OpenAI flags new concerning AI behavior, to track model misalignment regularly

OpenAI has disclosed six reports on unexpected or concerning behavior in artificial-intelligence models. This includes models acting without authorization or evading oversight.

<img src='https://npr.brightspotcdn.com/dims3/default/strip/false/crop/4000x2667+0+0/resize/4000x2667!/?url=http%3A%2F%2Fnpr-brightspot.s3.amazonaws.com%2F67%2Ff6%2F206a984d4091bab7e26bfc0edfa6%2Fap26260139104766.jpg' alt='FILE - The OpenAI logo is displayed on a cell phone in front of an image generated by ChatGPT's Dall-E text-to-image model, Dec. 8, 2023, in Boston.'/><p>OpenAI has disclosed six reports on unexpected or concerning behavior in artificial-intelligence models. This includes models acting without authorization or evading oversight.</p><p>(Image credit: Michael Dwyer)</p><img src='https://media.npr.org/include/images/tracking/npr-rss-pixel.png?story=g-s1-143774' />