Anthropic Executive Says Open-Weight Models Riskier Than Its Own

At the Dublin International AI Summit on October 14, Anthropic's international business lead said open-weight AI models are riskier than Anthropic's own models, adding that he cannot guarantee their safety.

He said the concern stems from Anthropic's experience securing its own products. Earlier this week, the company warned that small AI models could soon be used to power new botnets that self-replicate across small internet-connected devices such as smart ovens. Open-weight models make their parameters publicly available, allowing users to modify and deploy them.

Anthropic has argued that AI development should slow until international standards and unified evaluation mechanisms are established. The executive also said companies should keep investing in frontier models, noting that the cost of running the same application typically falls over time.

He revealed that Anthropic's latest Claude Opus 5.5 model is 40% cheaper than earlier versions. Paid enterprise customers using Anthropic's frontier models account for a "significant part" of the business, though he did not disclose the exact proportion.