OpenAI CFO Flags Hidden Risk in Cheap Chinese AI Models Moz Farooque ACCA Wed, October 14, 2026 at 3:39 PM EDT MSFT -1.68% OPAI.PVT Trade MSFT on Coinbase Trading disclosure Trading disclosure The above button links to Coinbase. Yahoo Finance is not a broker-dealer or investment adviser and does not offer securities or cryptocurrencies for sale or facilitate trading. Coinbase pays us for certain activity generated through this link. Prices displayed are informational.
Cheap AI models can save companies money. The harder question is what happens when those tools mishandle sensitive data or make a costly decision no one knows how to unwind.
Microsoft-backed (MSFT) OpenAI CFO Sarah Friar is warning businesses to think carefully before relying on lower-cost Chinese AI models, arguing that some open-source systems can create cybersecurity risks for corporate networks.
Warning! GuruFocus has detected 5 Warning Sign with MSFT.
Is MSFT fairly valued? Test your thesis with our free DCF calculator.
Speaking to Bloomberg Television, Friar said companies should favor Western AI software when protecting sensitive information.
Anthropic executive Guillaume Princen made a similar argument, saying open models can carry additional risks because users are able to alter the underlying code.
Those comments come as Chinese AI models gain traction partly because they are cheaper and easier to download than many premium U.S. alternatives.
That puts businesses in a familiar bind: lower software costs on one side, potentially higher security risk on the other.
A separate Wharton School study found that managers are still divided over who should take responsibility when workplace AI makes a serious error.
Respondents split roughly evenly among the employee using the bot, that employee's manager and the company's risk team.
That lack of agreement becomes more important as companies give AI agents access to customer records and internal systems.
The danger is not necessarily a dramatic system failure.
Sometimes it can begin with something much smaller.
The report cited an example in which a retail worker configured a bot incorrectly and it sent erroneous refund notices to 18,000 customers within hours.
That is what changes when software goes from answering questions to taking actions.
A human mistake that once affected one transaction can suddenly be repeated thousands of times.
Friar also argued that U.S. AI companies cannot respond by simply slowing development.
Her concern is that moving too cautiously could allow foreign competitors to gain ground.
OpenAI, meanwhile, is still scaling quickly. Friar said the company reached nearly $50 billion in annualized revenue during the third quarter and expects that figure to reach roughly $70 billion by year-end.
What stands out to me is that these two stories are really about the same problem.
Companies want AI to become cheaper, faster and more autonomous. But the more authority they hand to software, the more important questions around security, permissions and responsibility become.
For investors and businesses, I would watch enterprise AI security spending, adoption of autonomous agents and whether companies begin assigning clearer accountability when AI systems make mistakes.
The next stage of workplace AI will not be judged only by what the models can do.
It will also be judged by who takes responsibility when they do the wrong thing.
