OpenAI warns of Chinese AI cybersecurity risks and espionage threats

OpenAI executives and security protocols have highlighted growing concerns regarding cybersecurity and espionage within the artificial intelligence sector. At the International AI Summit in Dublin, OpenAI CFO Sarah Friar warned that companies adopting low-cost Chinese AI models may be exposing themselves to significant cybersecurity risks. Friar advocated for the use of Western models to mitigate these threats, noting that OpenAI is closely monitoring the rapid advancement of Chinese open-weight models, which are widely downloaded on platforms like Hugging Face.

In addition to external model risks, OpenAI has implemented internal security training to protect its technological secrets from human intelligence threats. The company has warned employees about “honey trap” tactics, where foreign intelligence services may use romantic or social approaches to gather information. Security measures advised include using privacy screen protectors, cleaning whiteboards after meetings, and avoiding company-branded clothing in public.

These concerns follow a recent incident involving the alleged arrest of a Saudi engineer, Abdulrahman Alfozan, who was accused of planting a recording device behind a television in an OpenAI meeting room. Similar security warnings have reportedly been issued to employees at other AI firms, including Anthropic and Google.