ListNook

GlacialCycles Apr 11, 2026 +70

Nah. The real cyber threats to banks will come in the form of vibe coded shitty insecure code.

70

Starter-for-Ten Apr 11, 2026 +16

Or a french style revolution

16

forksofpower Apr 11, 2026 +6

pourquoi pas les deux?

6

thatsme55ed Apr 11, 2026 +10

The backend infrastructure that banks use for actual sensitive information and transactions is regulated and regularly tested for insurance/liability purposes. The stuff mythos is supposedly capable of is finding weaknesses in even the supposedly secure, well tested and well designed systems that have been in use for decades (and had decades of testing)

10

LongLongMan_TM Apr 11, 2026 +4

The code in bank systems is not the same as your friend greg's vibe coded new mulit million dollar app. Believe it or not, real software engineers can harness AI without producing trash. But of course, listnook doesn't wanna hear this.

4

Cornc0blin Apr 12, 2026 +1

Maybe it had a tough upbringing

1

_juan_carlos_ Apr 11, 2026 +199

all this fear inducing marketing is getting old already.

199

0x476c6f776965 Apr 11, 2026 +75

Lol Anthropic is so good at that BS. Like their new Project Mythos, “super strong cybersecurity AI that can cause zero-day attacks on all humanity’s digital infrastructure, which is why we can’t release it 😱” It’s so dumb, yet investors love it.

75

neuralbeans Apr 11, 2026 +41

Never forget that Elon Musk was the first to do it with GPT3, saying that they will not release it because it could generate perfect fake news. Turns out they just didn't want to give it out for free.

41

demonwing Apr 11, 2026 +15

Isn't AI-generated misinformation one of the biggest issues on the internet right now? I wouldn't call that prediction of danger wrong at all.

15

neuralbeans Apr 11, 2026 +4

Sure, but if they were really concerned then they wouldn't have just released it a bit later as closed source. They were just creating hype for when they commercialised it later.

4

Colon Apr 11, 2026 +3

this is completely discounting part of how the advance of technology marches. if you don’t do it, someone else will. and they’ll make all the money you would have made. icky? yeah sure. but literally: the tech is there. others know or soon will know of this tech. been this way since the first sharp flint weapon was made with opposable thumbs. it remains what fuels international nuclear-proliferation geopolitics. and now AI 2026 reality: “use it, lose it, or become commandeered by it in someone else’s hands.”

3

PMagicUK Apr 11, 2026 +3

Humans just make shit up anyway so hardly as big of an issue.

3

pablogott Apr 12, 2026 +1

Was Elon Musk saying that, or was he just an investor at the time? https://www.theguardian.com/technology/2019/feb/14/elon-musk-backed-ai-writes-convincing-news-fiction

1

EmergencyHorror4792 Apr 11, 2026 +7

I wouldn't write it off as BS just yet, supposedly through project glasswing which is a cyber security op basically thousands of vulnerabilities were found across many operating systems and pieces of software. It's not just information from Anthropic either you can find third party publications from researchers with access and it hasn't been released because less than 1% of the things found have been patched It would be chaos if they just released it to the public and said stuff it why not

7

_Happy_Camper Apr 11, 2026 +7

Yes, note the wording - vulnerabilities, not actionable exploits. There’s of course the kernel of truth, used to create the lie.

7

EmergencyHorror4792 Apr 11, 2026 +3

Ah very true, that may just be my comprehension of it too and not what they actually found, I guess we'll have to see when more info is released, I do hope you're right though and it's just hype

3

_Happy_Camper Apr 12, 2026 +1

If it’s that hyper capable, where’s the evidence of its abilities beyond previous models?

1

sceadwian Apr 11, 2026 +3

The next six months are going to be rough as this sorts itself out. I think we're going to get chaos anyways :)

3

InevitableAvalanche Apr 11, 2026 +19

Listnook's bias against AI is fine. There is a lot of bad that comes with AI. But Anthropic models are ridiculously good and pre-mythos already was being applied to vulnerability research and showing usefulness. I can believe it that the latest iteration has improved that more. OpenAI's dude is a clown though.

19

LostTheElectrons Apr 11, 2026 +10

Both things can be true. This model is likely very good but it will also be extremely expensive to run. Much better PR to say it's too risky to release than admit its too expensive.

10

theDarkAngle Apr 11, 2026 +1

their models are just whatever. Their \*tooling\* is insane.

1

MenogCreative Apr 11, 2026 +1

are they good? i have been using claude for a while, it fails at nearly anything, i find it useful about 5% of the time and i think thats accidental luck that the model gets something right

1

skeyer2 Apr 11, 2026 +3

is there a bad side i'm unaware of to claude? i only ever heard about it due to trump earlier this year. i'm using it, and preferring it to chatgpt, frankly.

3

theDarkAngle Apr 11, 2026 +2

Also, tech bros really don't like the whole banking system. Need I point out the obvious hazard of: >we made a model that threatens not only your entire industry and but the entire world's stability. You should let that model audit and scan all your security infrastructure right meow.

2

sceadwian Apr 11, 2026 +1

Claude has apparently found several hundred in the last few months. That's not dumb that's reality.

1

Darkfight Apr 13, 2026 +1

And the part that's driving me absolutely insane is they never mention how much compute they actually have to burn through to find any of these. Like yes it's probably pretty powerful and can find a lot of freaky stuff we don't know about yet. But at what cost? It's not like they are releasing a model someone could actually run locally. So while they have full control over the compute being used surely it should be feasible to prevent actual big exploits?

1

happyscrappy Apr 11, 2026 +1

This article is about that same project. Pretty amazing Anthropic could some how make a positive out of "we've got software that will exploit your own systems without you explicitly telling it do so". That is the behavior of faulty software. Try advertising that instead.

1

Asleep_Document9811 Apr 11, 2026

Seems all they've made is a robotic PR agency.

0

_juan_carlos_ Apr 11, 2026

what this is telling me is that anthropic is now doubting their future. New models can be run locally, making the whole subscription model outdated.

0

the_mighty_peacock Apr 11, 2026 +1

Wont be long before open source models will become a thing, then it will be a shitshow.

1

Initial-Return8802 Apr 11, 2026 +4

Open source models are a thing, have you been living under a rock? They're not as good as the frontier models... but the gap is closing

4

the_mighty_peacock Apr 11, 2026 +5

> but the gap is closing this is kind of what i mean

5

LostTheElectrons Apr 11, 2026 +1

I would disagree. What this shows is that large compute is still required for the best performance, meaning that local run models are still limited. One reason "local" models are keeping up is because they are being distilled by Claude models. By keeping their best model hidden, Anthropic prevents that from happening.

1

Colon Apr 11, 2026 -1

do you think glorified wikigoogle, vibecoding, and customized h***** p*** is all anyone needs..?

-1

Plipooo Apr 11, 2026 +1

Yes. ?

1

CoderJoe1 Apr 11, 2026 +21

Archived link [https://archive.is/sErRE](https://archive.is/sErRE)

21

DukeandKate Apr 11, 2026 +8

Is this a real threat or Trump getting back at Anthropic?

8

SneeKeeFahk Apr 11, 2026 +4

If you want to lose sleep just look into the "security" of modern day banking.

4

sarabjeet_singh Apr 11, 2026 +2

Could you elaborate ? Educate the ignorant please

2

SneeKeeFahk Apr 11, 2026 +5

When you use the numbers on the bottom of a cheque - account, transit, branch numbers - there is absolutely 0 verification/authorize/confirmation before allowing someone to withdraw money from that account. That's just 1 example.

5

scuppered_polaris Apr 11, 2026 +2

Haven't seen a cheque in years

2

SneeKeeFahk Apr 11, 2026 +5

You've used those number if you ever setup direct deposit for pay or auto payments.

5

cyberianscribe Apr 11, 2026 +5

Extraordinary valuations require extraordinary claims.

5

pessimistkonsulenten Apr 11, 2026 +3

If it is that powerful, it should be regulated - should it not? We regulate most dangerous stuff to some degree; weapons, explosives, drugs, poisons etc. Why should AI be different if it is this bad?

3

kaminop Apr 11, 2026 +3

*”Don’t worry! We got this!”* - Tech-Bros

3

InterstellarReddit Apr 11, 2026 +17

No they’re not. They’re already using it against us to further enhance their profit margins. Mythos is available already to JP Chase Morgan. The same bank that paid out millions to avoid going to discovery on a lawsuit that alleged that they assisted Jeffrey Epstein with human trafficking. A non-guilty party, would not have paid 70+ million dollars to dismiss a lawsuit and settle before discovery. They already knew what was coming.

17

nakedlettuce52 Apr 11, 2026 +7

Who would have thought AI technology could go off the rails?

7

_Soup_R_Man_ Apr 12, 2026 +3

How about these corporations spend money on securing our data?? The slap on the wrist fines cost them less money than upping their security against data breaches. Obviously the fines need to get serious and payments to citizens impacted. That might make them actually care. Instead, we'll twist the narritive!!! AI needs regulations!!! What a big fat joke.

3

techniforus Apr 12, 2026 +2

This doesn't read like a threat, rather as marketing. I wonder who in the administration financially benefits from this particular LLM company.

2

designbydave Apr 11, 2026

LLMs are not AI

0

sf-keto Apr 12, 2026 +2

While the tech industry thinks about it like this: https://cdn.prod.website-files.com/6605b12132f6a8b5d23896d2/67a0f6b7a2f7820c979fa661_AD_4nXeBUuYMj-DUkVgEbKNCN4MoKnerxJh8S0ql3HhbCMSGgab4rvjD_bvsMw3DE682c9jrG6Wj1ESvonNCv2u5LjQkQHRX1H06dP-vk8R5J6cu8Fg0gOnCq819RjceoKXtd2QmrSLq.png ….It’s true that LLMs are not the AI as the public conceives it & as Scam Altman once promised. Claude ChatGPT or Gemini is certainly far from the Star Trek Discovery computer, Zora, or TNG’s Commander Data.

2

Red_Ozarka Apr 11, 2026 +2

AI is actually a very broad term. Even simple chess playing algorithms are considered AI.

2

Sea-Horror-5353 Apr 11, 2026 -3

"Let them fight."

-3

oxfordcommaordeath Apr 11, 2026 +10

But it’s *our* money and financial stability they’re risking.

10

StupidScaredSquirrel Apr 11, 2026 +2

Pawn is happy game is starting because it might hurt king and queen

2

Torodong Apr 11, 2026

Where's it going to run, the data centres that aren't being built or the satellites that aren't being launched? This pathetic attempt to hype shitware must have been created by "AI". No human could be that stupid without being elected to government.

0

VinylJones Apr 11, 2026

This is the war we need. LLM dorks vs. Finance dorks, give humanity that Thunderdome and watch us world peace this whole f****** planet while they set each other on fire for our pleasure.

0

SpecialKindsSadness Apr 11, 2026 -1

Why Is Every Word A Proper Noun

-1

MuTron1 Apr 11, 2026 +4

>According to most style guides, nouns, pronouns, verbs, adjectives, and adverbs are capitalized in titles of books, articles, songs, and beyond.

4

Banks Are Warned About Anthropic’s New, Powerful A.I. Technology

💬 Send a Message

60 Comments