OpenAI CFO Sarah Friar has urged businesses to think carefully before adopting Chinese AI models, warning that cybersecurity should be part of the decision. Her comments came as companies increasingly turn to cheaper AI tools from China, drawn by their performance and the flexibility offered by open-weight models.
Speaking to Bloomberg Television at the International AI Summit in Dublin on Wednesday (October 14, 2026) Friar said companies should pay closer attention to the safety and security implications of their AI choices. She suggested that Western models may be preferable when cybersecurity is a major concern, though her remarks did not establish that all Chinese models are unsafe.
Chinese AI developers have become serious competitors to US technology companies, releasing models that can handle coding, analysis and other demanding tasks at relatively low cost. Some open-weight models can also be downloaded and run on a company's own infrastructure, giving developers more control over how they use the technology.
That flexibility has helped these models gain traction among businesses and developers. However, the security implications depend on how a model is deployed, what data it can access and whether the organization has appropriate safeguards in place. For companies handling sensitive customer information, financial records or proprietary code, those questions can be as important as a model's price or performance.
Friar's warning comes as US AI companies face growing competition from Chinese developers. She argued that American firms must continue advancing their technology while taking safety seriously.
OpenAI has held back certain models to carry out additional safety work, she said. But slowing development too much could create an imbalance in who has access to powerful AI tools, including those that could present serious cybersecurity risks. The challenge for developers is to find a pace that allows innovation without overlooking potential misuse.
Anthropic has also raised concerns about the risks associated with open AI models. At the same Dublin summit, its head of international warned that open models require careful safety consideration.
For businesses, the issue is not simply where an AI model was developed. Data handling, hosting arrangements, access permissions and security testing all influence the level of risk. Running a model locally may offer more control, but it does not automatically make every deployment secure.
Also Read: OpenAI Safety Alert: 3 Fired Researchers Urge AI Monitoring
Organizations that choose AI technologies will need clear guidelines specifying which models the employees may employ and the information that may be exchanged through them. Proper security assessments and limitations of access to the critical systems will enable organizations to minimize the risk.
With the rivalry between AI companies from the USA and China growing, companies are forced to make a choice considering both the costs and capabilities.