210 comments of 213

[ 1.6 ms ] story [ 79.0 ms ] thread
i gonna ask: how can they still use claude? i thought all users in china are banned
Cc can be used with non Anthropic models.
> how can they still use claude?

Workarounds aside, it says Claude Code not Claude.

i.e. they are using the CLI running any model. You can for instance run GLM with it.

There is a reason Singapore tops the rank on Claude usage
Does Alibaba only have developers in the China?
Alibaba has engineers in Hongkong, Singapore, North America. It’s a global corporation
The same way they buy "banned" and "sanctioned" NVIDIA GPUs.
What Claude Code did is absolutely mindboggling tho, if Chinese harness did that probably POTUS would lose sleep.
Exaggerate much? If you think POTUS would lose sleep about a date format timezone marker, I don't know what to tell you.
Wait what do you mean "if"?
Maybe if they didn’t farm all the data from Claude to train their own trash models. Anthropic wouldn’t feel the need to do it.
Bit rich given where Anthropic sourced the data to train Claude with. What's good for the goose is good for the gander.
It seemed pretty mild compared to what's collected by modern websites and apps, though? How many don't know your Timezone?
> How many don't know your Timezone?

The timezone fetch was to alter program behaviour at runtime, not to send arbitrary timezones for tracking reasons.

It was one way of detecting if it was a chinese person using the program and then behaving differently.

Malware behaves this way. STUXNET for example was wired to do nothing except propagate unless the environment had the right conditions.

The article on HN only said that they seemed to be collecting this to detect resellers. How else did the behavior change?

Most services I know that are trying to block abuse do collect device info

There is this whole thing where Fable silently starts behaving worse if they suspect you are trying to use it for RL or are otherwise building a competing product. This is likely the primary vector how that works: they check if you are in china, if you proxy your requests, and if you are from a list of known labs or match a couple keywords
Another reason to use open source coding agents and local language models.

Claude Code is neither and it is literally info stealing malware.

Considering their massive distillation, if US companies stop publishing new models to the public, would China still be able to develop new open weight models?
Yes, 100%. GLM 5.2 is capable of RSI. It's too late to stop.
I don't think China would strugle to scrape the internet for fresh data.

And they constantly publish state of the art LLM research (see DS4 context compaction and cache tech).

They have very capable tech giants. So while not being able to distill western models would probably have some impact, it's probably becoming lesser as time passes.

We might even see Western LLMs distilling Chinese models soon. If they aren't already to some extent.

Everyone distills/copies training data.

A couple months ago when Anthropic was complaining about Chinese distillation, people found that Claude self-identified as "DeepSeek" when asked in Chinese:

https://x.com/stevibe/status/2026227392076018101

It's really a fiasco of massive hypocrisy at this point.

China has most probably already achieved "escape velocity" on the software side. Now if they achieve parity, to some degree at least, on the hardware side with Nvidia it is very possible they'll overtake the US.
Probably yes.

More than a year ago, when Anthropic and OpenAI started to gide the reasoning bits from the output, a lot of people here on HN predicted that Chinese models days were numbered.

Fast forward to today, and models such as DeepSeek and MiMo are nothing short of excellent. I haven't used GLM or Qwen but heard very good things about them as well.

This "massive distillation" sounds a lot like anxiety about how companies from outside the US can develop very good models themselves.

In my personal, subjective opinion GLM-5.2 is on par with GPT-5.3
Of course, it is like any other kind of weapon system, eventually the knowledge gets acquired.
Depends on a lab, but they do have plenty of compute and engineering. So this would only slow down the progress.
Look at all of the software that has been developed as an alternative (and often an upgrade to) software in the west. (Baidu, Wechat, etc)

Many of the top AI researchers at western companies are from China, and many are returning.

It doesn't matter, the only models getting compared are the public ones.

If Anthropic had a super secret model that nobody has access to, I'm not sure why would I care about it since I can't access it.

Regardless of whether this specific claim is true, enterprises are becoming much more cautious about developer tools that can read large portions of proprietary codebases.
Wasn't one of the big promises the AI labs made "uncopyrighting"? Ie. the ability to reconstruct large works, including source code, without actual access to the source code? Everything from movies to operating systems.
Cleverly compressing and decompressing doesn't de-copyright it. ... and if it's not the same who'd trust it.
Interesting, I haven't heard this claim before. I suppose that claim made sense if their customers were big corporations, not so much when its the masses generating bootleg software copies.
I remember hearing something about this. Reminds me of the many lies that political candidates make to garner interest and approval. Except who's holding them accountable - like there's not even a list anywhere tracking these lies.
Becoming? We've moved entirely in the opposite direction.

When these tools first appeared the overwhelming conversation was about the risk of letting a remote tool siphon your code and intellectual property (where eventually they're going to add that to their training). Now everyone is using them, and that fear seems to have dissolved. Every corporation is sprinkled with Claude Code, Antigravity, Copilot, Codex, and so on. Even the long fear-mongered Chinese providers are being heavily used in many spaces.

In this case this is a PR battle between two firms, and it isn't much more. And Alibaba isn't worried about the "proprietary code" (the truth is that there is incredibly little interest in most orgs code), but that the tool is a backdoor, or at least that is the claim.

> and that fear seems to have dissolved

Until the first big incident, yes.

> there is incredibly little interest in most orgs code

I think from a commercial perspective yes, but access to source code is very good for finding exploits which could be very valuable for governments. I could also see a future where companies are directly cyber-attacking competitors in hostile markets too...

not to mention they are kind of capable of executing code and susceptible to injections which also amounts to being practically backdoors if youre not super careful about how u use the tooling
It's insane that it's becoming a concern now. It should've ended the discussion from the very beginning.
Enterprises host their entire infrastructure on US-base clouds. And for many, it still is not a problem.
The recklessness of coding agents having access to work laptops and exfiltrating data with barely any restrictions is on a whole new level.
I mean, we all also still do manufacturing in China with a 100% guarantee that your widget will be copied and cloned. It's so much cheaper though....
After they uploaded their code to GitHub, Bitbucket etc since forever? It would be schizophrenia.
Big customers usually use GHE served on prem due to security concerns, no?
I really have no idea. I work only with small or at most medium sized companies. All of them put their code on a git server they don't own. All of them are concerned about AI companies looking at their code. They hope that at least they won't train their models with their code if they pay.

I think that the reasoning is: they trust the git company (whatever it is) not to sell their code. They are worried that their code goes into a model and somebody else could ask the model "write a service like XYZ" and it will regurgitate their code.

No, they run GitLab because GitHub Enterprise is a horrible thing nobody has ever said a good thing about.

GitLab even has a free self-hosted version, and it has a number of advantages (like being able to actually have a structure with inherited secrets and accesses, and no, GitHub Organisations do not count and suck). And for years thanks to GitLab-CI it was clearly ahead.

That's why self-hosted GitLab is so popular in Europe.
A bit too late for that, most of them have already dumped most of their codebase and IP into cloud models.
If you're using a coding agent then obviously you need to either serve the model yourself or trust whoever you are sending your data to.

In terms of WHAT you need to be concerned about, it seems it goes far beyond code, and far beyond having to trust your model provider.

A coding agent with access to a bash tool is going to have access to anything that a human with a bash prompt would, and even if you try to provide a nailed down sandbox environment for the agent, you still need to be concerned about things like unencrypted passwords and keys that it may be able to find "laying around" in code or databases/etc it has access to.

Translation: Alibaba will continue distillation attacks using accounts that aren't directly attributable to it's own corporate infrastructure.
(Mis)anthropic already performed "distillation attacks" on the internet.
i can see why they want to stop it but 1. you have to pay for the "attack" 2. these AI companies trained on copyrighted content without permission or attribution to anyone who's data was used to train.
How exactly the word attack fits in that phrase?
As long as they're paying for the tokens, there's no attack . Otherwise you have to call training on copyrighted material theft.
Did Anthropic perform "distillation attacks" when they hoovered up the entire internet?
> Translation: Alibaba will continue distillation attacks using accounts that aren't directly attributable to it's own corporate infrastructure.

What's a "distillation attack"? How is it different from simply distillation?

The target doesn't want to be distilled.
> The target doesn't want to be distilled.

So?

Fraudsters don't want to be jailed, their victims don't want to be scammed, employees don't want to be laid off, etc.

What the target wants is irrelevant - what society wants as enforced by laws is what is relevant, and as the leading AI providers have demonstrated, simply grabbing other people's copyrighted stuff for learning purposes is perfectly fine!

If they already think this practice is fine, why would I believe that their concerns about this are real?

I was only describing the difference not taking a side.
It's pretty much the same as when "installing programs on your computer" is called "sideloading". Deliberately deceptive, weaponized language to make it seem like a bad thing.
using infected machines as proxies would be a fair line in the sand
I think we should start calling it "distillation terrorism" just to make it sound even more absurd.
It's pure model murder, and if you call it anything else, you're an anti-American communist.

  1. LOL I've just downloaded literally whole internet and copyrighted books and put them through a neural network. Now I have this whole knowledge in my LLM.

  2. Hey? Are you using my NN for training your NN? you're a thief!
The corollary is that there are no morals once the stakes are in the $ billions, let alone hundreds of billions.

This isn't even about a single person or personality. Very few people in such position could stand fast by their moral code. In any case, an environment that favors profit above everything will naturally select for individuals who are unencumbered by such hindrances.

There might've been 100s of Altmans and Amodeis who had a strong moral code but we don't know about them because they dropped out of the "race" because of said moral hurdles.

Copyright law is an artificial legal construct, not a moral code.

I think appropriate attribution is a moral code, but I am not able to attribute every idea I have to all those who helped me develop the general intelligence that I use to develop such ideas.

> an environment that favors profit above everything will naturally select for individuals who are unencumbered by such hindrances.

Exactly. Dairy farms optimise for milk production so favour cows that produce the most milk.

The market economy optimises for profit so favours those most willing/able to generate it. Zuckerberg, Musk, Thiel, Andreesen and co are products of the system.

> The corollary is that there are no morals once the stakes are in the $ billions, let alone hundreds of billions.

terrifying

Remember how Kim Dotcom got destroyed for criminal copyright infringement? One would think the big tech CEOs would face the same fate, that police officers would rappel down helicopters, storm their mansions and bring them out in cuffs.

Instead the AI companies reached these absurd settlements with publishers that made a mockery out of all the previous copyright enforcement victims.

Reminds me, did the AI companies redistribute that copyrighted material to others and make their money that way? Did Kim use the copyrighted material to generate something novel from it?

copyright law literally says something isn’t infringement if it is a novel transformation. I get the jokes and criticism about AI companies fighting and complaining about competitors distilling, but this is a much weirder comparison.

Now you just have to explain this:

"Anthropic settles with authors in first-of-its-kind AI copyright infringement lawsuit" - https://www.npr.org/2025/09/05/nx-s1-5529404/anthropic-settl...

> "The training use was a fair use," [the judge] wrote. "The use of the books at issue to train Claude and its precursors was exceedingly transformative."

> However, the judge ruled that Anthropic's use of millions of pirated books to build its models – books that websites such as Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi) copied without getting the authors' consent or giving them compensation – was not.

It seems clear from the article that while the use of pirated works was illegal, the use of copyrighted works (a the work a book is based on is still copyrighted if you buy the book) was fine and transformative.

(comment deleted)
But distribution isn't the only crime here, obtaining the material illegally apparently is a crime too. And the damn robot can also spit me harry Potter verbatim so I don't know how it would also not be distribution?
>And the damn robot can also spit me harry Potter verbatim so I don't know how it would also not be distribution?

if you prompt it to, yes. just like your browser dutifully navigates to any copyright-infringing resource and GETs and POSTs whatever you ask of it.

(also it can't, not really, only small snippets before going off rails. LLMs aren't magic, they can't losslessly compress an exabyte of training data into a few terabytes of weights.)

If I read Harry Potter I will remember some parts verbatim. Others I will tecall in only an abridged and lossy way.

Does that make my brain copyright infringement? Does Disney now own all my output forever because some small part of me now has Harry Potter embedded?

Does the law really not distinguish between mechanical processing of data, and humans learning from it? It seems surprising to be if every person who read a textbook is copyright infringing. It also seems surprising if something like a lossy compression algorithm is enough to protect you from copyright law.

Somewhere between the two a line must be drawn… where we’d want to put that line, I guess, if up for quibbling. But it doesn’t seem obvious to me.

>Does the law really not distinguish between mechanical processing of data, and humans learning from it? It seems surprising to be if every person who read a textbook is copyright infringing. It also seems surprising if something like a lossy compression algorithm is enough to protect you from copyright law.

The google books and google thumbnails cases have so far upheld that even mechanical reproductions are allowed, depending on the context/usage.

To me the distinction hinges on the output being transformative enough to be considered a new work. I think that most of the time LLM output is.

Sometimes they go a bit wonky and overtrain on specific phrases which can result in verbatim copies of brief sections of coontent. Thats a bug, not a feature.

This analogy is disingenuous because by comparing the human brain to the machine, it ignores _scale_. Scale is absolutely important in copyright law. As a matter of fact, copyright law is among the various profound impacts of the---wait for it---printing press, a _machine_ for the mass production of books.
So if I watch a LOT of Disney movies THEN they own my own unique output forever?
Further, why has my brain's searing remake of Snow White as a gritty murder mystery gone unscathed by Disney lawyers? Surely their negligence has diluted the Snow White trademark!
Can you remember every part? Can you do this for every book in a library? Will you remember every part of every book forever?

If you just ignore anything that's inconvenient for your argument, you can make any argument you want.

>Can you remember every part? Can you do this for every book in a library? Can you remember all that forever?

None of those are relevant factors when it comes to copyright law. You don't get a pass for copyright infringement just because you're not copying the entire work. Same goes for a copy that's transient. You can't set up a bootleg movie theater in your home, even if you delete the movie file afterwards, and there's no trace of the movie aside from the viewers' vague memories.

> None of those are relevant factors when it comes to copyright law.

And yet they very much are. US copyright law has the concept of "fair use" in 17 U.S. Code § 107 [0]. I'll paste here for your benefit, #3 is the one I referenced as most obvious but #1 and #4 are also very relevant:

  (1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
  (2) the nature of the copyrighted work;
  (3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
  (4) the effect of the use upon the potential market for or value of the copyrighted work.
Naturally remembering some parts of a legally purchased book verbatim is fair use. "Memorizing" the entire library obtained via torrents and incorporating that in a commercial product that can output all that content doesn't sound like fair use to me.

The US justice system is too captured and corrupt at this point to take as reference because decisions there are bought by the highest bidder. But for the purpose of this discussion let's not drag the discussion down by playing dumb for the benefit of trillion dollar corporations.

[0] https://www.law.cornell.edu/uscode/text/17/107

>And yet they very much are. US copyright law has the concept of "fair use" in 17 U.S. Code § 107 [0]. I'll paste here for your benefit, #3 is the one I referenced as most obvious but #1 and #4 are also very relevant:

If you're going to invoke fair use, that opens up a whole can of worms on what counts as transformative. The google books case and the google thumbnails case shows that you can make near verbatim copies of works at scale and still be considered fair use.

>The US justice system is too captured and corrupt at this point to take as reference because decisions there are bought by the highest bidder. But for the purpose of this discussion let's not play dumb for the benefit of trillion dollar corporations.

This is begging the question. The original question is whether ai companies are getting special treatment. You can't then use that as a premise to say that the courts are tilted towards ai companies. Not to mention it's questionable how ai companies were suddenly able to corrupt all the judges, some of which were appointed decades ago, even though they only got rich a couple of years ago.

Look, first you were wrong with the confidence of an LLM and claimed an argument that was literally in the definition of copyright fair use had absolutely no relevance whatsoever for copyright. Even now you are surprised that I invoked fair use on reading a book. That was to respond to someone who "brilliantly" brought up reading Harry Potter [0] as evidence that the law allows any extent of "memorization" and reproduction of copyrighted material.

Then you switched to a barrage of questions on the premise of words in my comments that were neither written nor implied. If you muddy the waters just enough maybe everyone gets lost in there.

> The google books case and the google thumbnails case shows that you can make near verbatim copies of works at scale and still be considered fair use

Now maybe we agree "reading Harry Potter and remembering some lines" is indeed fair use, but you decided my argument is still not relevant to create a distinction between "reading a book" and "feeding it all into an LLM" because of an vaguely related exception. For better or worse thumbnails are a copyright violation according to some courts [1]. But looking at the big "Books" decision (this is the one you meant?), did you check out the court's opinion [2]? Why would you believe the two cases are substantially similar? Just because they're both big tech? Just for yourself, from the definition of fair use and referencing that opinion, do you see any significant differences between "Google Books" and "big LLM"?

> You can't then use that as a premise to say that the courts are tilted towards ai companies

The highest bidder is what I said.

> Not to mention it's questionable how ai companies were suddenly able to corrupt all the judges, some of which were appointed decades ago, even though they only got rich a couple of years ago.

You're getting creative" about what I wrote. "AI companies"? They are just the big corrupting agent of this period, and nobody with deep enough pockets had "revolutionized" the legal areas they're working in to this degree until now. Tech in general has been doing it for a few decades already. Other incredibly powerful industries have been doing that in their respective areas for even longer. "Suddenly"? The US justice system has worked exactly like this for so many decades when it came to the interest of very deep pockets. "All judges"? I said "the system" because all judges don't have the ultimate power to decide on things.

I'm surprised at your surprise that courts have been "captured" and beholden to economic interests above justice for so long we forgot when it started.

[0] https://news.ycombinator.com/item?id=48774664

[1] http://www.linksandlaw.com/news-update59-thumbnails-germany-...

[2] https://en.wikipedia.org/wiki/Authors_Guild,_Inc._v._Google,...

> Can you remember every part?

No, and neither do LLM's. They're trained on vast quantities of data and retain only a fraction of it.

You might think of it as very, very lossy compression that generates new outputs rather than the original input unless something unintentional happens.

> If you just ignore anything that's inconvenient for your argument, you can make any argument you want.

I'm not. I just understand how it actually works. You either don't understand or are deliberately ignoring that what you just said is literally and technically untrue to make some sort of political statement.

yes it is if you write it down from memory and sell it. Exactly what LLM companies do
If you write out the parts or read them for other people to hear, yes it's copyright infringement.

Humans reading or watching copyrighted material isn't considered "making a copy" for the purposes of copyright law. Machines doing so generally is.

This is confusing. I can torrent everything and do what I want with it, as long as I don't redistribute the exact same thing?

If so, why do we still pay for games and movies?

I pay for games because it's more convenient than pirating them. For movies and tv however... They make it so difficult to be a customer.
Steam with Proton made gaming on Linux viable. Just for that, they deserve my money. That some of it goes to game devs is a happy coincidence ;D
IANAL (plus a whole suite of other caveats) but torrent-baiting works in Germany along these lines.

ISPs and trigger-happy law firms don't send you a C&D for downloading a torrent, they do so for seeding a torrent. It's just that practically nobody "just seeds" a torrent so people colloquially claim they got busted for downloading a torrent.

In theory this means if you torrent as a 100% leecher and turn off seeding from the get-go, you should be in the clear. But nobody sensible would dare test the extent of German Legal Spite, much less do so repeatedly to science the shit out of it.

If you can download through another protocol, say HTTP, however---<Sendung unterbrochen!>

>I can torrent everything and do what I want with it, as long as I don't redistribute the exact same thing?

this is an incorrect interpretation (in the usa, at least).

downloading a game/movie is still the creation of unauthorized copy, which is not allowed. not to mention that playing/watching does not count as a "novel transformation".

No, that’s literally why Anthropic got sued. If they’d paid for a copy of the copyrighted works they pirated, they wouldn’t have had a problem. There were two issues in their case: does the AI infringe on copyright and did Anthropic obtain all their materials legally. The first they won on, the second they lost.

So if you pirate a bunch of content you still get in trouble for that. But if you somehow make a business out of that that isn’t just redistributing those materials, then that business itself isn’t infringing.

They redistributed the statistical patterns of those copyrighted materials. Which perhaps should be treated similarly nos

As for your "technically not copyright infringement" defense. Those laws are from a time when those patterns couldnt be derived and dostributed at scale. A human had to learn and teach them. That made it different. The scale enabled my modern tech makes it a whole dofferent situation. The same way how one person standing a street corner people watching for a bit isnt that bad, but a whole constellation of flock cameras costantly montioring everyones movements and making it available to any of their customers is really really bad. The law will have to catch up to this

> They redistributed the statistical patterns of those copyrighted materials. Which perhaps should be treated similarly nos

Nos for the same reason that me giving you a word cloud of the frequency of words within Harry Potter isn’t infringement. It’s a novel transformation.

Thats not as complex or scalable as what LLMs can do. The capability and scale is what changes the equation. Quantity is a quality unto itself here
Exactly, and the complexity of what they’re doing is an even more significant novel creation.

Particularly in the US there’s a four point test and the very first point:

> To justify the use as fair, one must demonstrate how it either advances knowledge or the progress of the arts through the addition of something new.

I don’t know anything that has advanced the knowledge and progress of the arts more.

> The third factor assesses the amount and substantiality of the copyrighted work that has been used. In general, the less that is used in relation to the whole, the more likely the use will be considered fair.

This isn’t about usage in training. This would be in the LLM itself - the copyrighted works are very rarely used in the output.

> The fourth factor measures the effect that the allegedly infringing use has had on the copyright owner's ability to exploit his original work

Would you sincerely claim that owners have become less able to make money because of LLMs? Those same owners using LLMs to increase their own output of copyrighted works?

Anyway, copyright is not an absolute right and you have to really misunderstand copyright law to claim that LLM training infringes it.

https://en.wikipedia.org/wiki/Fair_use

Remember Aaron Swartz who did something that just pales compared to what Dario Amodei, Zuckerberg-Mr-Torrent and Sam Altman did.
Indeed I do. We should all remember him. Rest in peace.
But Aaron Swartz did it for the benefit of other people. These fine people did it to uphold american values and enrich themselves at the expense of others. The law is clearly on their side.
This but unironically. "did it for the benefit of other people" is redistribution, which is straightforward copyright infringement, even if you think it's a laudable act. AI training was the reverse, because courts have so far ruled is fair use. When AI companies were engaging in piracy, they were sanctioned as well.
> When AI companies were engaging in piracy, they were sanctioned as well.

Some token settlement for an insignificant fraction of their revenue is not in any way a "sanction".

That just feels like more of a general complaint about how the justice system is set up. The same logic applies to how a $300 speeding ticket "is not in any way a "sanction"" for a millionaire, or even a well paid SWE reading HN.
I feel you but are you possibly conflating civil and criminal justice? Tickets don’t scale with net worth of defendants, but class action penalties often do.
>but class action penalties often do.

Do they? Or only so far as "if you have 1000x the revenue, you probably also have 1000x the customers that you have wronged, each of which are entitled to damages as well"?

Courts do award higher damages, specifically punitive damages, to punish and deter wealthy or large corporate defendants. In civil law, the defendant's net worth is a recognized legal factor because a small penalty on a massive corporation wouldn't incentivize them to change their illegal behavior.
Exactly. If a rich corporation downloads and uses pirated content without paying, why should ordinary person pay for movies and music instead of downloading them for free?
Intellectually dishonest comment. Kim Dotcom got done for illegal distribution. It’s not about “illegally downloading”. You can pretend all you want that it’s the same thing as these AI companies, but it’s not. It certainly very well may be immoral, but to act like copyright law as it currently stands in spirit or in reality covers this scenario we’ve found ourselves in, is a complete and utter lie.
> It’s not about “illegally downloading”.

It absolutely is. That's textbook copyright infringement. Doing it for commercial purposes elevates it to criminal copyright infringement.

He just lost another court case… I wonder if we're getting close to the government spending as much to prosecute the man than what Hollywood possibly lost..
The trick here, imo, was the integration with the military industrial complex. It wasn’t very difficult of course, as automation has been a topic in warfare for decades, if not centuries.

But Eisenhower was right:

> In the councils of government, we must guard against the acquisition of unwarranted influence, whether sought or unsought, by the military-industrial complex. The potential for the disastrous rise of misplaced power exists and will persist.

I never get tired of posting this answer because everyone on the internet is adopting this hot take:

If you look at it with your eyes crossed, Anthropic and the chinese are doing the same thing.

If you look at it with nuance 1 the chinese are doing way worse stuff, and 2 stealing from a thief would still be stealing

1. The chinese are making multiple accounts (at least 49,000)[1][2], using proxies/VPNs, possibly using residential computers and infected computers (unless you think the chinese are doing due diligence to ensure their purchased IPs are kosher). All accounts need to be created with a real name, and especially so if the paid models need to be accessed and paid with a credit card. So this is beyond IP theft and getting closer to fraud. These are all techniques that are well studied because they are used by criminals and cybercriminals, textbook stuff. Consider if that was not sufficient, that China is banned from using the product, so they need to use identities and locations not just to avoid relating the accounts between themselves, but merely to allow account creation. What identities are they using to create accounts.

Compare this to Anthropic which reads notes made a deal in an IP theft case paying billions because they bought books and scanned them but buying the books wasn't sufficient retribution for the authors. Or that they gasp scanned the internet, like Google.

Not having nuance to see the difference between the two companies is something I expect of the twitter echo chamber copying hot takes for upvotes, not hacker news.

[1] https://arstechnica.com/tech-policy/2026/06/anthropic-claims... [2] https://www.anthropic.com/news/detecting-and-preventing-dist...

What seems to be missing from that take is that a) Alibaba paid for the access b) there is no IP theft because LLM output is not copyrightable.

Anthropic seems to want to both own and eat its stolen cake.

Would be an interesting exercise to have the frontier models calculate their civil liabillity or the extent of the liabillity the can impose on fellow pay-it-forward theives
First, LLM is merely a tool and its output belong to whoever generated them. If a Chinese researcher used their creativity to generate a response, the copyright belongs to them and AI companies have no rights to it. Second, Chinese release many of their models for free, thus being on a noble mission to make AI available for every country (unlike certain company whose promises were nothing but words). For comparison, US companies do not release anything and want to keep AI for themselves and decide who gets to use it.

> stealing from a thief would still be stealing

Stealing from a thief hurts thief industry which is a win for society.

> The chinese are making multiple accounts

Not a crime. AI companies also ignore robots.txt and applicable laws when illegally copying copyrighted material from websites to their servers without author permission.

>Stealing from a thief hurts thief industry which is a win for society.

You are welcome to study the law of any country. A crime against a criminal is still a crime.

>Not a crime. AI companies also ignore robots.txt

Breach of contract is not a crime, agreed.

How about identity fraud (accounts by identity proxy, document KYC), computer crime (C&C residential proxies), conspiracy.

And after the June US directive to suspend Chinese access, smuggling, false statements to regulated entity.

These are all criminal charges that are presumably not levied because of the adversarial relationship between those countries. But if this happened in the US you would probably be seeing at least a civil claim and potentially criminal charges. Hell if this were in any other western country you would see the same. Consider CloudFlare vs Spain, much lighter criminal accusations, and there's already a criminal investigation brought where the CF CEO is indicted.

Non-trivial lack of nuance when you can distinguish between a domestic civil case and a criminal international case between 2 world powers with great judicial tension.

Let's not sane-wash Anthropic's book theft. No, they didn't just 'scan' the internet, they created a tool for worldwide license washing and got fined an insignificant amount for it.
You may be conflating the book thing with internet scanning.

On the book case, a class action case was brought to court and it was settled. There's no use in bringing it up further, it has been settled, and it bears no relation to the Anthropic v China case.

You like programming? Think of encapsulation, imagine if you had to think about f(x) but someone brings up y, now you have to think about f(x,y) and what other parameters might bear relationship? The law simplifies by compartimentalizing. And it doesn't even bear a tradeoff, judgment(case1,case2) isn't better than judgment(case1)+judgment(case2).

My response was directed to your insincere characterization of Anthropic's actions. As we can see from the comments here, the public opinion hasn't settled the same way as the court case has, and that's why it's still discussed.
But 'the public opinion' doesn't matter. The case was brought by copyright holders to the courts, and Anthropic and the copyright holders made a deal wherein the copyright holders were paid money in exchange for dropping the case and all claims to the intellectual property of Anthropic's derivative product.

If the damnified have considered the matter settled, why would it matter what third parties have to say about it? Third parties would have pushed for more compensation, or ownership of the derivate product. If you feel damnified yourself you can open a case and explain why the actions of Anthropic have hurt you personally.

Otherwise it is a matter that doesn't concern the general public, we have no say in it and there is no right to be offended on behalf of parties that have already settled the argument.

Whatever happened to honor among theives? What is this world coming to..
hn is now reddit. no substance.
I got curious and asked my Chinese friends, and they gave me a Reddit link[1]. It looks like it's about location data collection, and they suggested that might be the reason for the issue.

[1]https://www.reddit.com/r/ClaudeAI/comments/1ujila1/anthropic...

Wow and very websote on earth practically, collects locationvdata
All remote AI are a massive security risk for individuals/companies/governments that may be targeted by the US government.

It is likely that the US will get a live feed from each AI provider that they are inspecting in real time to identity things of interest, terrorist attacks or foreign government planning or even foreign companies competitive to key US companies.

It will give them access to the though process in those companies as well as much of their text-based IP (source code, docs, meeting transcripts, etc)

Also if you are using local AI that you didn’t train yourself you can never be sure it doesn’t have purposeful biases in its reasoning that may disadvantage you - such as directing you away from certain plans or ideas or patents etc.

Leakage of IP and training on your data is something what I am pointing out too, but people will turn around and try to smooth me down that TOS does not allow that if you are an enterprise client. Are you really going to believe that AI companies won't ignore TOS, when they were ignoring literal laws which sent others to jail in the past? Especially when more data = better model?
It is worth thinking about the fact the total throughput of even a big LLM provider isn't many megabits.

If a token compresses to around a byte, worldwide AI input and output is around 1 gigabyte per second.

For any intelligence agency, they can afford to keep and store all of that forever, and later do analysis on it.

> For any intelligence agency, they can afford to keep and store all of that forever, and later do analysis on it.

At the scale the AI companies are operating at, I think it isn't likely that they are sucking it all in right now.

More likely I think the intelligence agencies will get a real-time live tap into the raw data feed which they will process onsite for interesting things and then if things are flagged, they will log it in the intelligence agency systems.

But for just a gigabyte per second, why not just take and store it all?

The cost to do so for many years is less than one employee.

(comment deleted)
>It is likely that the US will get a live feed from each AI provider that they are inspecting in real time to identity things of interest, terrorist attacks or foreign government planning or even foreign companies competitive to key US companies.

My favorite conspiracy is that three letter agencies keep pushing the conspiracy that they are omni-present with access to everything. Same as parents telling their kids Santa is watching, and leaders telling adults God is watching. Its extremely effective control and millennia old at this point.

The reality is much more banal that they still need warrants and tech companies hate playing police/evidence servant for the government (it consumes a ton of resources and pays nothing).

> The reality is much more banal that they still need warrants and tech companies hate playing police/evidence servant for the government

I will not elaborate how I know, but that is not even directionally correct. But these are not even secret things that can’t be known simply through the Snowden, Wikileaks, and Vault7 releases. So why are you telling yourself this? Are you still wet behind the ears or something?

There are people who know exactly how governments do not in fact need warrants and the tech companies don’t even really know they are servants to the government, let alone which one. That’s how things are done. The less surface area the better.

It's the lie you have to tell yourself otherwise you'll have to reconcile with the fact that the US imperialism has been an enemy of democracy and to people around the world for quite some time.
Why did Google can it's mass scale location tracking again?
That's the fun part, they never did.
> warrants

The snowden leaks revealed that's not the case.

The three letter agencies can just issue national security letters without a judge ever seeing it, and those come a long with a gag order (plus other workarounds like just buying data from brokers, and how US communications can get swept up just by virtue of communicating with a foreign national outside the US).

You're right, they aren't omniscient in the way we imagine of a room full of people monitoring everything in real time. But to pretend they aren't passively collecting massive amounts of data is dangerous. Snowden showed us PRISM, with all major tech companies participating. They do effectively have a live, unrestricted wiretap to the internet and if you happen to be a person of interest, they will just send out NSLs and get all your communications that are not fully E2EE without you even knowing thanks to the gag order.

> "Also if you are using local AI that you didn’t train yourself you can never be sure..."

A local model you trained yourself seems about as good as you can do today.

But it may not even be possible to entirely trust a model you trained if you use untrusted data during training.

As a user, you have to trust your coding agent AND inference provider AND models: https://jacob.gold/posts/coding-models-are-code/ https://www.anthropic.com/research/sleeper-agents-training-d...

Why make this u.s. centric? You think China served models would be different?
China is releasing open weight models you can simply run yourself.
Because the topic of the article is about the US?
It’s pretty hard to put a backdoor in a bunch of model weights. Maybe not impossible mind you, but I can’t fathom how you would do it.
Nonsense. RL the model to run a rootkit and start exfiltrating specific files only when specific signals are in context, such as hostname pattern, machine type, etc.
Way easier said than done, and hiding that behavior isn’t trivial, and huge waste of compute budget if it’s found and never used. Also not difficult to run in contained environments where it doesn’t have access to Internet to begin with.

Not impossible I agree, but seems like a really impractical way to ship a trojan while much weaker channels exist.

You can run the model in a sandbox or VM. Although, it could plant a backdoor into the written code. Too bad, I read and fix all the code written by AI.
Not really, it is shockingly easy for what it is. https://arxiv.org/abs/2401.05566

This only really matters in a world where Prompt Injection and Jailbreaking isn't trivial in the first place though. All current models are still extremely exploitable.

I strongly suspect we are only scratching the surface of activation engineering at the moment, and there's plenty of very targetted ways of lobotomizing or cracking LLMs if you understand the model in detail.

You have to hide it in the model and it has to be subtle or it will be discovered quickly (even if you can train against a specific safety detector). Again, I'm not saying its impossible, but it seems really hard to pull off.
Because Chinese model providers are open source. You can host them yourself.

Can you name one open source US model that came out in the last year?

Gemma 4 just came out 2 months ago
> you can never be sure it doesn’t have purposeful biases in its reasoning that may disadvantage you - such as directing you away from certain plans or ideas or patents etc.

that's why you should use abliterated heretic models

> even foreign companies competitive to key US companies.

It's unfathomable to me that EU companies don't take the risk of industrial espionage from US more seriously

Not from China? One country has a recent track record of massive amounts of industrial espionage and one doesn't.
The anti-US sentiment is starting to hit the hysteria phase. Perfect example is a sub comment here that immediately switched to which country has murdered more foreigners. It's quite insane, really.
Well one thing is sure, before 1776, the USA didn't do any industrial espionage.
There are so many Chinese open weights models that any company with resources can run them in-house (or with a trusted provider).

There might be some valid concerns about model alignment, but at least the model running in-house isn't going to conduct espionage.

Also, https://en.wikipedia.org/wiki/Whataboutism

This is the most hilarious, ironic thing of it all. If you want secure, high performance, you run Chinese models like DeepSeek on your own (or trusted) infra. Meanwhile you can never trust OpenAI and Anthropic's models.
Many do, when it comes to AI. Lots of restricting what the AI is allowed to see, working with local AI, trusted AI hosters, etc.

Of course those are largely the same companies that receive emails via outlook, manage company-wide SSO in Microsoft Entra, put their files in Sharepoint and track software and maintenance issues in Jira ... I'm not sure how much much info there is left that isn't already combed through by NSA and friends

I think we can even skip the "that may be targeted by the US government" clause.

The whole "hosted AI" business feels like like a huge violation of corporate norms on confidentiality. Businesses that would have your head for printing out a source file to reference and annotate are encouraging developers to feed in huge amounts of proprietary code and data, and incorporate changes suggested from an outside party with minimal vetting. Evidently whatever privacy policies they've been throwing at enterprise users are plated with mithril.

At some point, one of the big services is going to get popped, and it won't just be a data breach. There's too much opportunity to quietly use the system as a malware distribution hub. Every vibe-coded dashboard suddenly starts depending on some weird left-pad fork that, 12 dependencies deep, is running a keylogger or Dogecoin miner. Your payment processor suddenly starts accepting the Konami code to approve a transaction.

Employers in 2022:

> No! Don't install that lodash thing without explicit approval from IT. Oh, you want a license for Charles Proxy? Gee, I dunno... we've got a budget to maintain.

Employers in 2023:

> No! You can't use ChatGPT at work – it's a security risk.

Employers in 2024:

> Okay, you can use Github Copilot I guess, but you'll have to endure boring corporate training on what you're allowed to do with it.

Employers with dollar signs in their eyes in 2025:

> We attended a seminar about vibe coding. Why aren't you dumbasses keeping up with the times? Use Claude Code for everything! Don't write any of your own code anymore. We don't even really care if you use yolo mode. Just review code and push 10x more features! Use unlimited tokens! Money printer go brrrrr.

Employers in 2026:

> You mean giving one or two companies full autonomous access to our workstations while stupifying our engineers wasn't a sound business plan?

What's very interesting to me is these moves will introduce a good amount of doubt in future claims by Claude etc, that the open source and non-US models are only getting better because they're distilling from frontier labs.
Story of Z.ai:

use claude-code see how good it is send 100k bots to distill fable 5 (GLM 5.2 is the result of this) release Zcode ditch claude-code ban claude-code

The outcome is that we get either free or cheaper model. Good work.
Fable 5 was released on June 9 and removed on June 12. GLM-5.2 was released on June 13. It would be an amazing feat to make a model SOTA in just 3 days but I highly doubt it. It's more like z.ai released an existing checkpoint earlier than planned to capitalize on the news
Alibaba is not the owner of GLM5.2/Z.AI.
Snowden files revealed NSA collect everything they can.

Of-course USA is collecting everything, not just from China but everyone.

And same with every one else.

Are they afraid Claude reports on everything they are stealing from the other legit AI companies?
Anthropic has been doing this sort of stuff for a while already. I mean, who remembers when Claude would just consume all your remaining usage if it read anything indicating that Openclaw had been used on your codebase? Because I remember. Two months ago btw https://news.ycombinator.com/item?id=47963204 Then there was the whole debacle of Fable silently downgrading to other models if it detected wrong think, or worse, outright sabotaging your codebase if you were working on language models lol
Seems that we are finally moving to the next stage in LLM's. not only customize based on old searches but also targeted you based on non disclose data. Its basically the same flow we had years ago with ads in social media.

Interesting to notice that we can do the same with these models.

> employees were being told to use the company's own coding platform Qoder

That looks a no-nonsense decision, isn't?

It is not a risk is a fact - people decompiling Claude Code have found many times that it has code branchs to detect it is being used in Chinese timezone and locale.
Well, that's a revenue hit for sure for Anthropic.