In recent weeks several open letters and position statements have appeared expressing differing views on so-called "open-weight" models and proposals to slow down frontier, automated AI development. The central question is how to enable innovation while reducing opportunities for misuse and safety risks.
Microsoft-led letter (July 24, 2024)
On July 24 a letter titled "Open Weights and American AI Leadership" was published, coordinated by Microsoft and signed by 235 AI-adjacent companies, including NVIDIA, Amazon, Y Combinator, The Linux Foundation and—later as a signer—OpenAI. The letter argues against banning or unduly restricting open-weight models on safety grounds and warns that doing so could harm U.S. leadership and the broader ecosystem.
The signatories contend that relying exclusively on closed models is not inherently safer: closed systems can be breached, misused, or fail in ways outsiders cannot detect. Concentrating advanced capabilities in a small number of closed models creates single points of failure, weakens competition, and places critical technology in the hands of a few providers. By contrast, open-weight models enable a broad community of researchers and developers to examine model behavior, identify vulnerabilities, develop safeguards, and improve models over time.
A notable element of the letter is its explicit defense of distillation—the practice of using one model’s outputs to help train or improve another. The letter frames distillation as a widely used technique for model improvement, evaluation, and validation, part of a longer tradition of learning from and building on existing technologies.
Anthropic's response (July 27, 2024)
Three days later, Anthropic published its own statement, "Our position on open-weights models." CEO Dario Amodei emphasized the risk that authoritarian governments could build AI models more powerful than those developed in the United States. He also warned that models could be misused to carry out cyberattacks or biological attacks.
Amodei called for a crackdown on industrial-scale distillation operations, arguing that such practices can increase risk; however, Anthropic also stated that it has never advocated for a blanket ban on open-weight models.
"Pacing the Frontier" (July 28, 2024)
On July 28 a separate letter, "Pacing the Frontier," was released and signed by 1,324 employees of frontier AI companies. Signatories include Jakub Pachocki (Chief Scientist, OpenAI), Ilya Sutskever (Safe Superintelligence Inc, formerly OpenAI), Dario Amodei (Anthropic) and Jack Clark (Anthropic). The letter asks the U.S. government to support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.
The authors express concern about intense competitive pressure combined with accelerated AI progress driven by automated AI research. They argue that automation and competitive dynamics can rapidly push frontier capabilities forward, increasing potential risks.
Examples cited in the debate
Participants and commentators have pointed to examples that illustrate how automation and internal tooling can speed development and lower costs:
- It was stated that Anthropic produces 80% of their code with Claude Code.
- OpenAI reportedly had Sol reduce their end-to-end serving costs by 20%.
- Kimi K3 designed a chip to serve a nano model built on its own architecture.
These examples suggest that internal automation tools and hardware innovations can materially reduce costs and accelerate development, which may intensify competition and associated risks.
Why this matters
The recent letters show there is no industry consensus on how to balance public safety and continued innovation. One camp emphasizes transparency and broad scrutiny via open-weight models; another highlights risks from state or malicious misuse and calls for constraints on certain practices like industrial-scale distillation.
The debate is likely to continue, and policymakers and international cooperation will play a key role in shaping regulatory and technical responses to align AI progress with safety concerns.
