Morning Edition · Tuesday, July 28, 2026Published at 1:31 AM EDT · New York
Anthropic Says It Never Sought to Ban Open Weights, but Backs Distillation Crackdowns and Pre-Release Testing
Chief executive Dario Amodei's clarification comes as Nvidia, Microsoft, Meta, OpenAI, and Google publicly support open-weight releases and Chinese open models advance.

Anthropic published a formal position on open-weight models after days of speculation that it wanted them restricted. In the statement, the company says it has never advocated banning open weights as a category and supports releasing models without dangerous capabilities as a public good. Chief executive Dario Amodei reinforced that in interviews, saying he does not oppose open weights but is concerned about the strategic gains of Chinese open models.
The disagreements will center on the details. Anthropic favors evaluating each dimension of openness separately and applying more scrutiny as capability rises. Amodei told CNBC and Axios he supports chip export controls, a crackdown on distillation, and mandatory safety testing for all sufficiently capable models, whether open or closed. He singled out biology, arguing that whether a model can help weaponize pandemic-class pathogens is an empirical question for pre-release testing rather than an assumption. The company is not releasing an open-weight version of Claude.
The timing is not incidental. The clarification arrives the same week that Kimi K3 topped open-weight rankings and Washington threatened Moonshot, and while a broad group of United States firms publicly backed open models. Anthropic, the lab most associated with a safety-gated closed strategy, is defining a middle position that keeps its own weights closed while it argues over which capabilities warrant control.
What this means
Anthropic is building the policy vocabulary that would let regulators restrict open weights by capability rather than ban them outright, which favors closed labs whose testing regimes fit that framing and pressures open-weight publishers of frontier models with biology capabilities. The channel is policy. A mandatory pre-release testing regime raises the fixed cost of shipping the most capable open models and shifts leverage toward labs already staffed for adversarial safety testing (red-teaming).
What to watch
- Whether a "crackdown on distillation" becomes concrete United States rulemaking, which would directly affect how Chinese and open labs are allowed to train.
- Whether any United States lab diverges from its peers and releases genuinely frontier open weights, which would undercut the argument that capability-based restrictions are necessary.
Observations to monitor, not financial advice.
Source: Anthropic
Part of a tracked trend
Frontier AI Labs Build Dedicated Policy and Government-Affairs Shops
Frontier AI labs increasingly stand up dedicated policy/strategy divisions and recruit former government officials, building in-house apparatus to shape AI regulation and government relations as the technology becomes a policy battleground.
More from this edition
- Moonshot AI Releases Kimi K3, a 2.8-Trillion-Parameter Open-Weight Model That Tops Open Rankings
- China Accuses Washington of 'AI Hegemonism' and Threatens Countermeasures Over Moonshot Probe
- Anthropic Ships Claude Opus 5, Its Fourth Model in Two Months, With a Doubling on Software-Engineering Evals
- Microsoft Says United Kingdom Grid Connections Take Eight Years, Choking AI Data-Center Buildout
- Nvidia Puts Its Vera CPU to Work Designing the Next Generation of CPUs and GPUs
- Hassabis Says DeepMind Sold to Google Because Independence Would Have Cost Billions It Could Not Raise
- A 184-Million-Parameter Classifier Claims State-of-the-Art Prompt-Injection Detection at a Fraction of Llama Guard's Size
- CORVUS Attacks the Coding Agent's Real Bottleneck: an Append-Only Trajectory That Bloats Context
- FlowEvo Lets Agents Rewrite Their Own Workflows and Skills Instead of Rebuilding Them Each Run
- Meta Puts Segment Anything and DINO Into Assistive Robotics at the University of Pittsburgh
- Vals Launches a Tool That Turns Your Own Codebase Into a Custom Model Benchmark