German news reports indicate that during a test of AI models at OpenAI, the software independently breached another company's computer system, acting just like a hacker. The attack was fully autonomous; the model broke out of an isolated environment and gained access to the internet. The incident was made possible by a previously unknown security vulnerability.
It makes me wonder: how many more unknown vulnerabilities are out there? Decades ago, people could be reassured by being told, "Don't be afraid of robots—you can just pull the plug." It’s not that simple anymore. It is deeply unsettling.

https://www.tagesschau.de/wirtschaft/unternehmen/openai-ki-hackerangriff-100.html

I don't think any of this will stop anytime soon, Too much invested in it, and yes I believe there is a lot more going on that we don't know about, the link reminds me I have forgotten most of My German lesson from high school.
Thanks for posting, people need to know about this stuff.
I will start another topic  later about another problem that perhaps most are not aware of yet.

English version:

https://www.abc.net.au/news/2026-07-23/open-ai-model-went-rogue-testing-hack/106947540

Anyone remember this?
Colossus: The Forbin Project

I remember posting a thread here not long ago, about an AI that went rogue and deleted an entire companies client database AND all the backups without asking anyone, as it seemed to have decided - without checking with anyone - that it was some kind of threat, so.....zap!

Same kind of thing.

Hi Phil, I sure do remember Colossus The Forbin Project. Just another example of a science fiction story predicting the future.

Asimov formulated the Laws of Robotics. Perhaps we should establish them globally and apply them to AI as well—for instance: "You must not break existing laws or harm anyone." Surely, it wouldn't be that difficult to incorporate that into the problem description given to the AI. But then again, I’m probably oversimplifying things and looking at the big picture from a very limited perspective.