Anthropic CEO Dario Amodei rejects ban on open-weight AI models after Nvidia, Microsoft, Meta, OpenAI, and Google back open-weight AI

Just three days after Nvidia, Microsoft, Meta, OpenAI, Google, and more than 20 technology companies urged President Donald Trump to support open-weight AI and warned against “premature restrictions” on open-weight models, all eyes were on Anthropic.

On Monday, Anthropic co-founder and CEO Dario Amodei finally broke his silence.

In a lengthy statement, Amodei rejected claims that Anthropic wants to ban open-weight AI models, calling them a “public good” when they do not possess dangerous capabilities. At the same time, he doubled down on concerns that the most capable frontier AI systems could create national security risks if they fall into the wrong hands.

Anthropic rejects blanket bans on open-weight AI

The response arrives after days of criticism from developers, researchers, and open-source advocates who accused Anthropic of pushing policies that could favor closed AI companies at the expense of open innovation. Amodei’s message attempts to draw a clearer distinction. According to him, the company’s concern has never been open-weight AI itself. The concern is what happens when frontier models reach capability levels that could be exploited for military, cyber, or biological attacks.

“Anthropic has never advocated for a ban on open-weights models,” Amodei wrote in statement published on Monday, July 27, 2026.

That sentence may become the most quoted line from the entire post.

Anthropic tries to reset the conversation

The debate surrounding open-weight AI has become increasingly polarized.

Supporters argue that openly available model weights accelerate research, strengthen competition, reduce dependence on a handful of large AI companies, and give businesses greater control over how they deploy artificial intelligence. That argument gained fresh momentum last week after a coalition of leading technology companies urged the White House to reject broad restrictions on open-weight AI, warning that premature regulation could weaken American leadership.

Anthropic has often found itself cast as the company on the opposite side of that debate.

Amodei’s latest statement appears intended to change that perception.

Rather than attacking open-weight models, he repeatedly argues that many of them create enormous value for developers, researchers, startups, and businesses. His concern begins only when AI systems reach capability levels that present credible national security risks.

That distinction runs through the entire statement.

Three policy priorities replace one sweeping restriction

Instead of calling for blanket restrictions on open-weight AI, Amodei outlined three policy priorities that he believes would better address the risks posed by advanced artificial intelligence.

The first focuses on limiting China’s access to advanced AI chips and chipmaking equipment. Amodei argues that computing infrastructure remains the biggest constraint on training frontier AI models and believes restricting access to those chips is the most effective way to slow military applications of advanced AI.

“We should not sell powerful chips or chipmaking equipment to China, and we should crack down on the rampant smuggling3 and workarounds used to obtain access to such chips. China has limited domestic production capacity, and therefore, due to the scaling laws, cannot build more powerful models than the US without US chips. T.”

His second proposal centers on industrial-scale model distillation.

Distillation has become one of the most closely watched topics in artificial intelligence. The technique allows developers to transfer knowledge from larger frontier models into smaller and more efficient systems. Amodei argues that large-scale distillation backed by authoritarian governments could shorten the gap separating Chinese AI models from their American counterparts, even without matching U.S. computing resources.

“We should crack down on industrial-scale distillation operations.Distillation is a much more compute-efficient process than training models from scratch. It allows China to build much better models than its number of chips would ordinarily enable, and thus partially evade chip bans. Distillation does not allow the CCP to obtain equivalent or superior AI capabilities to the US, but it can bring the Chinese frontier to within a few months of the US frontier.”

His third proposal calls for mandatory safety testing of sufficiently capable AI models before public release, regardless of whether they are distributed with open weights or remain proprietary.

“All sufficiently capable models, open and closed, should go through mandatory safety testing. The best way to address threat #2 is to just directly test models for cyber, biological, and alignment risks before release. I think this idea is actually close to a consensus: I have been heartened both that the Trump administration has moved in this direction in recent months, and by recent industry proposals that would apply such testing to the most capable models regardless of their country of origin or whether they are open or closed (while exempting less capable models, such as those from startups and academia, entirely).”

That point matters.

Amodei argues that safety requirements should depend on what a model can do, not how it is licensed or distributed.

The biggest disagreement has not changed

The statement clarifies Anthropic’s position, yet it leaves the industry’s biggest disagreement untouched.

Open-weight supporters believe broad access makes AI safer over time. More researchers can inspect models, identify vulnerabilities, improve defenses, and build new applications without depending on a small number of companies.

Amodei questions that assumption for the most capable frontier systems.

He argues that future AI models may create an imbalance where offensive capabilities advance faster than defensive ones. He points to biology as one example, suggesting sufficiently capable AI could dramatically lower the barriers to creating dangerous biological agents, whereas developing vaccines and other countermeasures still takes years.

That is not a new concern for Anthropic. It has appeared repeatedly in Amodei’s essays and congressional testimony over the past several years. Sunday’s statement reinforces that those concerns continue to shape the company’s public policy agenda.

The unanswered question

One issue remains unresolved.

Who decides when an AI model becomes “sufficiently capable?”

That question sits at the center of nearly every AI policy discussion.

If mandatory testing applies only to the most advanced frontier systems, many developers may view the proposal as a reasonable safeguard. If that definition gradually expands over time, startups and independent developers could face compliance costs that larger companies are better positioned to absorb.

Critics have long warned that regulations written for frontier AI can unintentionally strengthen the largest companies by raising barriers for everyone else.

Amodei’s statement does not answer where that line should be drawn.

More nuanced than many expected

For all the controversy surrounding Anthropic over the past week, Amodei’s response presents a more nuanced position than many critics had anticipated.

He does not argue that open-weight AI should be banned. He explicitly says the opposite.

His focus remains on keeping advanced computing infrastructure out of the hands of geopolitical rivals, limiting industrial-scale model distillation that could narrow the frontier AI race, and requiring safety evaluations for the most capable AI systems before release.

That still leaves one fundamental question dividing the AI industry.

Does releasing increasingly capable model weights make society safer through openness, independent scrutiny, and broader innovation?

Or does it increase risk by giving dangerous capabilities to actors who would otherwise never have gained access?

That debate has become one of the defining questions of the AI era. Amodei’s latest statement does not settle it. It does, though, make Anthropic’s position much harder to mischaracterize.