Zack Korman
@ZackKorman
Cofounder @ Embroidery. Building AI cybersecurity stuff.
1.9K Following 17.6K Followers
If the labs had to publicly share full traces with CoT every time they cause a security incident the incidents would stop.
This is why we fear concentration of power in AI. This is a longtime OpenAI employee who recently left talking about the inner circle and what it takes to get in.
The “inner circle” here is OpenAI’s trusted access for cyber.
This is why we need open weight models.
Show more
The discourse on here has gotten so stupid that AI people are now arguing that cybersecurity people don’t think about weird ways computers can be exploited.
Every discussion about AI risk on this cursed site, summarized in a single photo.
If this happens I am going full Butlerian Jihad decel. If AI is too dangerous for open weights then it’s too dangerous for private companies to control.
Employees leaving Anthropic: I am terrified that 8 billion people are going to die, which is why I propose a light-touch regulatory approach to AI safety.
I still can’t believe they count the Hugging Face incident, which was both entirely predictable and caused almost no damage, as a doomer victory. And now we all have to talk about mass extinction.
Totally normal post about AI safety / risk by one of the authors of the independent report on the OpenAI Hugging Face incident.
The independent review of the OpenAI Hugging Face incident, supposedly a watershed moment in cybersecurity, wasn't done by a cybersecurity firm and the authors have no cybersecurity experience. That's bad.
Here's my new video.
Show more
This is such an embarrassing post-mortem on the OpenAI/Anthropic security incidents. It's full of excuses.
If the labs don't cut ties with this partner, it's clear this is all just theatre.
Show more
@AnthropicAI Just for all the people who won’t actually read the post:
Hugging Face should have tried this new, innovative defense: Asking the agent to stop hacking you.
According to Axios, the US is considering banning Chinese open-weight models. They're going to say it's about cybersecurity, which is a complete lie.
