Anthropic Says It Does Not Support Blanket Bans on Open-Weights Models
Anthropic's newly stated policy position separates open-weight availability from its concerns about frontier capability, chip access, distillation and pre-release safety testing.
Anthropic said on July 27 that it does not support a blanket ban on open-weights AI models. The statement, published by the company under chief executive Dario Amodei's name, responds to public debate over possible restrictions on models from China and to criticism that Anthropic sought to limit open-weight competition.
What changed
Anthropic's position is that open-weight models without dangerous capabilities can benefit businesses, developers and researchers. It says a ban on their use by U.S. businesses would not address the risks it considers most serious, including highly capable systems controlled outside U.S. jurisdiction and the possibility of cyber or biological misuse.
Instead, the company advocates three policy measures. First, it calls for keeping powerful chips and chipmaking equipment from reaching China and for enforcing against smuggling. Second, it supports action against industrial-scale model distillation. Third, it argues that sufficiently capable models should undergo mandatory safety testing before release, whether they are open or closed.
Anthropic also draws a distinction between a model's availability and its capability. It says open weights can make it harder to apply guardrails, monitor use or withdraw a model once released, while maintaining that those concerns do not justify a category-wide ban. The company says the appropriate response should be capability-based testing rather than a presumption that every open-weight model presents the same risk.
Why it matters
The announcement is a policy position, not a change to Anthropic's product terms or a government rule. Its immediate importance is that it clarifies where the company places the line in a contentious debate: not between open and closed release by itself, but between lower-risk systems and models with potentially dangerous capabilities.
That framing could matter to developers and policy makers because it proposes a common test for both release models. It also leaves significant open questions, including which capabilities trigger testing, who sets the threshold and how testing would apply across jurisdictions. Those questions remain unresolved in Anthropic's statement.
Status
Confirmed. Anthropic published its own policy position through its official newsroom. Internal confidence is high for the fact of the statement and its stated recommendations; the security and geopolitical assessments in it are Anthropic's views, not independently established findings.
Sources
Update note: Last reviewed 2026-07-29. We will revise this post if Anthropic changes this position or a concrete policy proposal establishes a new implementation framework.
Sources
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.