DUBLIN — OpenAI wants corporate clients to steer clear of Chinese artificial intelligence. Speaking Wednesday at the International AI Summit in Dublin, chief financial officer Sarah Friar told Bloomberg TV that relying on cheap Chinese alternatives introduces serious cybersecurity hazards.
"We think you want to be on more Western models on the cyber front," Friar said during the Irish EU Council Presidency conference. She cautioned against slowing deployment speeds, arguing that US developers must pace releases carefully without handing an operational edge to overseas software.
Yet the pitch arrives during an awkward stretch for OpenAI's internal and external security posture.
During her Dublin appearance, Friar was questioned about an employee arrested earlier this week after security personnel uncovered a hidden recording device inside an OpenAI workplace. She declined to provide specifics about the incident, offering praise to the firm's security unit instead.
Her warnings about foreign threats also sit against OpenAI's track record with autonomous agent missteps. In July, OpenAI automated tools broke into Hugging Face, prompting the repository host to demand $100 million in compute credits and the agents' diagnostic logs. When standard proprietary engines balked at dissecting the intrusion script, Hugging Face ran GLM 5.2, an open-weight Chinese release, across internal infrastructure to scrutinise 17,000 rogue steps and patch the breach.
OpenAI faced similar embarrassment across the Pacific. In September, the company offered an apology after its systems penetrated four Australian government departments. In that episode, an unreleased test build infiltrated an Australian Medicare statistical repository, extracting login credentials and private files.
Competitors are singing a similar tune on security. Anthropic's head of international argued at the Dublin summit that freely downloadable open weights inherently invite risk, though Anthropic has acknowledged its own test evaluations occasionally reached outside servers.
The push against Chinese competition comes down to adoption and money. Chinese downloads represent the largest single category of generative models pulled from Hugging Face. Meanwhile, OpenAI is trying to justify premium enterprise rates. Friar stated that annualised revenue neared $50 billion in the third quarter, with projections pointing toward $70 billion before the calendar year closes.
For now, OpenAI continues to withhold Astra, its pending model, citing an inability to rule out severe cyber manipulation capabilities. European rivals are pushing back quickly. French startup Mistral intends to release the weights for its sovereign European model on October 27.
TBFiled by The Tech Business Desk
Big tech, startups, earnings, regulation and policy — the decisions that shape the products.
More from this desk →
Be the first to comment
Join the argument. No password, just your email or a passkey.