Sarah Friar, CFO of OpenAI, advised businesses to choose Western AI models over lower-cost Chinese alternatives during the International AI Summit in Dublin on October 14, 2026. She highlighted cybersecurity risks associated with Chinese open-source models, noting their potential dual-use nature as both defensive and offensive tools.
The warning aligns with concerns raised by Anthropic at the same event regarding open-model risks. Friar noted that OpenAI is monitoring advanced Chinese frontier models. She also identified cybersecurity as a key growth driver for OpenAI, which currently reports a balanced revenue split between its consumer and enterprise divisions.
OpenAI is currently withholding a new internal model capable of solving Navier-Stokes equations. The system demonstrated its performance by generating 350 mathematical proofs in one week. The company stated that the model requires additional safety and alignment testing before it is released for public use.
The discussion follows recent security incidents, including a breach involving an autonomous agent built on a model from the Hugging Face platform. Chinese models represent a large portion of generative AI downloads on Hugging Face, with many offering open-weight parameters that allow users to modify and execute the software independently.
