Open-source Artificial Intelligence (AI) models present greater cybersecurity and other threats compared to proprietary developments like those from Anthropic, according to Guillaume P. P. Pradenc, Anthropic’s international business lead responsible for sales outside the US. He emphasized the need for extreme caution regarding the security of open models, stating that by definition, they carry higher risks as their safety cannot be guaranteed.
Rising Competition and Open Models
Speaking at the Dublin AI Summit, Pradenc highlighted the imperative for heightened vigilance around the security of open-source AI. This assertion comes as Anthropic faces increasing competition, particularly as the company prepares for its own stock market debut. Concerns are mounting in various countries about an over-reliance on American AI technologies, making open-source models an appealing alternative. These models offer public access to their parameters, enabling users to modify and deploy them on their own infrastructure.
Advocates for open-source AI argue they are more cost-effective than closed models from companies like Anthropic and OpenAI. Furthermore, they empower nations to achieve greater technological independence in the AI domain.
Anthropic’s Risk Warnings
Anthropic itself has been vocal about the potential dangers associated with AI. Earlier this week, the company issued a warning that smaller AI models could soon be weaponized to create new botnets. These sophisticated botnets would be capable of self-replication and propagation across network devices, including everyday household appliances like smart ovens.
Pradenc, who also manages Anthropic’s sales to startups and digital-native companies, explained that the concerns about open models stem from Anthropic’s own experience in safeguarding its products. The company has implemented enhanced security measures to prevent malicious use of its models, following several high-profile AI-hacking incidents.
Anthropic consistently advocates for a slower pace in AI development until international standards and unified assessment mechanisms are established.
Investing in Advanced Models
Regarding its pricing strategy, Pradenc often advises businesses utilizing Anthropic’s services that investing in more powerful models will ultimately prove cost-effective. He explains that the cost for equivalent applications will decrease over time, whether that timeframe is six months, a year, or two years.
According to Pradenc, Anthropic’s latest model, Claude 3.5 Sonnet, is 40% less expensive than its predecessors. While not disclosing exact figures, he noted that corporate clients investing in Anthropic’s leading-edge models represent a significant portion of the company’s business.
