frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Five Eyes warns AI models capable of toppling governments are months away

https://www.theguardian.com/technology/2026/jun/22/anthropic-claude-fable-ai-model-artificial-intelligence-national-security
10•speckx•1h ago

Comments

recursivedoubts•1h ago
Nice try, you aren't gonna get me to love AI guys.
rich_sasha•1h ago
Governments able to topple themselves with 100% natural incompetence, meanwhile, have been invented millennia ago. AI is catching up.
AnimalMuppet•1h ago
Hmm. The article seems to be about AIs enabling cyberattacks.

How bad would a cyberattack have to be to topple a government? I guess I could see a parliamentary government losing public confidence if it failed to prevent a severe enough attack. I'm not sure I see it for something like the US, where elections are fixed. It might swing the next election, but that's not the same as "toppling a government", is it?

himata4113•1h ago

  misinformation (already exists)
  personalized attacks to influence voting behavior (already exists)
  attacks of the financial system (the big one here)
Basically more of the same, but at much larger scale. The only way I could see five eyes react to this is if it turns out that our financial system is compromised.
floxy•45m ago
Aren't there plenty of countries which aren't exactly super-stable anyway? Might not take much to have select countries pushed over the edge. Especially if you have territorial disputes with neighboring countries looking to take advantage. For the U.S., what if Social Security payments suddenly stopped for 6 months? What if the U.S. lost records of Treasury Bills, and they couldn't be redeemed for a while? What if electricity went out for 30% of the population for three months? I would think it pretty cold comfort that the government didn't topple, because the next election results were a mix of democrats and republicans.
himata4113•1h ago
Suddenly when we invest 100m into cyber security (project glasswire) and 100b in training toppling governments is months away, but if you were to train and hire a team of people for the same amount of money it's nothing new.

There might be a bigger reason why five eyes is doing this, but even with all the knowledge I have of the (publically known) capabilities of five eyes I really don't see why this statement exists. Either we're on the verge of real, efficient computer intelligence or this is just fearmongering that has been elevated from mouth to mouth making it sound way more severe than it actually is.

State sponsored actors have already had the capability of mythos, the only thing mythos allows is reducing the amount of precursor knowledge needed in order to perform explotation.

I've been slowly working on "mythos at home" article with solid proof of replicating many of the (claimed) capabilties that mythos has with opensource models and GLM-5.2 has been pretty instrumental in advancing it forward.

I strongly believe there is either something extremely serious being hidden from the public (nothing new, five eyes operates in secrecy) or this is overblown fearmongering being brought up to the upper decision center that are afraid that non state sponsored entities now have the capability to achieve what was previously only possible when you're backed by a state.

On the bright side, glm-5.2 is pretty good at autonomously optimizing software so at least we're not going to be locked out of the frontier for long.

--

Side note: The amount of vulerabilities found in blockchains around gpt and opus 4.7 release does show that it is a real problem and I am not denying that. There have been several government agencies that have suffered from data leaks last few months as well as general more public CVE's such as copyfail. But I still believe if the same kind of money was invested into people we'd have similar if not better results.

Bender•50m ago
If AI can do mass exploitation then it can also do mass patching. Some malware in the past has been used to mass patch machines.

Get to it. Give the teams doing this presidential pardons and full immunity. Apply patches and mitigations on all the things. Impress us all. One big downtime, get 'er done.

AnimalMuppet•47m ago
Do you trust the AI to build perfect patches? Do you trust it to not leave backdoors in the process?

If it's smart enough to do the first, I don't trust it to do the second.

Bender•41m ago
I don't even believe that AI is real. To me it's just a predictive chat bot using LLM with big-data shoved up it's back-side. As for trust, I do not trust anyone or anything but if our choices are mass exploitation and mass patching, I will take the latter if people are not patching their own stuff especially if AI can gain access to it already. If AI can get to it so can spooks right now. There is a lot of hardware with state operated unofficial remote access and some of their companies are listed in In-Q-Tel's website.

If there is a third option where companies drop everything, all hands on deck to patch all the things I would take that but it's just never been a priority. That's why I was always a big fan of ransomware separating customer data away from companies that should not have had it in the first place.

floxy•42m ago
That's pretty much the plot for one of the latter Terminator movies, right?

https://en.wikipedia.org/wiki/Terminator_3:_Rise_of_the_Mach...

Steam Machine

https://store.steampowered.com/hardware/steammachine
387•theschwa•1h ago•340 comments

Moebius: 0.2B image inpainting model with 10B-level performance

https://hustvl.github.io/Moebius/
145•DSemba•5h ago•36 comments

Deno Desktop

https://docs.deno.com/runtime/desktop/
911•GeneralMaximus•13h ago•337 comments

Show HN: Oak – Git replacement designed for agents

https://oak.space/oak/oak
69•zdgeier•3h ago•82 comments

My Mathematical Regression

https://blog.dahl.dev/posts/my-mathematical-regression/
45•aleda145•3d ago•12 comments

Prompt Injection as Role Confusion

https://role-confusion.github.io
69•x312•3h ago•38 comments

Nintendo Wii U games running from a 1980's Bernoulli disk [video]

https://www.youtube.com/watch?v=8GZDOpV2OXk
49•zdw•22h ago•20 comments

Codex logging bug may write TBs to local SSDs

https://github.com/openai/codex/issues/28224
378•vantareed•11h ago•208 comments

Pledging another $400k to the Zig software foundation

https://mitchellh.com/writing/zig-donation-2026
586•tosh•5h ago•187 comments

Charge Robotics (YC S21) Is Hiring Software and Hardware Engineers

https://jobs.ashbyhq.com/charge-robotics
1•justicz•2h ago

The text in Claude Code’s “Extended Thinking” output

https://patrickmccanna.net/the-text-in-claude-codes-extended-thinking-output-is-not-authentic/
213•0o_MrPatrick_o0•4h ago•153 comments

Bain tests software takeover targets by vibecoding AI replicas

https://www.ft.com/content/e5bac4d1-b1f8-43a4-bd54-b182d5357af0
16•macleginn•3h ago•31 comments

Finding the Best Dog Treat with Statistics

https://www.wespiser.com/posts/2026-06-19-best-dog-treat.html
21•wespiser_2018•1h ago•3 comments

GLM 5.2 vs. Opus

https://techstackups.com/comparisons/glm-5.2-vs-opus/
405•ritzaco•11h ago•280 comments

Die analysis of the 8087 math coprocessor's fast bit shifter (2020)

https://www.righto.com/2020/05/die-analysis-of-8087-math-coprocessors.html
54•Jimmc414•5h ago•9 comments

DisplayMate

https://www.displaymate.com/
40•skibz•2h ago•13 comments

Mexican government unveils a prototype for a new homegrown, ultra-affordable EV

https://gizmodo.com/mexico-just-showed-off-a-new-extremely-cheap-government-backed-ev-2000769080
85•speckx•2h ago•35 comments

Blogger defeats photographer's copyright claim

https://blog.ericgoldman.org/archives/2026/06/blogger-defeats-photographers-copyright-claim-sokol...
29•speckx•2h ago•8 comments

How it feels to practice for IOAI in Iran

https://howitfeels.netlify.app/
15•linmer•2d ago•0 comments

window.showDirectoryPicker opens up a whole new world

https://steveharrison.dev/showdirectorypicker-opens-up-a-whole-new-world/
78•steveharrison•6h ago•82 comments

Show HN: Got sick of ads, so I made my own logic puzzle site

https://puzzlelair.com/
77•HaxleRose•6h ago•69 comments

NSF slashes research programs to support new tech initiative, insiders say

https://www.science.org/content/article/exclusive-nsf-slashes-research-programs-support-new-tech-...
104•strangeloops85•2h ago•40 comments

Help I accidentally a wigglegram

https://lmao.center/blog/wiggle-accidents/
438•gregsadetsky•2d ago•111 comments

Alan Greenspan has died

https://www.washingtonpost.com/obituaries/2026/06/22/alan-greenspan-most-powerful-central-banker-...
147•helsinkiandrew•7h ago•152 comments

Granularity comes at a cost

https://www.sidhantbansal.com/2026/Granularity-comes-at-a-cost/
39•sidhantbansal•2d ago•6 comments

Chevron signs 20-year power agreement with Microsoft for West Texas data center

https://www.chevron.com/newsroom/2026/q2/chevron-signs-20-year-power-agreement-with-microsoft-for...
77•cdrnsf•5h ago•71 comments

Show HN: Selector Forge – browser extension for AI-generated resilient selectors

https://github.com/Intuned/selector-forge
22•ahmadilaiwi•4h ago•0 comments

Show HN: CleverCrow: give tokens to your favorite projects

https://clevercrow.io
42•zhubert•1d ago•61 comments

DHL Set to Transport Goods on New Wind-Powered Cargo Ships

https://www.wsj.com/pro/sustainable-business/dhl-set-to-transport-goods-on-new-wind-powered-cargo...
103•julienchastang•4h ago•43 comments

There is minimal downside to switching to open models

https://www.marble.onl/posts/cancel_claude.html
364•amarble•22h ago•296 comments