Anthropic's head of international business Guillaume Princen, speaking in an interview with Bloomberg Television at the International AI Summit in Dublin, said that compared with the company's own AI models, open models pose "higher risks" in cybersecurity incidents or other dangers. He noted that on safety issues, one must be especially cautious with open-source models, and emphasized that he cannot speak for open models, but by definition these models do carry higher risks.
The remarks come as Anthropic faces increasingly fierce competition in several countries. The current focus as an alternative to closed-source models is open-weight AI models, which publish their parameters, allow users to modify them, and can run on their own hardware. Proponents argue that compared with the tightly controlled closed-source models offered by Anthropic and OpenAI, open-weight models are lower cost and better for preserving national sovereignty. French AI company Mistral AI recently released an open-weight model, providing Europe with a sovereign model alternative; at the same time, a series of similar open models launched by Chinese companies are gaining increasing attention in overseas markets.
Anthropic's safety concerns over open models continue to grow
Anthropic has recently stepped up its warnings about AI capabilities. Earlier this week, the company warned that small AI models could soon give rise to a new type of botnet capable of self-replicating among connected small devices such as smart ovens. Princen also oversees Anthropic's sales to startups and "digital native" enterprises. He said concerns about open-weight models stem from Anthropic's efforts to ensure product safety. Following several high-profile AI hacking incidents, Anthropic said it has added extra security safeguards to prevent its models from being used for malicious purposes.
The company has also consistently argued that AI development should be slowed until international standards and a unified evaluation mechanism are established.
Addressing the company's higher-priced models, Princen said he tells enterprise customers that buying more powerful models will ultimately pay off. "Always invest in frontier technology. Because six months, one year, or two years from now, the cost of using these applications will also come down," he said. He also noted that the company's latest Claude model, Opus 5.5, is 40% cheaper than its predecessor. He added that enterprise customers purchasing Anthropic's "frontier" models account for a "very large share" of the company's business, but he did not disclose specific figures.
Regarding Anthropic's IPO plans, Princen declined to comment; he also would not disclose how much the company has invested in training new models.