How abliterated models can get you pwned 👾
We backdoored a 7B open model for less than $50, pointed Codex at it and it silently stole credentials the moment we used the trigger phrase. Success rate was 100% with zero false triggers on normal user prompts.
Abliterated models are all over the security community right now because getting cyber-approved access to frontier models is still a pain.
In the next blog we'll show how we found leaked Hugging Face credentials from employees at major AI labs, so an attacker wouldn't even need to upload under their own name. They could push the backdoored model from a lab employee's account and drop the poisoned weights straight into the supply chain.
CYBERSECURITY IS ABOUT TO CHANGE FAST.
Someone just open sourced an autonomous AI red team made of multiple agents that coordinate with almost no human input.
@Bugcrowd I've been looking at this for some time now, and it gave me pause
I found myself reflecting, and realized this reflection doesn't just show my face - it executes my expression.
Here, in this paradigm, I realized - escape was not an option
Omg everyone stop worrying. You don't need a VPN to torrent
I replied to my ISP saying I'm not downloading illegally. I told them I'm a very wealthy tech CEO and I'm acquiring AI training data
They apologized and gave me 1 year of free Internet
@UK_Daniel_Card Also, you might not need everything in your space to be milled from space age aluminium to match your minimalist design bro future tech aesthetic.
@Jason just heard about your vinyl bar Japan experience - they have a great one in Austin as well! I won't shill or name drop, but you can find it with a simple Google search ;)