Archive — US — techcrunch
OpenAI tightens the screws on security to keep away prying eyes
OpenAI has strengthened its security measures against espionage, particularly after a rival Chinese startup reportedly used 'distillation' techniques to copy its models. Enhancements include limiting staff access to sensitive information, offline systems for proprietary tech, biometric controls, and a strict internet policy to protect against both external threats and internal leaks.
Key points
- OpenAI revamped its security after DeepSeek's competing model was launched.
- Allegation: DeepSeek copied OpenAI's models using distillation techniques.
- New security measures: limiting staff access to sensitive algorithms, isolating proprietary technology, implementing biometric controls, and maintaining a deny-by-default internet policy.
- Increased physical security and cybersecurity personnel reflect concerns over foreign adversaries and internal security vulnerabilities.
Why it matters
- Rising concerns over corporate espionage in the technology sector.
- Ongoing AI competition and technology poaching wars in the U.S.
- Past incidents of intellectual property theft in the tech industry and the need for enhanced cybersecurity.
OpenAI's proactive security overhaul underscores the escalating tensions in the AI field, highlighting the need for companies to safeguard their innovations against both external and internal risks.