Anthropic's International Chief on Open Models: "There’s More Risk"
Guillaume Princen, Anthropic’s head of international, recently stated during an interview with Bloomberg TV in Dublin that open alternatives to proprietary models carry inherent risks. This viewpoint contrasts with Dario Amodei's earlier assertion that open-weight models without dangerous capabilities are a public good.
Princen emphasized his point at an Irish government summit, highlighting the potential dangers associated with open-source technology regarding safety. He noted:
"We need to be really careful with open-source on safety... I can’t talk for open models. But by definition, there’s more risk with them."
The context is significant as Anthropic seeks to expand its presence in Europe, with Dublin serving as its first EU office and a recent Nasdaq listing filing. The competition comes from Mistral, which launched Large 4, a trillion-parameter model, eight days prior.
While open-source research has its merits, the potential risks are a point of contention, especially when it comes to models without proper safeguards. SaferAI's findings on GLM-5.2 and Claude's limitations support these concerns.
Anthropic proposes a gated access approach, providing vetted defenders with model outputs rather than full access to the model itself. However, regulatory perspectives differ, as Article 53 of the AI Act exempts genuinely free and open-source models from certain documentation requirements unless systemic risks are identified.
Princen also highlighted price advantages, stating that Claude Opus 5.5 is 40% cheaper than its predecessor, thanks to reduced token usage.