.
- contradicts: Adversarial Distillation — rejects treating distillation as misappropriation.
- related: Open-Source AI / Open-Weight Models, Dual-Use Frontier AI, Regulatory Capture in AI Policy.
- related: Microsoft, Nvidia & TSMC — AI Compute Infrastructure, Meta AI, OpenAI, Google DeepMind.