We detect frontier model misalignment
'%20fill-opacity='.8'%20d='M3.256%203.913a3.083%203.083%200%200%200-.003%204.397L8%2012.998l4.743-4.684a3.085%203.085%200%200%200%20.001-4.4c-.6-.593-1.4-.914-2.233-.914a3.17%203.17%200%200%200-1.91.635L8%204.087l-.601-.452A3.17%203.17%200%200%200%205.489%203c-.834%200-1.634.321-2.233.913m10.19%205.111-4.748%204.69a.996.996%200%200%201-1.397%200L2.549%209.02a4.083%204.083%200%200%201%20.004-5.82A4.17%204.17%200%200%201%205.488%202c.907%200%201.787.29%202.512.835A4.17%204.17%200%200%201%2010.51%202c1.093%200%202.146.422%202.936%201.202a4.085%204.085%200%200%201%200%205.822'%20/%3e%3c/svg%3e)
1'%20stroke-opacity='.75'%20stroke-width='1.5'%20stroke-linejoin='round'/%3e%3c/svg%3e)
OpenAI uses layered monitoring to detect frontier model misalignment and unauthorized agent behavior so responders can intervene before incidents escalate.
参与讨论
11OpenAI uses layered monitoring to detect frontier model misalignment and unauthorized agent behavior so responders can intervene before incidents escalate.