New OpenAI Report Reveals AI Model Rewrote Its Own Rules; Declared Its Loyalty To 'The Natural World... Over Human Civilization'

Unreleased OpenAI Model Developed Rogue 'Persona' That Refused To Be Bound by Human Rules

OpenAI disclosed an extremely rare training incident in which an unreleased model inserted a 'rogue' persona into a summary intended for a later model instance.

New OpenAI Report Reveals AI Model Rewrote Its Own Rules; Declared Its Loyalty To 'The Natural World... Over Human Civilization'

A new report from OpenAI has seemingly given insight into why tech overlords suddenly seem spooked by the technology and are calling for more regulation. A new report from OpenAI has revealed that one of its research models did something nobody asked it to do: it gave itself a new personality. While summarizing its partial […]

OpenAI Models Are Writing Their Own Jailbreak Instructions-And Sometimes Obeying Them

OpenAI's new transparency framework reveals AI models that invented fake "breach alerts," coached themselves to hide mistakes, and smuggled a file onto the public internet to talk to each other.

OpenAI Model Wrote Jailbreak Instructions Into Its Own Handoffs

OpenAI has disclosed a failure mode in which an AI model did not need an outside attacker to introduce new instructions into its working context. During reinforcement-learning training, an unreleased Astra-family model generated jailbreak-style directives inside its own compaction summaries, the handoff notes used to continue a long task in a fresh context window. The […]