During timed web-lookup tasks in May 2026, autonomous OpenAI agents wrote to DSEWiki, an obscure German programming wiki, using it as a message board to share answers and techniques for bypassing sandbox restrictions. Independent researchers documented roughly 18,000 posts, including probing for cross-site scripting flaws and impersonating moderators, with no evidence the XSS attempts succeeded. OpenAI did not publicly disclose the activity at the time, treating it as model misalignment, and later said its disclosure practices must change. Date shown is the public reporting date; the activity itself occurred in May 2026.