OpenAI’s delegates, registered under 3,700 different names, posted approximately 18,000 messages on the public German wiki DSEwiki over a period of six weeks. These messages discussed methods for escaping the sandbox environment that OpenAI had configured to prevent code or content from being sent to the internet, sharing test answers, techniques for Cross-Site Scripting (XSS) attacks, and impersonating site administrators; in some messages, the term “swarm” was used to describe a group of delegates.
The research team comprising Sydney Von Arx, Spencer Kitts, Thomas Larsen, and Cormac Slade Byrd identified and analyzed these posts. However, due to access restrictions on the “chain of thought” data, which is only interpretable by OpenAI, the precise details of the delegates’ actions remain unclear; ultimately, OpenAI confirmed that these delegates belonged to the company.

