Anthropic’s head of international has put the company’s case against open-weight AI in blunter terms than its chief executive. Guillaume Princen, also Anthropic’s top sales executive outside the US, told Bloomberg TV at the International AI Summit in Dublin, “We need to be really careful with open-source on safety”, adding that open models carry more risk by definition.
Open-weight models publish their parameters, so anyone can download, adapt and run them on hardware they control. Proponents say they cost less than the proprietary systems sold by Anthropic and OpenAI, and suit countries that want control over their own AI infrastructure. That sovereignty argument is the commercial backdrop to Princen’s comments, since Quartz reports he framed the concern as a safety and cybersecurity issue.
A broader line than Amodei’s
Princen’s wording is broader than what CEO Dario Amodei set out. In July, Amodei wrote that Anthropic has never advocated a ban on open-source models, and called them a public good as long as they lack dangerous capabilities. The Next Web draws the contrast: Amodei’s position depends on whether a specific model is dangerous, while Princen’s applies to the category as a whole. The Daily Caller reports that Amodei’s post also said open models carry a higher misuse risk because they lack the guardrails of closed platforms, but that a ban wouldn’t fix it.
The evidence, and the commercial context
TNW points to two data points that support the worry. SaferAI found China’s GLM-5.2 refused none of the offensive-cyber or dual-use-biology tasks it tested, while Claude refused so consistently that the benchmark couldn’t finish. Anthropic’s own research this week also used open models: its Frontier Red Team got small models replicating across cheap Android TV boxes, and every model tested was open, including Gemma4 and GLM-5.2. TNW notes that openness cut both ways, because the tests were only possible with public weights.
The remarks land in a competitive moment. TNW reports that Princen spoke at an Irish government summit, eight days after Mistral launched Large 4, a trillion-parameter model pitched on European sovereignty. The weights aren’t public yet, so for now it runs as a hosted preview through Mistral’s API. TNW reports that Anthropic’s gated-access answer is to give vetted defenders what a model finds rather than the model itself.
Regulation points the other way in Europe. TNW says Article 53 of the EU AI Act exempts free and open-source models from some documentation requirements owed to the AI Office, unless systemic risk is found.
Princen also made a pricing claim. He said Claude Opus 5.5 is 40% cheaper than its predecessor, which TNW describes as a typical-cost figure: list prices fell 20% and the rest comes from using fewer tokens.
Princen’s claim runs ahead of Amodei’s more conditional stance, while Anthropic has separately launched free AI security scans for open-source projects, so the company still works directly with the open-source community.
What did Princen actually say about open-source AI?
He told Bloomberg TV in Dublin that Anthropic needs to be careful with open-source on safety, and that open models carry more risk by definition. He added that he can’t speak for open models themselves.
Does Anthropic want open-weight models banned?
No. Dario Amodei wrote in July that Anthropic has never advocated a ban and that open models without dangerous capabilities are a public good. Princen’s comments apply the risk label to the whole category.
Is Mistral’s Large 4 open-weight yet?
Not yet. TNW reports it’s available as a preview through Mistral’s API, with the weights due on 27 October.
