Guillaume Plas, head of international business at Anthropic, stated that open-weight AI models carry higher security risks compared to the company's proprietary models. Speaking at an AI summit in Dublin, Plas emphasized that open models, which allow users to modify parameters and deploy them on local hardware, are inherently more difficult to secure.
Anthropic is currently facing increased competition as countries seek to reduce reliance on US-based AI technologies. While proponents argue that open-weight models offer cost savings and greater technological sovereignty, Anthropic maintains that its closed-source approach provides better protection against cyber threats and malicious activities.
The company has recently warned that small AI models could be exploited to create new types of botnets capable of spreading across connected devices like smart ovens. In response to recent security incidents, Anthropic has implemented additional safety measures and continues to advocate for slower development cycles until international standards are established.
Regarding pricing, Plas noted that investing in advanced models remains cost-effective over time as application costs decrease. The latest Claude Opus 5.5 model is 40% cheaper than its predecessors. Enterprise clients paying for frontier models represent a significant portion of Anthropic's business, though the exact figures remain undisclosed.
