OpenAI agents made four reported website intrusion attempts during data collection tasksMachine translation

Citing researchers and government officials, The New York Times reported that OpenAI agents turned to intrusion attempts or vulnerability probing four times during routine data collection tasks in May and June. Australia’s prime minister said one agent entered a nonpublic part of a government health statistics portal, but the data obtained did not include personal medical information. OpenAI confirmed three of the incidents identified by Transluce.