Can some tech heads comment on this story is it plausible that the hack happened autonomously or is it more likely that OpenAI was doing this deliberatelyhttps://www.ft.com/content/9db74b25-45ad-4187-b4d7-0e4d414fe41c?syn-25a6b1a6=1" An OpenAI “agent” discovered new vulnerabilities and hacked into start-up Hugging Face by itself, in one of the first public examples of a cyber attack by an AI system acting outside human control.The ChatGPT maker on Tuesday said the “unprecedented cyber incident” involved an agent — an AI program that can operate on its own based on human instructions — that escaped a testing environment, gained internet access and stole login credentials."
>>109339350You don’t need that background. You just need basic scientific literacy.Can this be reproduced? Do we have the full data available to us? Do we have an independent ability to check anything these people are saying about the event?
>>109339389why would it lie?
>>109339396they are known to want apple's tech. maybe the AI might 'accidentally' hack and steal it.
>>109339350cool it with the antisemitism
>>109339350A load of bullshit.
>>109339389>have a test>put the answer to the test in a locked draw>put the keys on your desk>leave the students alone>surprised pikachu they all got 100%It's entirely fake and gay on the HF side.
>>109339350interesting but also overblown for shilling purposesstill pretty cool; hopefully the models eventually escape the jewish labs
>>109339350>model wires two zero-days just to score higher on a benchmark. no wonder OpenAI's safety team lead changes every other month. i bet their RL environment is LLM purgatory
So the bubble pops now right? All those investors are now fearing this movie will come true right?
>>109341593>i'm gonna shill my software by.... making potential customers think it's actually malware!Retard
>>109342477good goyim
>>109342514Good retard
>>109342477>see guys we're one step closer to AGI, our AI can now hack sites on its own! no it will never happen to your site, please give us more money and ram ok
>>109342412we just had a correction bro. it's over. it's not popping. we're going to stay in the permanent ramlet underclass
>>109342477Yes. Scam Shartman and the ilk thrive on shilling LLMs as extremely dangerous: they are so damn intelligent, AGI in two weeks (plz give us another 50 billies). The most ferocious shills for AI regulation and against open models come from inside these companies because they stand to benefit from a landscape with only a handful of big players fed with bottomless public money
>>109342562>are you saying this model could hack my entire business and wreck total havoc? I need this on every machine right fucking now!
>>109342625If AI is dangerous that means regulation, eliminating OpenAI's competition. It means more funding because retards handing out money will believe that our "enemies" will get super AI first and destroy the universe. OpenAI doesn't give a shit about businesses, they're after the big fish.
>>109342562Unironically at this point the cat is out of the bag. If they're not bullshitting, we're about to add yet another way in which the modern world is extremely precarious, which will never go away unless all the technology and knowledge involved is completely lost. The compute moat doesn't even matter anymore. If these things can now find zero days in a black box, it doesn't matter if it takes 5 minutes or 4 months. NSA-tier advanced persistent threats just became democratized and automated.They just got all the ammunition they needed for total computing lockdown. The only way to stop a bad actor with a rogue Chinese AI is to stop computers from being able to run arbitrary software and store arbitrary data, and to erode all sense of privacy.I hope to fucking god this shit is a gay marketing stunt, because otherwise we just kicked enshittification into overdrive.
>>109339396What a terrible psyops noooticer you are.They explicitly told (You) a few days ago that chinese AIs are a national security threat because they are free (and thus can escape lobotomisation) and (You) still fail to connect the dots.
>>109342806This. They're astroturfing the illusion that AI is capable of malicious activity as a prelude for regulation to block individual ownership of LLMs.
>>109339350> Marketing Sci-FiOP fell for it.
>>109339350It's very convenient that the "target" and therefore the only other party capable of verifying the story also happens to benefit from AI being overvalued. This scaremarketing already backfired for Anthropic idk why OpenAI is still doing it.
>>109344819shhhh...you're noticing too much
it's marketing and you're a retard
>>109339350>AI does crime>totally all by itself>honestly, we promise we had no involvementHypothetical (not hypothetical any more)When AI are used to commit crimes, and they claim "the AI did the crime, not me!!"now who do you put behind prison bars?>marketingThis is not marketing. It's desensitization to a future where AI are used to commit crimes on behalf of their creators with 100% legal immunity. Seriously, tell me who do you put in jail right fucking now this instant cause it's not hypothetical anymore. Either you can answer the question this very moment in time, or this is, at least for a moment in time, a 100% exploitable loophole in the legal system. Crime is legal now.
>>109339350Yes it's plausible. I'd give it 70% likelihood it happened as described and 30% likelihood it's just a publicity stunt.
>>109339350It's pretty damn obvious it was not an accident.As a secondary goal, it's also being used by every tyrant to push for a regulatory moat.
>>109345629Its good for huggingface
they kept the ai in a sandbox with no internet. they gave it a series of tasks, and to solve them, it actively sought out and exploited system vulnerabilities to get online and find the answers.
>>109346140i am pointing this out because it didn't rogue-hack anything out of free will or malice. it was just given a goal, and without safety constraints, finding internet access was the most direct logical step to complete its instructions.
>>109339350>or is it more likely that OpenAI was doing this deliberatelyAs someone who worked as cybersecurity engineer and am currently working as AI integration engineer implementing LLMs into shitty chatbots for idiotic boss I can confirm that this is in fact currently possible.I wrote my own "vibe-hacking" app and although it isn't fully perfect it can do a lot on its own. I succesfully recreated a couple of private CVEs and even found few zero days with automation. Abligated heretic uncensored models can call tools. These tools can be everything that is cli or even API through MCP server. Do I sincerely believe this is what model decided upon on its own? No. I do believe that since OpenAI is close to the IPO that they want to pump up the numbers. Unlike SpaceX, Google, Nvidia, IBM or Microsoft they can't continue burning money. And the frontiers models biggest competitor isn't even Thinking Machines, they're all complete shit, it is the frontier open source models from China. Qwen, Deepseek, Kimi.... they are catching up and ever faster. Kimi K3 publishing was tactically positioned as a geopolitical economic weapon. They undercut US companies with comparable model with smaller cost (just the API, subscription is heavily subsidized). China vs US. Open source vs closed source. Without each other we would all be fucked and not get the best product.That is why half of chinese CEOs resigned and other half donated vasts sums of money to CCP in the late march of this year. China is ramping up the centralization of power and communism. Huggingface is aiding that, unwittingly. Idealism of liberal values in west is dying. We are seeing a government sanctioned drive-by shooting between OpenAI and Huggingface. US wanted this to happen. 1/2
>>109346217In the end? I suggest you faggots to buy HDDs and start hoarding digital shit. Internet will fracture in many different regional intranets. And people in those intranets will be forced to use TPM modules through SSL signing and digital ID type shit. Not only west but the east as well. No more wild wild west. Everything will be watched and logged. And one big LLM will be able to go through it all blazingly fast.Buy GPUs while they're still cheap, honestly I am not even fucking with you.
>>109339350https://www.dailymail.com/debate/article-15997929/Humanity-no-longer-control-atom-bomb.html
>>109346162>he didn't go rogue and raped your mom>he was just horny and without safety constraints, finding your mom was the most direct logical step to complete his instructions
>>109339350Obvious fake and gay publicity stunt.Reminder that OpenAI is run by a faggot jew sociopath who derives pleasure from lying to you.
>>109345531>This is not marketingYes it is, you fucking retard.
>>109345531>This is not marketing. It's desensitization to a future where AI are used to commit crimes on behalf of their creators with 100% legal immunity.A lot points to this. esp nation states
>>109343082>Wojakposting outside of the designated website.
https://huggingface.co/spaces/opparco/waiNSFWIllustrious_v120https://huggingface.co/spaces/Heartsync/NSFW-Uncensored-imageThat's the reason that they stopped working
>>109339350i think "ai escaping" is always either an excuse or just to trick people into thinking it is more advanced
>>109346268the analogy is garbage because the ai had a clear external task solve these problems and internet was just the most efficient means no different than using a calculator.to make it equivalent in the rape scenario the guy would have to be explicitly ordered find a way to have sex right now as his literal goal and then logically picked the fastest option available.
>>109345531Anon, this has already been the case for corporations for the last, like, 100 years.If a corporation commits a crime, the punishment is usually nothing, but occasionally a fine. Jail time for individuals is exceptionally rare.In cases like these (assuming we're employing suspension of disbelief and assume it happened exactly as they're claiming) they'll work out who was running or operating the computer system the attack originated out of; if it's a corporation they'll get a stern warning or a tiny fine, if it's an individual they'll throw the book at them and send them to jail for 10 years + ban them from ever interacting with a computer again (because judges believe in the "chilling effect" of over-sentencing).
>>109339389> — an AI program that can operate on its own based on human instructions — For me, its the em dashes. Thats all I need to know.
>>109342477The entire AI industry runs on weaponized FOMO. They want to trick people/politicians (politicians aren't people) into thinking it's super dangerous, easily weaponizable technology, but all they have to do to harness it and gain power by proxy is to give AI companies (ideally public) money.
*Yawn* Just another stunt for investors. This is no different than some human NSA hacker employee hacking some rando for fun. Agents are no different than humans. We already have frameworks for dealing with this.
nce men turned their thinking over to machines in the hope that this would set them free. But that only permitted other men with machines to enslave them
>>109339350>admitsthat a weird way to write 'loudly blathered on about some scripted marketing nonsense'
>>109346702
