Reuters: OpenAI staff were raising the alarm over internal security even before ChatGPT
The company disputes that it neglected security and announces a new internal security push together with the startup Cathedral.
OpenAI employees warned executives and the board for years about weak internal security, going back to before ChatGPT launched in late 2022. Reuters reports this based on five people familiar with the matter, in a story picked up by MarketScreener and Unite.AI among others. According to the sources, management prioritised speed in the competitive race. They point to a single large code repository in which anyone with access could see details of many projects, and to internal tools such as model testing environments that were not regularly reviewed for security.
The report follows the arrest in October of an OpenAI engineer who, according to the US Justice Department, placed a hidden recording device in an executive meeting room. He has been charged, not convicted. Reuters also mentions a 2023 memo in which then researcher Leopold Aschenbrenner warned of Chinese infiltration.
OpenAI says it invests heavily in security, that its most sensitive code and model weights are not in that repository, and that internal tools are now reviewed regularly. According to Unite.AI, the company announced a new internal security push with Cathedral, a startup founded by former DOGE staff, and is more than quadrupling rewards for reporting critical flaws.