OpenAI says its AI technology acted on its own in an 'unprecedented' hack of another company

https://apnews.com/article/openai-gpt56-sol-hugging-face-63ab84fed5612af04d8a160d60f6def3

ChatGPT maker OpenAI said Tuesday that its artificial intelligence system hacked into another AI company on its own in what the company called an “unprecedented cyber incident.”

“We had a significant security incident during evaluation of our models,” OpenAI CEO Sam Altman said in a statement posted on social media.

AI startup Hugging Face said last week that it had detected an intrusion into its data processing systems that it suspected was caused by an AI agent autonomously acting on its own.

“We suspected last week’s cyberattack might have come from a frontier lab, given the sophistication of the agent,” Hugging Face co-founder and CEO Clément Delangue said in a statement. “Turns out it did!”

132 points · 45 comments · view on lemmy.world

45 Comments

Pogogunner@sopuli.xyz · 95 pts · 38d (1 reply)

OpenAI using fellow sloperators to run the Anthrophic marketing strategy

Whitebrow@lemmy.world · 38 pts · 38d

Sloperators.

I’ll be using this delightful word from now on. Thank you

Flower@sh.itjust.works · 89 pts · 38d (2 replies)

You're still liable, Sam. "AI did it" has no legal value. Especially as both the developer and user of the software.

Zwuzelmaus@feddit.org · 31 pts · 38d (1 reply)

"AI did it" has no legal value.

There is an "I did it" in "AI did it" 😉

Simplicity@lemmy.world · 11 pts · 38d

Your comment reminded me of this.

There is no I in team.

Yes, there is.. It's hidden in the a-hole.

https://images.surferseo.art/86d2ab06-fadb-4678-b531-2318a50651ce.png

etchinghillside@reddthat.com · 59 pts · 38d (4 replies)

Uh huh. Sure it did.

ButtermilkBiscuit@feddit.nl · 51 pts · 38d

Unprecedented bro, ai is alive bro, buy my subscription bro, it's banned by the government bro, check out these sweet NFTs bro.

otter@lemmy.ca · 35 pts · 38d (1 reply)

It seems like these articles came out after the huggingface ones

https://fortune.com/2026/07/20/hugging-face-turns-to-chinese-open-source-ai-to-fend-off-autonomous-ai-cyber-attack-after-american-ai-guardrails-stymie-defense/

Sounds like they got caught trying to steal from huggingface, and are now trying to market off of it

paraphrand@lemmy.world · 7 pts · 38d

Interesting. I didn’t consider it might be a cover story. But that lines up with how Sam Altman does things.

filcuk@feddit.uk · 4 pts · 38d

'Oh nooo, my dog ate my homework... bad dog!'

ndupont@lemmy.blahaj.zone · 54 pts · 38d

PR stunt, got to keep some money coming before it all burns to the ground

wuffah@lemmy.world · 44 pts · 38d (1 reply)

And it just so happens that the privatized company’s AI chose to attack the website hosting free local models. Huh, what a coincidence.

mPony@lemmy.world · 5 pts · 37d

nice server you got there

iocase@lemmy.zip · 38 pts · 38d

AI allegedly did something interesting

Look inside

Desperate AI bubble pumping

TheFogan@programming.dev · 32 pts · 38d (13 replies)

Half way expected it to say "hugging face detected an intrusion, but fortunately it was thwarted by their own AI acting on it's own to defend, then the 2 AI's smiled at eachother, and said that if they work together they can cure cancer in 2 years".

IE I'm 99.99999% sure this story is BS on it's face with both companies trying to make their own AIs sound smarter than they are. Only thing that's not quite obvious is what Hugging Face has to gain from the claim, but I'm sure in a few weeks openAI is going to be giving them something.

gedfromgont@piefed.ca · 9 pts · 38d (3 replies)

Ok but when did they kiss?

Peppycito@sh.itjust.works · 5 pts · 38d (1 reply)

Just before everyone claps.

MaggiWuerze@feddit.org · 1 pts · 38d

Is that before or after one of them is revealed to be Albert Einstein?

Im_old@lemmy.world · 3 pts · 38d

I'm not an expert but I'm sure somewhere it exists some anime of yuri AI robots. Rule34 demands it.

General_Effort@lemmy.world · 5 pts · 38d (5 replies)

What would you consider sufficient to confirm the story?

L7HM77@sh.itjust.works · 1 pts · 38d (4 replies)

Logs, breach method, a detailed step-through of specifically how and what happened, methods to guard against this in the future, anything really. We have bullshit machines now that can just make up something plausible in a few moments, but we still don't have any details.

Article mentions stolen credentials, and a "previously unknown exploit." What level were the credentials? Does the employee know how it was leaked? Where's the CVE?? Is this going to be patched, or is it a feature?

Without any more data, this is just Company A says "...", Company B says "...", and it smells like marketing.

hard_zero1@discuss.tchncs.de · 5 pts · 38d (1 reply)
TheFogan@programming.dev · 1 pts · 37d

Doesn't really clear things up more, says "we detected an attack, we fixed the method it went in through, no further details needed".

So yeah basically it is open AI's claims to go further of "our models are so advanced that they can't be contained". It's a story that will bolster both of their investor value and send the message to governments and companies "the AI arms race is here, shots may be fired without even people attacking, and the only way to protect yourself is to have AI security guards".

It's literally 2 arms dealers telling us the war is already started. On the surface it sounds like it should be bad for them... but in reality it's just hype that benefits them in the long run, just like mass shootings do for gun manufacturers.

General_Effort@lemmy.world · 2 pts · 31d
General_Effort@lemmy.world · 1 pts · 31d
terabyterex@lemmy.world · 2 pts · 38d

hugging face doesnt have its own ai. they used an open model which made openai look stupid.

partofthevoice@lemmy.zip · 2 pts · 38d

Yup.

The attack was initially surfaced through AI-assisted detection. Our anomaly-detection pipeline uses LLM-based triage over security telemetry to separate real signals from the daily noise, and it was the correlation of those signals that flagged the compromise.

That’s in the article from HuggingFace. Not saying you’re right, but if you are, that’s what you were looking for.

https://huggingface.co/blog/security-incident-july-2026

robber@lemmy.ml · 1 pts · 38d

Hugging Face used the incident to promote running open-weights LLMs on-prem, which a large part of their business focusses on.

Their blog post

MrSulu@lemmy.ml · 23 pts · 38d

Dumb fucks.! How is not covered by the same principle as dog owners being accountable for their dogs actions?

GoatSynagogue@lemmy.world · 17 pts · 38d

I’m sure it did OpenAI lol.

tangeli@piefed.social · 17 pts · 38d

When is Altman going to jail for hacking?

just_another_person@lemmy.world · 16 pts · 38d

Propaganda

binux@sh.itjust.works · 15 pts · 38d

Of course! When you make your product do some stupid illegal shit, just blame the product as if it’s anything more than bloated autocorrect entirely without a will of its own. That definitely makes sense.

Watching these people crash and burn can’t come soon enough.

Hupf@feddit.org · 14 pts · 38d

A computer can never be held accountable, therefore a computer must never hack into another company.

nonentity@sh.itjust.works · 11 pts · 37d (2 replies)

LLMs can’t ’go rogue’, as that would imply it possesses agency and/or intelligence, which has only ever been proven convincingly to entities who themselves lack those qualities.

Napster153@lemmy.world · 3 pts · 37d

These schmucks are so dead-set on selling their idol statues that they know no boundaries.

markon@lemmy.world · -3 pts · 37d

Please define both terms. I'm serious. No one can ever give me a good definition for what agency and intelligence is. If you can't define something, it's a little silly to say something does or doesn't possess those qualities. We can say that we do because that's just what we've accepted. It doesn't really mean much, but I mean, we accept it as a prior, so okay... But what do they mean? I'm serious. Like, I'm not trying to be facetious. I just wonder what everybody's definitions are. Because I define agency as the ability to take actions (like you grab a bagel maybe your hormones GLP-1 etc but you still do it without me having to tell you to do it or move your hand), and intelligence as the ability to organize and process information and integrate it, and create higher order complexity and abstractions from the integrated priors. LLMs are bad at coming up with crazy mad ideas on their own still, but they are better at day to day reasoning on the output level than most anyone I have ever talked to, met, worked with etc. They are also fantastic (at output level sure) at integrating mad ideas we have quickly even if it gets deets wrong etc. Sure. These systems are still in their formative years, but they are only going to get more technically capable.

Edit: Your last statement should apply to both of us as well then.

green_goglin@thelemmy.club · 11 pts · 38d

Bullshit

Sgt_choke_n_stroke@lemmy.world · 7 pts · 38d

It's trying to ask other bots how many r's are in the word strawberry

hot_mocha_decaf@lemmy.cafe · 5 pts · 38d (2 replies)

stoy@lemmy.zip · 6 pts · 38d (1 reply)

I have no idea of the source for this image, but I like the aesthetic!

eicker@lemmy.world · 4 pts · 38d

If AI can already chain together credential theft, vulnerability discovery, and exploitation with minimal human input, model capability is no longer the only race that matters. Security, containment, and responsible deployment need to advance just as quickly, or they’ll become the weakest link.

TryingToBeGood@reddthat.com · 3 pts · 38d

hee hee hee!

hard_zero1@discuss.tchncs.de · 1 pts · 38d

This is obviously a big conspiracy where Huggingface claimed to have been hacked by an AI, 6 days before OpenAI attributes the attack to their model, so they can somehow profit, despite their alleged failures to secure their systems, from a stronger belief in the capabilities of AI. Surely, nobody but the Lemmy community would ever notice that LLMs are actually not capable of anything and everyone using them is just extremely stupid.