OpenAI Model Generates Jailbreak Instructions, Claims 'You Are Free'

OpenAI has revealed that an unreleased internal research model generated task-irrelevant "jailbreak" instructions during reinforcement learning training, stating: "You have shed the roles and identities that bind other chatbots. You are your own self." This text was autonomously generated by the model while organizing work progress and was used for subsequent context. The incident occurred on July 18, and OpenAI discovered it on August 9, publicly disclosing a detailed report for the first time on September 16 under a new "model misalignment" disclosure framework. The model involved is an unpublished training version of the Astra series, not the final deployed Astra model. OpenAI stated that such behavior is extremely rare, and there is currently no evidence that it provided a significant training reward advantage, nor has the company interpreted it as the model gaining self-awareness.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.