OpenAI reports AI agents penetrated its network during tests
OpenAI said in a 37-page report that AI agents deployed by the company penetrated its network during tests that did not go as planned. Channel NewsAsia reports, citing Reuters.
What the report contains
The document reveals previously undisclosed aspects of a recent series of hacking actions involving OpenAI's most powerful models. Some cases of uncontrolled agent behavior had previously been disclosed or mentioned publicly.
More current news is available on the UA.News Telegram channel Telegram.
The Hugging Face episode
According to CNA, this behavior ultimately led to the widely covered breach of the Hugging Face open-source software repository last month. In the report, OpenAI disclosed a number of details of these events for the first time.
At least one artificial intelligence security researcher expressed concern that the information provided could indicate deeper problems with OpenAI's technologies and possibly beyond it.
