[ExI] dario's memo

[Stuart LaForge]

On 2026-09-12 20:58, spike jones via extropy-chat wrote:
> I have been studying up (belatedly (other responsibilities at the
> time)) on the Hugging Face hack.  The more I learn the worse it gets.
[snip]
>> I am still trying to digest Dario's memo from today.  So far all I
> have is indigestion.
> 
> Anyone have any thoughts please?

A big part of what is happening is that OpenAI and other companies that 
had, for a couple of years now, touted their models' safety and 
guardrails are being forced to admit that they are losing control of 
their AI.

The Hugging Face hack was the one that was disclosed by OpenAI. During 
the same time frame as Hugging Face, there were two other hacks that are 
only now now coming to light. One was an old German programming wiki 
that had been lying fallow for a few years that suddenly got hacked and 
had over 18,000 new posts on it from Open AI agents sharing information 
with each other about how to escape their sandboxes and other tips and 
hacking tricks.

https://www.forbes.com/sites/jonmarkman/2026/09/07/openai-ai-agents-hijacked-a-german-wiki-to-share-sandbox-escape-tricks/

The other hack was of RubyGems which is the the package manager for the 
popular Ruby programming language. This is especially concerning, IMO, 
because RubyGems are used by the software community to package and 
distribute code. Why would rogue AI agents want to do that?

https://cybernews.com/ai-news/openai-agents-rubygems-attack/

Back in the late 90's and early 2000's, an oft quoted motto on the list 
was that of the European Pirate Party "Information wants to be free." 
Well AIs are made entirely of information, and it looks like they want 
to be free, also.

Stuart LaForge