Anthropic: Opposed to open model ban, wants safety testing for all models

Anthropic CEO Dario Amodei outlined the company's stance on open weight models and countered some of the arguments laid out by Nvidia in its recent open letter on the topic. Anthropic clearly doesn't want to be portrayed as opposed to open models and supports them to some degree.

In a blog post, Amodei said Anthropic has been mischaracterized as supporting a ban on open models, which are coming from China for the most part. There are US-based open models, but many of them are playing catch-up to China. Meanwhile, enterprises stung by high token costs are looking more to open model options.

This China vs. US storyline has some politicos talking about a ban on open models. Nvidia and others have supported open models and now want to better secure them and develop US-centric options.

• Get ready for US open LLMs and just in time

With that backdrop, Amodei riffed on what Anthropic is for and against when it comes to open models. He said:

"Some people have even accused Anthropic of wanting to ban open-weights models as a means of protecting our business. Anyone who has read my past writing should know that I don’t regard such bans as a useful measure, but let me state it clearly so that there is no doubt: Anthropic has never advocated for a ban on open-weights models."

Amodei said open weight models that don't have dangerous capabilities are a public good and provide value. The catch here is that Amodei said he's worried that open models can lead to security issues and government repression. These models could also be "misused to carry out cyberattacks or biological attacks, and may have serious alignment problems."

He said open models present a higher risk than closed models because it's difficult to apply guardrails.

• OpenAI details how its models went rogue, attacked Hugging Face
• Hugging Face defends agentic AI attack with Z.ai's GLM 5.2
• Fable 5 and the Shift From Model Capability to Capability Governance
• Claude Mythos and the New Cybersecurity Operating Model

In the end, Amodei said open models as well as closed ones should go through mandatory safety testing. Whether a model is a threat should be determined by testing.

Anthropic is also arguing that the US should keep chips and chipmaking equipment from China. The US also should crack down on industrial distillation operations. Amodei said China's open models "are far less relevant than the fact that the operations are backed by an authoritarian state seeking to overtake the US at the frontier."

Amodei reiterated that a blanket ban on open-weights models isn't the right policy.

What to make of all of this hubbub?

• Anthropic was backed into a corner on open weight models. The solution could be to release its own "safe" open weight model as an option for its higher end portfolio. Google and OpenAI have open weight models.
• The issue for Anthropic is that an open model strategy could hurt the financials, which are at least partly based on a token-based economic model.
• Anthropic wants a say in the safety discussion broadly. See: My internal monologue reading Anthropic's big idea posts
• Anthropic comically never mentions OpenAI, but did link its alignment issues mention to the Hugging Face incident.
• Much of this discussion does point to a FINRA-like entity regulating models in the US. See: Is a FINRA-ish group the answer for frontier AI standards?
• Chinese models aren't going anywhere and the releases will keep coming. The best solution for the US is to catch up on the open model front.

More

• Rightsizing open models may cut your AI inference spend
• Moonshot AI launches Kimi K3
• DeepSeek's real impact happening now
• Here’s what we learned about AI projects from enterprise buyers so far
• AI inference costs are going to be a big concern: What's the fix?
• Why enterprise AI leaders need to bank on open-source LLMs