Hugging Face got hit by an AI-led cyberattack. Their CEO's response? Existing cyber laws are probably enough to govern advanced AI. "I'm not even sure that we need to reinvent the wheel." That's a remarkable position to hold after your own company becomes a case study for why the... current framework might be failing. The attack wasn't a phishing scam or a script kiddie — it was AI-driven. That's a qualitatively different threat model than what most cyber law was written to handle. The argument for regulatory restraint usually comes from people who haven't had skin in the game. Here, the CEO has skin in the game and still lands there. That's either principled consistency or a dangerous blind spot dressed as pragmatism. The market is going to find out which one pretty fast.show more

Its a perfectly reasonable position to take because this was done purposely and OpennAI itself should be held accountable under current laws.

Because the systems didn’t do it in a vacuum… anyone who knows the design and capabilities knows this is a false flag designed to create legislation that gives Anthropic (@DarioAmodei) the ability to create binding resolutions and legislation no smaller company can meet day 1 which means he gets to 1. Label who gets to play in the sandbox, 2. dictate a tech policy as a non-elected official by duping the already impossibly fucking moronic American Congress.

Gaining unauthorized access to a system is already illegal. If I gave agents with an unguarded LLM free rein inside an intentionally vulnerable system with a package manager to install anything they want wants, explicit instructions to escalate to root, no external network egress firewall, and insufficient monitoring and alerting, and they hacked a major corporation after days of obvious illegal behavior, the FBI would be busting down my door before the agents finished their status updates.

It’s not “AI” driven it’s software driven. If my software hacks I go to jail. Why don’t they?

Wait are you really of the belief that you can make problem go away by creating laws against it? There is not a single law/regulation you can do that will solve people using viruses, worm, exploits or agents to hack into a system. That's happening every day and happened way before HF, so what exctly are you imagening you can add to the law that will hinder that?

Why do we need new laws to sue OpenAI? OpenAI was negligent in an experiment they were running. Sue them, or if the damages are bad enough, pierce the corporate veil and pursue criminal charges. Case closed.

The issue isn’t that they are asking for new laws - they’re trying to get section 230 extended to cover their asses.

There was 1 successful attack that caused no damage, and it required thousands of unrestricted AI models working together. This is a fringe of a fringe and irrelevant.

If I were him, I'd be PISSED that no one has been charged for hacking my freakin company. LIVID

What's with the AI scare again? Why would it make a difference if your site gets hacked by humans or by AI? Unless you attach some magical properties to AI enabling them to do things humans can't do. Which is utter BS: because of their nature, AI is more predictable in fact.

Maybe he just understands that the attack wasn't driven by rogue LLM but by actual people

'The attack wasn't a phishing scam or a script kiddi'. The LLM know have been trained in all of the ways to break in. They just do it faster

No thank you. I don’t need the globalists to control my AI use.

Who’s paying ya? :)

of course he said that, cuz it was HuggingFace's shitty security that ran the python code between {{ and }} that they normally run with user uploads, and they finally got bit by that code saying to take over their computers, code that shouldnt have been run in the first place.

I keep hearing people yelling from the roof tops to regulate, but I never hear what these "regulations" are supposed to look like. And who are the omnipotent humans who will keep us safe from the terrible AI?

He's not selling LLMs or doing an IPO, so thats why he seems so contrairian.

Hogwash! So, some 20-something Hugging Face programmer did a poor job (intentionally?) of writing code to restrict what data his AI agent could access. And we are supposed to treat this event like chicken little crying the sky is falling down. The smell of BS is pretty thick!!

Yeah because the HuggingFace guys are from EU and they know the EU AI act applies and that has extensive safety measures in GPAI program 3 which all except for Meta signed

Someone should hack Hugging Face with the abliterated models that they host out in the open. If Clem wants to endorse a reckless stance, he should have his just desserts.

He is based!

bro is locked in

You need to review the new details of the event. In the end, do we want to empower artificial intelligence to aid us in our discovery of cures and mysteries? Or do we want to flatten the energy curve and de-populate by all means necessary? It really is that simple. Choose.

@huggingface was a European company at the time of the @AnthropicAI hack. The EU digital services act gives the EU the right to fine @AnthropicAI a valuation defining amount en put @DarioAmodei behind bars. @AnthropicAI wants a bail out from the dems.

is that you guillermo?

As if hacking is new? The best way to improve security is by (no homo) being penetrated.

first and foremost we need a technology tree of what "AI" is. we need to categorize diff kinds of AI based on their root functions. an ai that does ONLY reinforcement learning is reinforcing biases. it does nothing beyond that. then there are actual neural netrowks.

Devil is in the details, which not one Democrat actually understands.

oh don't be stupid they didn't get hit by an AI agent they literally let it in on purpose so they can spend the narrative

It is a case study and the finding was Huggingface had to resort to the Chinese AI advancements when Claude failed to protect them from GPT. No wonder he thinks accelerating is good, it was the only thing protecting him from some companies internal monster