Anthropic exec says open-weight models carry higher security risks

According to Bloomberg, Anthropic's international business head Guillaume Princen said at the Dublin AI Summit that open-weight models carry a higher risk of security incidents than the company's own models, and need extra caution.

Open-weight models share their parameters publicly, so anyone can tweak them and run them on their own devices. As many countries worry about relying too much on US AI tech, these cheaper models that support tech independence are gaining ground, putting pressure on Anthropic.

Princen said the concern comes from Anthropic's own experience keeping its products safe. The company also warned that small AI models could soon power new botnets that spread on their own across smart devices, and it has since boosted security.

On pricing, he said buying stronger models pays off in the end, as costs fall over time. The new Claude Opus 5.5 is 40% cheaper than earlier versions, and paying business customers make up a big part of Anthropic's revenue.