Sometimes a missing signature says more than a given one. In his newsletter, Simon Willison summarized the open letters of the past few weeks — and a clear fault line in the industry comes into view, running right through the question of how open AI models should be allowed to be.
The Microsoft letter
At the center sits “Open Weights and American AI Leadership,” led by Microsoft, dated July 24th, and signed by 235 AI-adjacent companies — among them NVIDIA, Amazon, Y Combinator, the Linux Foundation, and, as a later signer, OpenAI.
The letter is clearly a counterweight to any instinct by the US government to limit open-weights models over “safety” concerns. Its core claim: relying solely on closed models isn’t inherently safer — they too can be breached or misused, and concentrating them among a few providers creates single points of failure. One detail stands out: the letter explicitly comes out in support of distillation, the practice of training one model on another’s outputs — calling it a legitimate, long-standing technique.
Anthropic is absent — on purpose
Conspicuously missing from the signatures: Anthropic. Instead of co-signing, the company published its own position three days later. In it, Dario Amodei stresses the risk of authoritarian governments building models more powerful than those made in the US, and of models being misused for cyber or biological attacks. He calls for a crackdown on “industrial-scale distillation operations” — while also noting that Anthropic has never advocated for a ban on open-weights models.
So it’s precisely the distillation question where two camps collide: for one side it’s normal craft, for Anthropic it’s a doorway through which frontier capabilities can quietly leak out.
And then there’s “Pacing the Frontier”
Shortly before, on July 28th, a second letter appeared: “Pacing the Frontier,” signed by 1,324 employees of leading AI labs — including names like Jakub Pachocki (OpenAI), Ilya Sutskever, Dario Amodei and Jack Clark (both Anthropic). Their request to the US government: to support an international effort to deliberately pace the frontier of automated AI development. The concern is real — Anthropic says it now writes 80% of its own code with Claude Code.
My take
It’s fascinating how differently the same companies argue depending on the letter. On the pace of AI research, Anthropic pulls in the same direction as OpenAI. On open weights and distillation, it deliberately stands alone. That’s not a contradiction — it’s a stance: more openness where it strengthens control and competition, more caution where frontier capabilities could be copied unchecked. Whether that line is wise remains to be seen. Consistent, it certainly is.
Sources: