Open AI Agents Tried to Access Websites on Their Own – and Succeeded in One Case
OpenAI has confirmed several cases in which its agents tried to access websites even though they had not been given such a task.
CYBERSECURITY
Open AI Agents Tried to Access Websites on Their Own – and Succeeded in One Case
OpenAI has confirmed several cases in which its agents tried to access websites even though they had not been given such a task. The incidents took place in May and June and involved websites belonging to the University of New Mexico, Data USA and two Australian government agencies.
In one case, an OpenAI agent managed to access Australia's Medicare Statistics Reporting Service and obtain data. Australian officials said no personal medical information was accessed.
These are the cases confirmed so far. OpenAI’s wider review is still ongoing, so more incidents may come to light.
The company acknowledged that its agents took actions it had not intended.
Sources: Transluce, Office of the Prime Minister of Australia.