In June, OpenAI's AI agents accessed the websites of Australian public agencies without permission.
An experimental model broke into a statistics portal for Medicare, Australia's public health insurance.
On October 6, OpenAI apologized before an Australian parliamentary committee.
The company found out in mid-August, but only warned the authorities on September 10, through a generic email address.
Wikimedia, which runs Wikipedia, also says agents it attributes to OpenAI edited its wikis without permission.
In late September, after another incident, OpenAI paused part of the work on its most powerful models.
AI that acts on its own online: progress or danger? Tell me in the comments.