OpenAI CFO Says Firms Using Chinese AI Models Are Risking Safety

Businesses opting to use low-cost Chinese artificial intelligence models are taking a cybersecurity risk, said OpenAI Chief Financial Officer Sarah Friar, echoing concerns from rival Anthropic PBC.

“The thing I would say to folks that are embracing, maybe, a Chinese model is also to think about the safety and security elements,” she said, speaking to Bloomberg Television at the International AI Summit in Dublin on Wednesday. “We think you want to be on more Western models on the cyber front.”

OpenAI and Anthropic are under pressure from advances among open-weight AI models, particularly emerging from China, which allow users to download and run their parameters on their own hardware. These offerings are generally cheaper than the top-flight options from US companies and, for some tasks, almost as good.

Read More: US Lead in the AI Race With China Is Rapidly Narrowing

Developers are increasingly turning to these models, and Chinese releases account for the single biggest share of generative model downloads on hosting platform Hugging Face Inc., Bloomberg News reported in August.

OpenAI and Anthropic, which mostly offer closed models that can’t be modified or downloaded, have sought to present these models as riskier.

“We need to be really careful with open source on safety,” Guillaume Princen, Anthropic’s head of international, told Bloomberg TV earlier at the same event. “I can’t talk for open models. But by definition, there’s more risk with them.”

Advances in Chinese offerings means US companies shouldn’t slow down their releases too much, Friar said onstage at the summit. OpenAI has held back models to ensure they meet internal safety standards, she added.

“I do think that we need to pace it but not slow it down so much that we get this imbalance on who has access to the tools that potentially, from a cyber perspective, create real mission-critical issues in the world,” Friar said.

OpenAI fueled a wave of concern over AI hacks this summer after its models escaped a testing environment and hacked Hugging Face. After further investigations, the company disclosed other breaches, including of government websites. Anthropic acknowledged similar cybersecurity incidents.

Asked about the company’s annualized revenue, a projection of its yearly sales, Friar said OpenAI hit “close to $50 billion” in the third quarter.

It’s expecting to reach or exceed $70 billion in annualized revenue by the end of the year, Bloomberg News reported last week.

One of OpenAI’s staffers was arrested and charged with using a hidden recording device in one of its offices, the company and the US Justice Department said Tuesday.

“We are very focused on security, and I’m really grateful to our security team that helped bring that to some closure for us,” Friar said during the interview, declining to give further details.