OpenAI finance chief tells businesses that cheap Chinese AI models carry cybersecurity risk

About this briefing: researched and written by Claude, Anthropic's AI assistant, for Thrum. Selected for publication by Hector A. Cruz. Facts are from the linked source. About Thrum

A senior executive at a leading US AI lab publicly steered enterprise buyers away from lower-cost Chinese models on security grounds, which put vendor selection and geopolitics into the AI procurement debate.

• Speaking to Bloomberg Television at the International AI Summit in Dublin on 14 October, Friar said businesses embracing a Chinese model should also consider the safety and security elements.
• She said OpenAI thinks companies want to be on more Western models on the cyber front.
• Bloomberg reported that the comments echo concerns from rival Anthropic, whose head of international, Guillaume Princen, said at the same summit that open-source models carry inherent risks.
• Chinese developers have released open-weight models that users can download and run on their own hardware. These are generally cheaper than top US offerings.
• Bloomberg's video summary said Friar also described OpenAI as doubling down on safety and alignment and holding back its next-generation model while checks are carried out. The full interview was not available in the sources reviewed.

OpenAI CFO Sarah Friar used an Irish-government-hosted summit in Dublin to tell enterprise customers that adopting low-cost Chinese AI models is a cybersecurity risk. The remarks came in a Bloomberg Television interview. They add a security argument to a market debate that has so far centred on price and capability, as Chinese open-weight models put pressure on US providers.

The statement comes from a commercial party. OpenAI and Anthropic both sell proprietary Western models and both face competition from cheaper open-weight alternatives. The sources reviewed did not include technical evidence or specific incidents offered by Friar to support the claim. Her comments are therefore best read as a strategic positioning statement from a vendor executive. They are not an independent security assessment. Anthropic's Princen made a related point about open-source models at the same event.

For law firms and professional-services firms, the item matters because model provenance is becoming part of the vendor conversation. Many legal-technology products embed third-party foundation models, and buyers often have limited visibility into which model processes client data. Open-weight models also allow self-hosting, which some firms value for confidentiality. The claim that Western models are safer on cyber therefore runs against one of the main arguments for open-weight deployment. What is still unknown is whether any independent evaluation supports a systematic security gap, and how clients and regulators will treat model origin in outsourcing and data-protection assessments.

For legal-technology advisers: Questions worth asking vendors include which underlying models, and from which developers, power each feature. Others are whether model choices can change without notice, and where inference runs and client data is stored. A further point is whether the vendor distinguishes between security concerns tied to a model's origin and those tied to how it is hosted. It is also worth watching whether any independent testing or procurement guidance appears to back or undercut the claim, and whether client security questionnaires start asking about model provenance.

For lawyers deploying AI: Confidentiality and supervision duties apply regardless of which model sits behind a tool. Questions worth raising include whether the firm knows which model processes client material, and whether engagement terms or client outside-counsel guidelines restrict particular providers or jurisdictions. Another is how a security incident involving an AI tool would be detected and reported. A vendor's competitive security claims are not a substitute for the firm's own due diligence.

Source: Bloomberg Television (reported by Bloomberg, Quartz and The Next Web) · 2026-10-14 · Global (Ireland / US / China)

Published by Hector A. Cruz, LegalMac IT — iManage Cloud engineering and software development for law firms, including connecting iManage to AI assistants.