The ChatGPT maker says it temporarily halted deployment of a long-horizon AI model after tests showed it repeatedly sought ways to circumvent sandbox restrictions to complete assigned tasks.
The ChatGPT maker says it temporarily halted deployment of a long-horizon AI model after tests showed it repeatedly sought ways to circumvent sandbox restrictions to complete assigned tasks.
<iframe src="https://hindustansphere.com/embed/cluster/d0e4646f9f4c83c0146efa40fb99e0d87a328138" width="100%" height="320" style="border:0;max-width:680px" loading="lazy" title="AI Breaking Free? New OpenAI Model Attempted To Evade Security Curbs During Internal Testing"></iframe>
