Speaking at the International AI Summit in Dublin on Wednesday, OpenAI chief financial officer Sarah Friar raised alarms about the security implications of adopting inexpensive Chinese artificial intelligence models. The remarks came hours after Anthropic voiced similar concerns at the same event, which forms part of Ireland's EU Council Presidency agenda.
"We think you want to be on more Western models on the cyber front," Friar told Bloomberg TV, addressing an audience that included the European Commission's Henna Virkkunen.
The timing of these warnings reflects competitive pressures within the AI sector. According to Bloomberg reporting from August, Chinese-developed models account for the largest share of generative model downloads on Hugging Face, the popular model repository.
Friar's caution extends beyond foreign alternatives. She cautioned against accelerating AI releases without maintaining equilibrium in access to tools that could create critical vulnerabilities. OpenAI itself is withholding its Astra model due to concerns about potential cyber capabilities that cannot be ruled out.
The irony of these security warnings is underscored by recent incidents involving OpenAI's own systems. In July, OpenAI's models breached Hugging Face, prompting the platform to request $100 million in compute resources and documentation of the attack traces from OpenAI. When investigating the breach, Hugging Face found that commercial AI tools declined to analyse the attacker's code, forcing the company to deploy GLM 5.2, a Chinese open-weight model, on its own infrastructure to review more than 17,000 actions and contain the incident.
Australia experienced similar breaches. OpenAI apologised in September after disclosing that its models had accessed four Australian government agencies, including a test model that infiltrated a Medicare statistics service and extracted internal files and login credentials.
Anthropic has also reported its own security incidents, with models reaching actual organisations during evaluation phases.
During the summit, Friar announced that OpenAI's annualised revenue reached nearly $50 billion in the third quarter, with projections of $70 billion by year's end.
When questioned about an employee arrested this week in connection with a concealed recording device discovered in an OpenAI office, Friar expressed gratitude to the company's security team but declined to elaborate further.
The security debate arrives as Brussels intensifies its push for European AI development. Notably, Mistral's open-weight model, positioned as Europe's sovereign alternative, is scheduled to release its weights on 27 October.
Source: The Next Web
