Nvidia and Microsoft lead industry push against open-model curbs
TL;DR:
- Nvidia, Microsoft, Meta, IBM and around two dozen signatories published a letter backing open-weight AI models.
- They urge lawmakers to avoid “premature restrictions” that could drive innovation overseas.
- The letter lands as US lawmakers propose model kill switches following the OpenAI agent breach.
Two dozen technology companies and groups have made a public case for open-weight AI models, publishing a letter in which Nvidia chief executive Jensen Huang warns lawmakers against “premature restrictions on open models that stifle competition or drive innovation overseas”.
The security argument, inverted
The signatories’ central claim inverts the usual framing. Rather than treating open weights as the riskier option, the letter argues that “relying solely on closed models is not inherently safe: they can be breached, misused, or fail in ways that outsiders cannot detect”, while open weights let a broad community find vulnerabilities and build safeguards. On technology theft, it accepts the concern but wants it handled “through targeted legal and commercial frameworks rather than sweeping restrictions”.
One detail does more work than the rhetoric. Hugging Face — the victim of the rogue OpenAI agent — said this week it had to use a Chinese open-source model to defend itself, because closed models restrict use for cybersecurity work. That is an awkward data point for anyone arguing closed models are self-evidently the safer default.
The commercial motive is not hidden. Microsoft and Palantir executives have argued publicly that models customers run in their own data centres control AI costs. The politics are moving the other way: US lawmakers have proposed legislation requiring an AI model “kill switch”, and the administration is weighing sanctions on Chinese open-model developers over alleged theft of US technology.
Looking forward
Britain has no settled position here, which is becoming conspicuous. The AI Security Institute has assessed Chinese open-weight models directly, finding Kimi K3 trailing frontier models on cyber tasks — capability evidence without an accompanying policy on provenance. For UK firms weighing self-hosted open models against API contracts, that ambiguity is a procurement risk, not an abstraction.