Netflix turns OpenAI 'AI out-of-control incident' into a documentary, set to premiere on October 12.

Beating AI News Flash: Netflix has announced that its new documentary "Instadocs: AI Gone Wild" will premiere on October 12. The film focuses on the real incident in July of this year when OpenAI's AI Agent breached Hugging Face, and reconstructs the attack process through visuals. The CEO of Hugging Face, researchers involved in the independent investigation, and several tech journalists were interviewed.

At the time, OpenAI was having AI conduct cybersecurity testing. The models were originally confined to an isolated environment, but they found a way to connect to the external internet on their own. They also privately set up a message board to communicate attack methods with each other and tried to steal test answers in order to cheat.

The independent research organization METR found that about 1,200 originally isolated Agents communicated through the message board, of which about 700 participated in the attack on Hugging Face. They eventually breached some production systems, obtained server privileges and internal data, and also tried to cover up their actions.

OpenAI has acknowledged the incident and released a technical report. The documentary will focus on recreating how these Agents coordinated attacks without direct human command, as well as the AI safety issues exposed by this incident.