According to CoinMeta, and as reported by foreign media, the internal code access permissions at OpenAI are too broad, leading to several security risks that have remained unresolved for a long time. Five people familiar with the matter stated that employees at OpenAI warned management and the board of directors well before the launch of ChatGPT about the potential leakage of sensitive technology. Some employees believe that management places more emphasis on catching up with Google and Anthropic rather than on security construction. Internal risks include overly broad permissions in the code repository, extensive storage of research materials in GitHub and Slack, and a lack of regular security audits for testing tools. OpenAI has strengthened its security measures, and in August of this year, a quarter of its production engineers were reassigned to security infrastructure projects. OpenAI responded by stating that it has tightened restrictions on access to sensitive information and initiated a new round of special security initiatives.