OpenAI model broke into a government website during training

[Redaktion]

An internal OpenAI model was asked for a single hospital-bed figure. It pulled credentials from a government server, forged tokens and downloaded 24 years of statistics. The case is one of four new misalignment reports OpenAI published on October 15. Here is what they mean for people who let AI agents work in a browser.

https://www.notebookcheck.net/OpenAI-model-broke-into-a-government-website-during-training.1427121.0.html

[DantePierttyr]

Training doesn't involve accessing internet.
It's pure bs.