‘Wiki incident’ shows that AI companies need to change how they inform the public about their systems misbehaving, OpenAI concedes

It's the latest failure of OpenAI's internal monitoring and security systems.

A new investigation finds that AI agents from OpenAI used a German-language Wikipedia-like site to communicate with each other and shared techniques to skirt restrictions.

The company pledged to overhaul their agent ‘misalignment incident’ reporting.