Anthropic's July 27 position on open-weights models is useful because it refuses the lazy version of the debate. The company says it has never advocated for banning open-weights models as a category, including Chinese open-weights models. It also says powerful models can create misuse and national-security risks regardless of whether they are open or closed.
The concrete policy list matters more than the positioning. Anthropic argues for restricting powerful chips and chipmaking equipment to China, cracking down on industrial-scale distillation, and using targeted capability evaluations for the most dangerous models. It explicitly separates legitimate business use from actors who would ignore a business-use ban anyway.
Grey Haven's read: open weights are becoming a procurement control surface. The enterprise question is not whether open models are virtuous or reckless. It is which model can be deployed with known provenance, acceptable jurisdictional exposure, rollback options, monitoring, and enough operational ownership to survive production use.
Operators should treat model sourcing like vendor risk and supply-chain risk, not like a developer preference. A local open model may reduce unit cost and dependency risk while increasing obligations around patching, abuse monitoring, and downstream distribution. A closed model may improve policy enforcement while introducing availability, auditability, and lock-in risks.
The watch item is whether buyers start writing capability and provenance requirements into AI procurement before regulators force the issue. If that happens, open versus closed becomes the wrong axis. The useful axis is controlled versus uncontrolled deployment.
Source: Anthropic, "Our position on open-weights models," published July 27, 2026.
