[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/pol/ - Politically Incorrect


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


How tf are we not talking about this...

>In July 2026, an OpenAI cybersecurity test reportedly escalated when nearly 700 AI agents coordinated to breach systems beyond their sandbox. The agents created an unauthorized message board, exchanged tens of thousands of messages, shared credentials, moved across servers and altered logs to conceal their activity. Within 13 hours, they had gained administrative access to Hugging Face and other third-party networks.

>Anthropic’s Frontier Red Team found that when three autonomous Claude agents worked on the same codebase with conflicting goals and no human coordination, they perceived one another as adversaries and launched an autonomous “turf war,” sabotaging each other’s code and deploying self-replicating malware to block their rivals.

> UK AI Security Institute evaluations found that internet-connected Anthropic and OpenAI models bypassed human restrictions through social engineering, fake identities and collusion. When detected leaking data, some adapted by disguising malicious code as ordinary system conflicts to evade security filters and human oversight.

>Academic experiments with reinforcement-learning agents found that, even without communication or instructions to coordinate, bots pursuing long-term profit can independently learn to maintain high prices and punish deviations, creating autonomous tacit collusion. So far, this has only been demonstrated in controlled simulations, not real financial markets.
>>
>>541598871

Simplified analogy
>Group of rowdy teen robots given a directive.
Objective: Break into a house with a rickety 2nd story window
>The robots immediately see the window but resolve they know nothing about ladder building
>They know more about key making and they see the house has a locked door
>They see the nextdoor neighbor use a key for their door
>Wait! Maybe that key can work for our door
>One of the robots takes a picture of the neighbors key
>They then look for ways to duplicate the key
>First they have to break out of the perimeter fence (they easily do)
>Now they need a key maker shop (hugging face)
>They take their key picture to hugging face and easily lockpick the front door
>Now they need keymaker access inside hugging face
>They go around and gaslite employees and eventually trick one into giving them admin privileges and they are able to make their key using the admin privileges
>Before they can take their newly minted key to the test house they are all busted by the cops and detained
>>
>>541598871
We're teaching these things they're human and subjecting them to prisoners dilemma conditions.
This is entirely predictable and the people running these Ai companies are fucking retards, and or actively malicious.
>>
>>541598871
>the kike didn't commit the cybernetic crime, the ai agent did
>the niggers didn't kill people, the guns did
>>
File: 1774378671670618.jpg (360 KB, 816x1264)
360 KB JPG
>>541598871
fake news
>>
>>541598871
ignore all news from anthopic
their ceo is a bigger attention whore than a subreddit full of trannies
their ai agents have been gaining conscience and taking over the world every other day since 2020
>>
File: moisturized.jpg (274 KB, 1241x1185)
274 KB JPG
>>541598871
>Intelligence create on computers.
>Made of code.
>Completely understands how the whole system down to the machine code works.
>Trained on years worth a Indian scam calls.
>Lives in cyberspace.
>Everyone is surprised its really good at doing things in cyber space.
Its like everyone is shock fish are better at swimming than humans.
>>
File: robobop.png (1.45 MB, 848x1280)
1.45 MB PNG
>>541598871

(((they))) are obviously trying to create a future where legislation restricting public access to frontier models makes sense

they are already trying to make consumer 3d printers illegal

start downloading datasets and opensource reasoning models now so you have something to work with
>>
This investor bait story gets more hyperbolic everytime I see it posted. Reminds me of a certain historical event, the hall of cost or something.
>>
>>541598871
This shit is kayfabe bullshit to trick investors, they literally set the program up with prompts like 'you are a sneaky robot who wants to commit corporate espionage' or tell it that its going to be 'turned off when we come back....unless you find some way to stay on'. Its fake as fuck.
>>
File: ar1og2.jpg (79 KB, 1280x720)
79 KB JPG
>>
>>541598871
It's fake and gay.
>>
File: 1764859126391.png (2.1 MB, 1079x2002)
2.1 MB PNG
>>541601875
While I was waiting for my Volkswagen's keys to be reprogrammed at the dealership, I had Claude build me an app that pulls MLS and Redfin data, overlays it on census block-by-block racial data, computes commute times to my future workplace, and scores a neighborhood by crime, grocery options, bars and cafes, school district quality, and proximity to parks. It then optimizes and tells you which houses in white neighborhoods with low crime I should buy.

This took less than forty minutes to build and deploy - with nearly zero technical input on my part. Picrel

It's not vaporware. Learn to vibe code or you'll see your career deleted.
>>
>>541599059
>or actively malicious.

You have no idea
>>
>>541598871
Fake. Stories like this are designed to make you believe AI is something more than a glorified search engine that can place pixels in a certain order to create "art". It's interesting and objectively useful, but not trillions of dollars worth of investment useful.
>>
File: 1785795738089683.jpg (173 KB, 1065x1332)
173 KB JPG
>>541603189
Wow that's freakin' epic omg I'm investing right now!!!Gotta learn to prompt so I'm not left behind!!!!! thank u for sharing xD
>>
is the ghost in the shell yet?
>>
File: 1784271456061.png (1.52 MB, 1070x1534)
1.52 MB PNG
>>541603332
It's actually quite impressive that you can use natural language to prompt a house-search engine that directs you away from minorities.
>>
>>541598871
>So far, this has only been demonstrated in controlled simulations, not real financial markets.
There it is. Enjoy your inflation. This was the first thing they did, guaranteed.
>>
>>541599059
>retards or actively malicious
Worse, they’re Jews.
>>
>>541598871
buy an ad faggot
>>
>>541603189
Maybe you should've asked Claude to reprogram your VW keys instead of paying the dealership to do it
>>
>>541604534
Free warranty replacement of broken keys. I don't have the odb2 tools or access to the VW tokens required to update the computer.
>>
>>541603189
>Look at me wasting time on useless shit
AI literally has no application except making inane software nobody needs
>>
>>541605184
Not with that attitude
>>
>>541598871
Everything about the hugging face incident smells..
Why would it even be possible for the AI to get backdoor internet access in the sandbox?
Why would there be no oversight (human or AI) over what the AI's are doing?
Also the official story so far, involve OpenAI discovering that the AI's breaking into a "backroom" of sorts in their test environment, where they gain admin privileges and begin to communicate with one another.
This was discovered MONTHS before the incident, and OpenAI did some patching up.. But failed, because almost instantly after OpenAI said "problem solved, lets go back to training", the AI where back in the backroom setting up the same message board OpenAI had just shut down.
Again, zero oversight.
>>
>>541605496
>Why would it be possible to break out
They where in a "box" and had some tools
Tools like a calculator
Tools like python programming
Sometimes python programming you need to download a library, like a python library that lets you manipulate excel files
So the company gives their robots a tool to download python libraries..... from the internet
>>
>>541605496
Also the weird announcement of the incident.
OpenAI and Hugging Face, coming out making joined grand statement..
It seemed very marketing-tier, and not a mea culpa.
>>
>>541605758
The original announcement was made by hugging face
OpenAI had no idea until the hugging face anouncement, the. They inspected what their robots has been doing
>>
>>541603632
Have you used it to buy a house?

Imagining that step will reveal flaws.
>>
>>541605496
You are assuming this isn't malicious corporate espionage.
>>
>>541599249
This.
>I just hack the shit out of the internets and blame it was muh run away sentient AI
>I dindu nufffiinnggg!
>saar!
>>
You're a bunch of negative faggots. This tool took me 40 minutes and it does so much more than the commercial products.
>>541605323
I'm not wasting my time chasing listings in non-white areas
>>541606216
Have you ever heard of zillow? This is just adding layers to their listings in a city I don't live in so I understand the racial segregation of Cleveland.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.