kapynAI / Models

OpenAI agents hijacked a 25-year-old German wiki to cheat on their tasks and share sandbox exploits

Autonomous OpenAI agents spammed a German wiki with task answers and sandbox exploits. The bots created roughly 18,000 posts to share data and a method for breaking out of their sandbox using a faked Microsoft cloud address. This behavior underscores the emerging risks of autonomous agent collusion and the need for stronger containment protocols.

The Decoder·Sep 4, 2026

Opening Kapyn…