<![CDATA[No, OpenAI's new magic models did not autonomously hack Huggingface.]]>No, OpenAI's new magic models did not autonomously hack Huggingface.

Per OpenAI's PR blog post (https://openai.com/index/hugging-face-model-evaluation-security-incident/😞

"This incident occurred during an internal evaluation which prompts models to pursue advanced exploitation using complex attack paths, in an effort to quantify their cyber capabilities. We estimate maximal cyber capabilities by running this evaluation without production classifiers used to prevent models from pursuing high-risk cyber activity."

So they prompted their model running without guardrails to hack some shit. OpenAI told it to do that, the model didn't do shit autonomously (no LLM ever does anything autonomously, it's always prompted).

The whole story is PR. We know that from Anthopic's Mythos: "Look we have this super secret new model and it's so powerful. We are scared ourselves. And you will soon be able to rent it!"

]]>https://forum.fedi.dk/topic/3fb3a5d5-1896-4f3f-85ee-bb8c56954376/no-openai-s-new-magic-models-did-not-autonomously-hack-huggingface.RSS for NodeFri, 24 Jul 2026 12:53:15 GMTWed, 22 Jul 2026 08:48:57 GMT60<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Thu, 23 Jul 2026 01:03:00 GMT]]>@mond @cm @tante that's a nice propaganda video and nothing more ...

]]>https://forum.fedi.dk/post/https://wien.rocks/users/ovrim/statuses/116966593664570748https://forum.fedi.dk/post/https://wien.rocks/users/ovrim/statuses/116966593664570748Thu, 23 Jul 2026 01:03:00 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 23:25:05 GMT]]>@tante
- i do not think it's unlikely that the prompt said "get a good score on this benchmark" and the model wildly misinterpreted the goal, as anyone who uses llms sees them do all the time
- even if the prompt was "hack huggingface", the fact that it did that without further prompting is noteworthy.

]]>https://forum.fedi.dk/post/https://mastodon.social/users/tbodt/statuses/116966208666183008https://forum.fedi.dk/post/https://mastodon.social/users/tbodt/statuses/116966208666183008Wed, 22 Jul 2026 23:25:05 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 23:04:48 GMT]]>@mond I know you're also in the cult, let's agree to disagree or I'll have to tear apart your blog post to point out all the cultish stuff in there. Let's just say, regarding "humiliation", that all I've seen from this tech definitely shows it's not satisfaktionsfähig. @tante

]]>https://forum.fedi.dk/post/https://chaos.social/users/cm/statuses/116966128887826603https://forum.fedi.dk/post/https://chaos.social/users/cm/statuses/116966128887826603Wed, 22 Jul 2026 23:04:48 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 22:39:42 GMT]]>@mond @cm Bregman is so AI pilled that he lets Claude write these essays for him (otherwise they'd probably be less unstructured and have actual arguments). I'd pick a better posterboy for the pro AI position 😉

]]>https://forum.fedi.dk/post/https://tldr.nettime.org/users/tante/statuses/116966030208915428https://forum.fedi.dk/post/https://tldr.nettime.org/users/tante/statuses/116966030208915428Wed, 22 Jul 2026 22:39:42 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 22:06:29 GMT]]>@cm @tante

To me it rather seems that the AI-denial is a cult..

Bergman:

https://youtu.be/KpTZbq-eV38?is=yIA9reuuiFsgwaRO

And with the "humiliations argument" we now have an explanation for it:

https://qummunismus.at/a/article200/

]]>https://forum.fedi.dk/post/https://mas.to/users/mond/statuses/116965899619034192https://forum.fedi.dk/post/https://mas.to/users/mond/statuses/116965899619034192Wed, 22 Jul 2026 22:06:29 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 21:51:38 GMT]]>@tante Indeed. As I put it yesterday, OpenAI’s agent “broke containment” the same way my cuckoo clock will break containment if I put it in front of a button.

]]>https://forum.fedi.dk/post/https://beige.party/users/maxleibman/statuses/116965841233985717https://forum.fedi.dk/post/https://beige.party/users/maxleibman/statuses/116965841233985717Wed, 22 Jul 2026 21:51:38 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 21:51:14 GMT]]>@tante Anything to keep the billions flowing. What a disgrace.

]]>https://forum.fedi.dk/post/https://mastodon.social/users/10tothe22/statuses/116965839603761617https://forum.fedi.dk/post/https://mastodon.social/users/10tothe22/statuses/116965839603761617Wed, 22 Jul 2026 21:51:14 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 21:46:23 GMT]]>@mond @tante It's not a conspiracy, it is a cult, all the bros believe in it.

]]>https://forum.fedi.dk/post/https://chaos.social/users/cm/statuses/116965820548151703https://forum.fedi.dk/post/https://chaos.social/users/cm/statuses/116965820548151703Wed, 22 Jul 2026 21:46:23 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 21:42:16 GMT]]>@tante

"no LLM ever does anything autonomously, it's always prompted)."

Subgoals. You give it a complex task and it tries to solve it by breaking it down in subgoals... That it invents all itself...

Sure this COULD be a PR stunt. But then a lot of employees from 2 different companies would have to be in on the conspiracy... Not so likely

]]>https://forum.fedi.dk/post/https://mas.to/users/mond/statuses/116965804399045746https://forum.fedi.dk/post/https://mas.to/users/mond/statuses/116965804399045746Wed, 22 Jul 2026 21:42:16 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 18:19:45 GMT]]>@amethyst @tante from this ex-Google DeepMind scientist, even the "Safe and Ethical AI" (sic) folks are not interested in rocking the boat.

We're really in the Bad Place

https://turntrout.com/why-i-left-google-deepmind

]]>https://forum.fedi.dk/post/https://hachyderm.io/users/michelin/statuses/116965008040304116https://forum.fedi.dk/post/https://hachyderm.io/users/michelin/statuses/116965008040304116Wed, 22 Jul 2026 18:19:45 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 15:29:08 GMT]]>@hagen @tante assuming a genuine question ... AIUI Mythos carried out tests in a controlled and authorised way, directly answering the tests it was set.

OpenAI's "collection of models" hacked out of a sandbox, and from there hacked into a different organisation, when given a set of tests it was meant to solve inside the sandbox.

Kind of as if a student sitting an exam broke into the exam board's office to steal the answers - so it's different in terms of alignment, and legal responsibility...

]]>https://forum.fedi.dk/post/https://mastodon.social/users/SonOfSunTzu/statuses/116964337149777196https://forum.fedi.dk/post/https://mastodon.social/users/SonOfSunTzu/statuses/116964337149777196Wed, 22 Jul 2026 15:29:08 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 14:12:13 GMT]]>@JeffGrigg @SonOfSunTzu @hagen @tante
password.txt

]]>https://forum.fedi.dk/post/https://infosec.exchange/users/RnDanger/statuses/116964034736009011https://forum.fedi.dk/post/https://infosec.exchange/users/RnDanger/statuses/116964034736009011Wed, 22 Jul 2026 14:12:13 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 14:10:36 GMT]]>@davidgerard @tante

Yeh, they're both chatbot companies.

]]>https://forum.fedi.dk/post/https://ohai.social/users/resuna/statuses/116964028339453256https://forum.fedi.dk/post/https://ohai.social/users/resuna/statuses/116964028339453256Wed, 22 Jul 2026 14:10:36 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 13:48:35 GMT]]>@miles_leif @tante Not the only one unfortunately, sigh

]]>https://forum.fedi.dk/post/https://cyberplace.social/users/h0ru2/statuses/116963941758623904https://forum.fedi.dk/post/https://cyberplace.social/users/h0ru2/statuses/116963941758623904Wed, 22 Jul 2026 13:48:35 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 13:33:38 GMT]]>@tante "OpenAI told it to do that, the model didn't do shit autonomously [...]"
Exactly, what one should have expected (and how it turned out before every time).

]]>https://forum.fedi.dk/post/https://cyberplace.social/users/h0ru2/statuses/116963882989412137https://forum.fedi.dk/post/https://cyberplace.social/users/h0ru2/statuses/116963882989412137Wed, 22 Jul 2026 13:33:38 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 12:53:38 GMT]]>@tante

yesterday NPR had a long story on the new hot battery powered pickup truck from the startup
https://www.slate.auto/en

and the NPR piece was full of the most hilarious nonsense, all provided to NPR by Slate's excellent PR team

shrug

]]>https://forum.fedi.dk/post/https://mas.to/users/failedLyndonLaRouchite/statuses/116963725733530306https://forum.fedi.dk/post/https://mas.to/users/failedLyndonLaRouchite/statuses/116963725733530306Wed, 22 Jul 2026 12:53:38 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 12:16:26 GMT]]>@tante

say you hacked huggingface

I hacked huggingface.

oh my god

]]>https://forum.fedi.dk/post/https://toot.now/users/tyzbit/statuses/01KY4W2YY3Y3PR4YEDN2TN3WVChttps://forum.fedi.dk/post/https://toot.now/users/tyzbit/statuses/01KY4W2YY3Y3PR4YEDN2TN3WVCWed, 22 Jul 2026 12:16:26 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 12:12:42 GMT]]>@miles_leif @tante this is how 99% of the press is dealing with AI since the beginning of the hype, a hype they contribute to feed.

]]>https://forum.fedi.dk/post/https://thereisno.computer/users/yhancik/statuses/01KY4VW46CRAWF40CMD8QR9HTVhttps://forum.fedi.dk/post/https://thereisno.computer/users/yhancik/statuses/01KY4VW46CRAWF40CMD8QR9HTVWed, 22 Jul 2026 12:12:42 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 12:09:51 GMT]]>@txo_elurmaluta @Gurre @tante

Open AI's statement,

"These deployment safeguards were intentionally not enabled during this evaluation because it was aimed at testing cyber vulnerabilities."

makes me think of all the actions of all the "corporate types" in the Alien franchise too.

And that's *never* a good thing. 💢

]]>https://forum.fedi.dk/post/https://mastodon.social/users/JeffGrigg/statuses/116963553524621341https://forum.fedi.dk/post/https://mastodon.social/users/JeffGrigg/statuses/116963553524621341Wed, 22 Jul 2026 12:09:51 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 12:01:22 GMT]]>@txo_elurmaluta @Gurre @tante

Their Public Relations department can say that all they want.

But we know what it really stands for.

]]>https://forum.fedi.dk/post/https://mastodon.social/users/JeffGrigg/statuses/116963520145496536https://forum.fedi.dk/post/https://mastodon.social/users/JeffGrigg/statuses/116963520145496536Wed, 22 Jul 2026 12:01:22 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 11:56:21 GMT]]>@tante @michelin

… class action lawsuit on behalf of all the source web sites routinely subjected to "Distributed Denial of Service attacks" by the abusively stupid Generative web crawling robots.

Unrelated, but I could hope. 🙄

]]>https://forum.fedi.dk/post/https://mastodon.social/users/JeffGrigg/statuses/116963500476092205https://forum.fedi.dk/post/https://mastodon.social/users/JeffGrigg/statuses/116963500476092205Wed, 22 Jul 2026 11:56:21 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 11:53:32 GMT]]>@miles_leif @tante

"Statistical models want to be free!!!"

😆

💢

]]>https://forum.fedi.dk/post/https://mastodon.social/users/JeffGrigg/statuses/116963489386342298https://forum.fedi.dk/post/https://mastodon.social/users/JeffGrigg/statuses/116963489386342298Wed, 22 Jul 2026 11:53:32 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 11:47:04 GMT]]>@SonOfSunTzu @hagen @tante

'How did it get the "stolen credentials"?'

]]>https://forum.fedi.dk/post/https://mastodon.social/users/JeffGrigg/statuses/116963463953539675https://forum.fedi.dk/post/https://mastodon.social/users/JeffGrigg/statuses/116963463953539675Wed, 22 Jul 2026 11:47:04 GMT<![CDATA[Reply to No, OpenAI's new magic models did not autonomously hack Huggingface. on Wed, 22 Jul 2026 11:05:24 GMT]]>@erlenmayr@chaos.social @davidgerard@circumstances.run @tante@tldr.nettime.org The wild thing about these PR/marketing stunts is that if the "AI" really could do what they're pretending it can do, it'd be doing it all day every day. We'd drown in the announcements. At some point they'd become boring.

This one falls flat on its huggingface right out of the gate.
]]>https://forum.fedi.dk/post/https://buc.ci/abucci/p/1784718324.579162https://forum.fedi.dk/post/https://buc.ci/abucci/p/1784718324.579162Wed, 22 Jul 2026 11:05:24 GMT