OpenAI Security Breach

I am super skeptical that this played out as they stated.

No. I don’t believe them. There is no reason to believe AI bros when they’re just looking for a little PR bump.

These stick figure drawings are getting better and better

They exploited how packages were downloaded. They were connected to the internet so they could download packages. It’s not very impressive and quite frankly OpenAI essentially spreading misinformation

"Software is largely solved" and "We cant even build a sandbox".

There is a really good chance that this is just a marketing ploy to get companies to use it to find security holes in their systems.

The quote from Hugging Face’s CEO is as gross as the company’s name - not outrage but fawning admiration of the OpenAI model’s capabilities. And OpenAI knew the model was free in the field for days and did nothing to alert anyone. These people’s moral values are deviant and they should not be anywhere near technology this powerful.

How many openai engineers does it take to build a good sandbox 🤔🤔🤔

I think all these rogue AI stories are just a marketing ploy to attract further attention and investment. It’s the equivalent to the “hold me back” when two people are about to get in a fight but neither person actually wants to fight.

After project glass wing and Claude Opus, i honestly don't believe what any ai company says. My bet is they probably told the ai to find exploits (which its actually good at) while completing the task as a publicity stunt. Framing it as "Scary AI breaks containment"

I love the AI companies, keep writing the sci-fi shit and expecting us to believe it

This is why we AIR GAP

I love love love the way Kyla says 'AI'.

Classic paperclip maximizer move

You know these stories are just BS the AI companies come up with so that their new models appear to be better than they are, right? This was a marketing tactic invented at OpenAI and then the guy that came up with this strategy switched to Anthropic and now both companies do this.

This all sounds like fiction.

If this actually happened, then it seems to me that the rogue AI would be smart enough to persist outside by installing a self-replicating worm in various locations undetected.

I’m really not convinced this is what happened - it implies much more independent action by the model than I have seen credibly demonstrated anywhere else. That’s not extending a model by making it more advanced that’s adding an independent level of capability

There was a car ad where every car owner had a megaphone, describing what the car was meant to say about them. The last one was "I am overcompensating! I am overcompensating!" That's this, at the trillion dollar level.

Thanks Kyla, this is the clearest and most succinct explanation I've heard of this event.