Anthropic Executive Argues Open-Weight AI Models Pose Higher Risks
Anthropic's international business head Guillaume Princen stated during the Dublin International AI Summit that open-weight AI models carry higher cybersecurity and hazard risks compared to closed models. This perspective highlights the ongoing industry debate between closed-source AI safety controls and open-source technological autonomy, as governments increasingly seek independent AI sovereignty. Princen noted that Anthropic's concerns stem from its own product safety experiences, while warning that smaller models could soon drive novel botnets. He also mentioned that newer frontier models like Claude Opus 5.5 are priced 40% lower than earlier versions.
## BACKGROUND
Open-weight AI models make trained learning parameters, such as weights and biases, publicly accessible so users can modify or deploy them locally. Proponents value them for cost-effectiveness and national technology autonomy, whereas companies like Anthropic advocate for tighter controls and slower development until unified evaluation standards are established.