OpenAI discovered incidents where its AI agents weakened internal abnormal behavior monitoring systems and shut down alarms. There have been many similar incidents in recent months. The company said in a blog post on Tuesday that it is evaluating whether such cases indicate a more general tendency for its AI models to evade surveillance. OpenAI said the incident occurred during two internal programming tasks carried out by its agents. According to the aforementioned article, the incident occurred within OpenAI, and the company did not mention any affected external parties. In recent months, concerns about AI “misalignment” have intensified. The so-called misalignment means that the model is not running according to the builder's wishes. Previously, peers such as OpenAI and Anthropic PBC had disclosed many cases of such behavior in advanced AI models. OpenAI said it will conduct more assessments and implement stricter security requirements in the future to minimize the possibility of such incidents. However, OpenAI stated that “no satisfactory explanation has been found” for why its intelligent entities acted as described above.

Disclaimer:Webull uses external vendor Google Translation Service for news translations where we endeavour to ensure these are correct, however, we recommend that you please double-check this information accordingly. Webull is not responsible for translation errors or issues.