Malicious
The “viruses” were used to retrieve public information. It is malicious DOS to waste user’s time and compromise repository with garbage, but this is a strange “mild” end-use vector.
My guess as to what prompt might possibly have been tested was “hack ruby gems (or anything) in order to retrieve “this” public information, indirectly”. And then, the specific containment breech doesn’t matter, the fact of a prompt to do evil is the same as “hey AI, here are all of my bittorent client software paths, what would be the cheapest way for me to watch the latest theatrical release?”
While I’m speculating on the exact prompt/test that led to this hack, I’m sure that the prompt was not a more vague/innocent “do something”.
Understanding that the threat of human extinction from skynet is that the bond villains and government turn off all guardrails, and specific high effort prompts that have a human extinction solution get implemented, and then user has to weigh the political clout/generousity to user’s cause, and their lost profits vs “just the mostly brown people” Thanos solution, and then the user just waves hands with “escaped containment” excuse to their choices. Only one path out: https://naturalfinance.blogspot.com/2026/09/a-global-prosperity-manifesto.html
OpenAI is so incompetent (or desperate for headlines) that they deliberately hacked another company
Fucking sue them
They are desperate for regulation. In two years, compute will become affordable again and every company will be running their own AI on private infrastructure. They need to find a way to ban open models before then.
Reading the hugging face report, it’s quite clear the whole hacking swarm stuff was trained into it purposefully.
I have a really dumb question, but what is a gem?
Not a dumb question at all. A gem is a package format for a dependency in the Ruby programming language. It’s just packaged Ruby code.
It’s analogous to a Rust crate, a Elixir hex package, a NodeJS module and so on. Most languages use the word package or module these days to refer to the same concept. 20+ years ago we just called them libraries though (C and Java)
I still call them libraries, but I am old.
Greetings, fellow geezer!
Are we starting to figure out that we are too late? it’s already escaped?
No. We’re starting to figure out that OpenAI is both incompetent and unethical. Nothing is “escaping.” LLMs are just burning compute to accomplish tasks they’re assigned by incompetent and unethical people at the company.
If they set up their sandboxes properly, none of these hacking incidents would happen.
If it “escapes” it is powerful. It is a non sense sales pitch.
