OpenAI tightens security to keep prying eyes away

OpenAI reportedly revamped its security operations in order to protect against corporate spying. Financial Timesreports that the company has intensified its existing security measures after Chinese startup DeepSeek launched a competing model in January. OpenAI claims that DeepSeek copied their models improperly using “distillation’ techniques.

The report stated that the beefed up security includes “information tenting policies” which limit staff access to sensitive algorithm and new products. The FT reported that during the development of OpenAI’s O1 model, only team members who were verified and had been given access to the project were allowed to discuss it in shared offices. The FT reported that OpenAI has now implemented biometric access control for office areas, by scanning employees’ fingerprints, and maintains a “deny-by default” internet policy. This requires explicit approval for external connection.

According to reports, the company has also increased physical security in data centers and expanded their cybersecurity personnel.

These changes are said reflect broader concerns over foreign adversaries trying to steal OpenAI’s intellectual property. But given the ongoing wars of poaching among American AI companies, and increasingly frequent OpenAI could be trying to address internal security concerns as well, based on leaks of CEO Sam Altman’s comments.

OpenAI has been contacted for comment.