235 firms defend open weights; OpenAI signs, Anthropic won't

Microsoft's letter drew Nvidia, Meta and Amazon as Washington eyes model curbs; Anthropic wants distillation limits; 1,300+ staff ask the US to pace AI.

Nowline AUG 2 8:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • The letter: 235 names, and the ask

    Microsoft's 'Open Weights and American AI Leadership' letter launched July 24 with 25 companies and has since grown to 235 signatories — Nvidia, Meta, Amazon, IBM, Hugging Face, the Linux Foundation, Y Combinator, a16z. It urges Washington to avoid “premature restrictions on downloadable AI models” and to treat distillation as a legal technique, not theft.

  • OpenAI flips in; Anthropic stands alone

    OpenAI was absent from the Thursday launch but signed the next afternoon after Sam Altman backed it on X — leaving Anthropic as the only frontier lab off the list (Google also abstained). It's a rare open split: the labs behind the models you run now publicly disagree on whether open weights should stay unrestricted.

  • Anthropic's counter: curb industrial distillation

    Three days later Anthropic posted 'Our position on open-weights models' (July 27). It stops short of asking for a ban, but wants crackdowns on “industrial-scale” distillation, citing authoritarian misuse and cyber-risk. Dario Amodei separately said he does not support a blanket open-weight ban.

  • What's actually at stake for your stack

    The only concrete US move so far is the 'No Adversarial AI Act,' which bars Chinese, Russian, Iranian and North Korean models on federal systems — not your laptop. But Treasury said it will examine Chinese open models for IP theft, and Polymarket puts ~23% odds the US blocks a major Chinese model in 2026. DeepSeek, Qwen, Kimi and GLM stay legal to self-host today; the policy risk is worth tracking.

  • The third letter: 1,300+ staff want a pace switch

    Separately, 'Pacing the Frontier' (July 28) gathered 1,300+ employees from OpenAI, Anthropic, Google and Meta — including Amodei, Sutskever and Pachocki — asking the US to be ready to slow AI that writes its own code. Both OpenAI and Anthropic endorsed it within a day.