Three separate open letters about AI development landed within days of each other in late July and early August 2026, and Simon Willison's roundup of them is worth reading precisely because the letters disagree with each other. A 235-company coalition led by Microsoft wants fewer restrictions on open-weight models. Anthropic wants limits on industrial-scale distillation while opposing outright bans. Over a thousand AI employees want international coordination to slow the pace of automated AI research itself.
This article covers what each letter is actually asking for, why the disagreements between them are more informative than any single letter's position, and what the split suggests about where real regulatory pressure is likely to land.
Letter one: open weights, defended by scale
The Microsoft-led letter, signed by 235 companies including Nvidia and OpenAI, advocates for open-weight models and distillation techniques against potential government restrictions. Open-weight models are AI models whose trained parameters are published for anyone to download and run, as opposed to being locked behind a hosted API. Distillation is the practice of training a smaller model to mimic a larger one's behavior, often using the larger model's outputs as training data — a technique that makes frontier capability more accessible but also lets competitors replicate a lab's expensive training investment more cheaply.
The scale of this coalition — 235 companies, including firms that also compete fiercely with each other — signals a shared commercial interest that cuts across the usual rivalries: nearly every company outside the very top tier of frontier labs benefits from open weights and distillation staying broadly legal, because those techniques are how smaller players access capability they couldn't otherwise afford to train from scratch.
Letter two: Anthropic wants a narrower carve-out
Anthropic's separate letter breaks from that industry consensus in one specific way: it raises concerns about authoritarian AI development and calls for limiting industrial-scale distillation, while still opposing outright bans on open models. That's a meaningfully different position from "no new restrictions" — it's "some restrictions, targeted narrowly at large-scale distillation operations, not at open-weight models generally."
A company that already leads on frontier capability has less to lose from restricting industrial-scale distillation than a competitor trying to catch up by training on that company's outputs.
The self-interest here isn't hidden — Anthropic sits closer to the frontier than most of the 235 signatories on the Microsoft letter, which makes restricting distillation a position that happens to protect the value of Anthropic's own training investment as much as it addresses any stated safety concern. That doesn't make the concern illegitimate, but it's a useful lens for evaluating why the two letters land in different places on the same underlying issue.
Letter three: a warning from inside the labs
The third letter is the hardest to dismiss as pure corporate positioning, because it comes from 1,324 AI company employees — including leaders from both OpenAI and Anthropic — rather than from the companies themselves. Its ask is for international governance to pace automated AI research amid competitive pressures. That's a request from people building the technology, arguing that the competitive dynamics between their own employers are moving faster than anyone currently has the ability to govern responsibly.
This third letter matters because it doesn't fit neatly into either company's official position — employees signing across company lines suggests a level of shared concern that isn't fully captured by either the Microsoft coalition's deregulatory stance or Anthropic's narrower distillation-focused ask.
What this means for builders
For anyone building on frontier AI infrastructure, this split is a signal about where regulatory uncertainty is likely to concentrate first: distillation and open-weight distribution, not model capability itself. If Anthropic's position gains traction, expect narrower rules targeting large-scale distillation operations specifically, rather than blanket restrictions on running or fine-tuning open-weight models — a distinction worth tracking if your product roadmap depends on either.
The employee letter is also worth watching as a leading indicator — cross-company employee coalitions expressing shared concern often precede formal policy proposals, since they represent people with technical visibility into risks that aren't yet public.
Conclusion
Three open letters in one week isn't noise — it's a rare moment where the AI industry's internal disagreements are visible all at once, rather than smoothed into a single lobbying position. The interesting story isn't that the industry wants a voice in its own regulation; it's that the industry can't agree internally on what that regulation should even target, and the shape of that disagreement is a better predictor of future policy than any single company's stated position.
Further Reading
- Anthropic's cryptanalysis research shows the kind of frontier capability work these governance debates are ultimately about