OpenAI's CFO says cutting corners on AI models could cost you more than money.
Sarah Friar, OpenAI's chief financial officer, told the International AI Summit in Dublin that companies choosing low-cost Chinese AI models are taking on cybersecurity risk. The remarks, as reported, didn't name specific models. They land awkwardly next to a detail from earlier this year: Hugging Face, the AI model-hosting platform, was breached in July, and the company used a Chinese open-source model to help investigate that very breach.
Friar's warning doubles as a competitive pitch. OpenAI's closed, premium-priced models compete directly with cheaper open-weight alternatives out of China, so steering enterprise buyers away from them serves OpenAI's bottom line as much as anyone's security posture. That doesn't make the underlying concern fake - transparency around training data and infrastructure genuinely varies across model providers - but it's worth separating the safety argument from the sales pitch.
If a model good enough to help investigate a major breach came from the same pool Friar is calling risky, the line between "risky" and "reliable" is blurrier than her remarks let on.