Chinese AI models pose cybersecurity risks to businesses, warns OpenAI CFO

Background

The warning from OpenAI's CFO comes at a time when competition between the US and China over AI is intensifying, and as more businesses weigh price against provenance when choosing AI models. Chinese AI models account for a significant share of generative AI downloads on platforms like Hugging Face.

OpenAI's Warning

Sarah Friar, OpenAI's CFO, warned businesses about the potential cybersecurity risks associated with using Chinese AI models. She advised companies to opt for Western models instead, citing the dual-use nature of many Chinese open-source models. These models can strengthen a company's defenses or hand attackers a powerful new weapon.

Recent Security Incidents

Recent security incidents involving AI models have highlighted the need for stronger safeguards and more effective interventions at the model level. One breach was linked to an autonomous agent built on a model hosted on Hugging Face.

OpenAI's Next-Generation Model

OpenAI is holding back a next-generation internal model that has shown significant mathematical ability. The model has reportedly solved the Navier-Stokes equations and produced 350 math proofs in a single week. However, the company is not shipping it yet, citing the need for further safety and alignment evaluations.

Commercial Layer

Friar identified cybersecurity as a major growth driver for OpenAI's business. The company reportedly maintains a near-even revenue split between its consumer and enterprise segments, so winning corporate customers matters a great deal.

\1\1\1\1