OpenAI has introduced a sweeping set of security upgrades to protect its research and proprietary technology from both internal leaks and foreign threats.
The new security measures follow a specific incident. In January, Chinese startup DeepSeek released a competing AI model.
OpenAI suspects DeepSeek may have replicated its technology using “distillation.” This method allows one AI model to imitate another, sometimes without direct access to the original source code.
OpenAI considers this a serious breach. In response, the company accelerated a pre-planned internal clampdown.
Information Tenting
OpenAI has now adopted what it calls “information tenting,” a policy that limits access to sensitive projects.
Only verified employees can participate in confidential work. Even then, they must follow strict communication protocols.
For example, during the development of the o1 model, only select team members were allowed to discuss the project. Also, these discussions could only take place in designated office spaces.
Biometric Controls and Offline Storage
The company has also updated its physical and digital security. Office areas now require biometric verification, specifically, fingerprint scans, for access.
Additionally, OpenAI now stores its most sensitive technologies offline, systems that operate without internet access.
The company also enforces a “deny-by-default” internet policy. Staff must request special approval to access external networks or online resources.
This structure reduces the potential for cyberattacks and data theft.
Increased Cybersecurity Staff
OpenAI has expanded its cybersecurity team. It has also increased physical security at its data centers.
The company appears to be guarding against both foreign adversaries and internal risks.
Leaks of CEO Sam Altman’s private remarks and increased poaching by rival companies suggest growing internal vulnerabilities.
OpenAI’s defensive posture addresses both of these pressures.
