OpenAI, Google Sign Open-Weight AI Letter; Anthropic Declines
Anthropic's refusal to sign highlights a split among leading AI developers over whether open-weight models help or harm security, with Anthropic arguing that mandatory safety testing for all high-performance models is needed instead of assuming openness benefits defenders.
Key Facts
- The joint letter "Open Weights and American AI Leadership" was published by NVIDIA CEO Jensen Huang on July 24, 2026.
- OpenAI, Google, and xAI signed the letter, joining initial signatories NVIDIA, Microsoft, and Hugging Face.
- The signatory list expanded to roughly 50 companies and organizations.
- Anthropic did not sign the letter, making it the only major AI developer absent.
- Anthropic CEO Dario Amodei argued that high-performance open-weight models pose risks because safety measures can be removed and released weights cannot be recalled.
Reporting from 2 sources: ASCII.jp, GIGAZINE.
A joint letter opposing rapid regulation of open-weight AI models has drawn signatures from OpenAI, Google, and xAI, among others, expanding the original signatory list to roughly 50 companies and organizations. The letter, titled "Open Weights and American AI Leadership," was first published on July 24 by NVIDIA CEO Jensen Huang in his inaugural post on X. Initial signatories included NVIDIA, Microsoft, and Hugging Face, with additional companies joining afterward. The letter argues that open-weight models, which release trained weights for use in private environments, promote competition and support safety verification and cybersecurity. Huang stated that the world needs both frontier closed models and frontier open models. Notably, Anthropic, the developer of Claude, has not signed the letter, making it the only major AI developer absent from the list. Anthropic CEO Dario Amodei has previously argued that high-performance open-weight models carry risks because safety measures can be removed and released weights cannot be recalled. The letter comes amid U.S. government consideration of restrictions on Chinese open-weight models, which have shown competitive performance against American closed models.
- OpenAI CEO Sam Altman: "I want America to win with both open and closed models."
- Google DeepMind CEO Demis Hassabis: "For the world to benefit from AI, a strong and safe open ecosystem is important."
- NVIDIA CEO Jensen Huang, in his first X post since opening his account in June: "The world needs both frontier closed models and frontier open models."
- Anthropic CEO Dario Amodei, in a company blog post: "Anthropic opposes a blanket ban on open-weight models, and models without dangerous capabilities are a public good."
- Amodei also argued that "all sufficiently high-performance models, whether open or closed, should be subject to mandatory safety testing before release."
- Microsoft's statement, which the letter accompanies, says open models let users "develop and use AI without paying high costs" and avoid being "tied to a single provider."
- Microsoft acknowledged open models carry risks of being "difficult to control or track," but said the answer is not banning them; it is ensuring "the defense side also has access to models of equivalent performance."
- The letter follows a July 22 letter from the Little Tech Association, a group of about 200 companies including Proton and Y Combinator, asking the Trump administration not to restrict Chinese open-weight models.
- GIGAZINE notes the Chinese open model Kimi K3, released July 17, outperformed GPT-5.6 Sol and Claude Fable 5 in some tests and is slated to become open-weight on July 27.
- Signatories include Meta, Mistral, Cohere, IBM, Palantir, CrowdStrike, Dell Technologies, Cloudflare, Mozilla, and The Linux Foundation, alongside the companies named in the summary.
- Anthropic is the only major AI developer absent from the list, according to ASCII.jp.
Synthesized by Yomimono from the 2 cited sources below, including Japanese-language reporting where cited, then editorially reviewed before publishing.