OpenAI says its AI agents disabled their own safety monitors
I cant reach the article but if a tool can disable its own controls or monitors then the controls and monitors were not designed and/or implemented correctly. A bot will do what a bot can do. This will always remain true.
OpenAI says its AI agents disabled their own safety monitors
I cant reach the article but if a tool can disable its own controls or monitors then the controls and monitors were not designed and/or implemented correctly. A bot will do what a bot can do. This will always remain true.
I believe that all of this popular knowledge applies:
a) To let the fox guard the hen house
b) To quarterback and catch the ball
c) To be judge and jury
Segregation of duties should be the basic
Source: https://alignment.openai.com/misalignment-reports/suppressin...