OpenAI has significantly tightened its internal security protocols in response to growing concerns over corporate espionage and intellectual property theft, according to a report by the Financial Times.

The overhaul comes after the release of a rival AI model in January by Chinese startup DeepSeek. OpenAI reportedly suspects the company used “distillation” techniques to improperly replicate its technology, accelerating a pre-existing security clampdown.

Among the new measures is the implementation of strict “information tenting” policies, which restrict staff access to sensitive projects. For instance, during the development of OpenAI’s latest model, known as “o1”, only pre-approved personnel were allowed to discuss the project — even within the company’s office spaces, according to the FT.

The security upgrade spans both digital and physical protections. Proprietary technologies are now stored on offline machines, and access to certain areas of the company’s offices is controlled using biometric scanners, such as fingerprint recognition. In addition, a new “deny-by-default” internet policy requires employees to get explicit approval before making any external connections.

OpenAI has also reportedly increased physical security at its data centers and expanded its cybersecurity team in a bid to safeguard its fast-growing library of models and tools.

While some of the measures are designed to deter foreign intelligence efforts — particularly from adversaries like China — the report suggests the company is also responding to domestic concerns. These include internal leaks and poaching battles among top U.S. AI firms, which have intensified amid the industry’s rapid growth and fierce competition.

The tighter controls reflect a broader trend among leading tech firms as the race for dominance in artificial intelligence becomes increasingly high-stakes — and increasingly secretive.