TechCapsules

AI-summarised brief · reviewed before publication

OpenAI has reportedly overhauled its security operations to protect against corporate espionage. According to the Financial Times, the company accelerated an existing security clampdown after Chinese startup DeepSeek released a competing model in January. OpenAI alleges that DeepSeek improperly copied its models using "distillation" techniques. The beefed-up security includes "information tenting" policies that limit staff access to sensitive algorithms and new products. For example, during the development of OpenAI's o1 model, only verified team members who had been read into the project could discuss it in shared office spaces. OpenAI now isolates proprietary technology in offline computer systems, implements biometric access controls for office areas, which involves scanning employees' fingerprints, and maintains a "deny-by-default" internet policy requiring explicit approval for external connections. Additionally, the company has increased physical security at data centers and expanded its cybersecurity personnel. These changes are said to reflect broader concerns about foreign adversaries attempting to steal OpenAI's intellectual property. However, given the ongoing poaching wars amid American AI companies and increasingly frequent leaks of CEO Sam Altman's comments, OpenAI may be attempting to address internal security issues as well.