# Anthropic Rejects Open-Weight Ban as Nvidia Argues for Open Models > Source: > Published: 2026-07-28 16:15:13+00:00 # Anthropic Rejects Open-Weight Ban as Nvidia Argues for Open Models Anthropic CEO Dario Amodei said on July 27 that the company does not support a blanket ban on open-weight AI models, instead calling for chip controls, action against industrial-scale distillation, and safety tests for sufficiently capable models. Nvidia's July 24 policy paper argues that open weights expand access, competition, and customer control while acknowledging that released weights cannot be withdrawn. Anthropic CEO Dario Amodei said on July 27 that the company has not advocated for a blanket ban on open-weight AI models. His post responded to criticism that Anthropic's absence from an industry letter opposing restrictions reflected support for a ban. Amodei called open-weight models without dangerous capabilities a public good. He argued that banning their use by legitimate U.S. businesses would not address his main national-security concerns and could shield domestic AI companies from competition. ### Anthropic favors controls tied to capability Instead of a category-wide ban, Amodei proposed three measures: tighter restrictions on powerful chips and chipmaking equipment reaching authoritarian governments, action against industrial-scale model distillation, and mandatory safety testing for all sufficiently capable models before release, whether their weights are open or closed. That position does not treat all open models as equally risky. It ties proposed testing to capability while arguing that lower-capability open-weight systems should remain available. Anthropic also says open weights can create additional misuse risk once safeguards can be removed and copies can be distributed beyond the original developer's control. Bloomberg, Computerworld, and SiliconANGLE separately reported the policy split and the criticism surrounding Anthropic's decision not to sign the industry letter. ### Nvidia makes the competition case Nvidia's July 24 paper, "Open Weights and American AI Leadership," argues that downloadable and modifiable model weights widen access to AI. It says startups, enterprises, universities, and public institutions can adapt models without training from scratch or paying frontier-model prices for every task. Nvidia also links open weights to customer control, local deployment, and reduced dependence on a single provider. The company acknowledges that weights are difficult to trace or withdraw after release, but argues that broad access can support evaluation, red teaming, and defensive security work. The two positions overlap in opposing blanket restrictions but diverge on how strongly openness itself contributes to safety. Nvidia presents broad access as a source of defensive capacity and scrutiny. Anthropic says that claim should be tested rather than assumed, particularly for cyber and biological risks. For data and AI teams, no new U.S. restriction was announced in these statements. The immediate operational issue is preparedness: track model provenance and license terms, document deployment controls, preserve evaluation evidence, and plan for policy changes that may apply differently to low-risk local models and frontier-capability systems. ## Key Points - 1Anthropic opposes a blanket open-weight ban and instead supports chip controls, action against industrial-scale distillation, and capability-based safety testing. - 2Nvidia argues that open weights can expand access, competition, local deployment, and customer control while acknowledging irreversible release risks. - 3The policy debate has not yet produced a new U.S. restriction, so practitioners should preserve model provenance, licensing, evaluation, and deployment evidence. ## Scoring Rationale The dispute concerns potential U.S. rules affecting access to self-hostable and modifiable AI models, a consequential issue for model deployment and governance. It is a notable policy-position development involving major AI companies, but no regulation or product release has been announced. ## Sources Primary source and supporting public references used for this report. Practice interview problems based on real data 1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)