Hacker News

> "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 fully trust a model you trained if you used 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...

> 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

Why make this u.s. centric? You think China served models would be different?

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.

> 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

>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).

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 seemed pretty mild compared to what's collected by modern websites and apps, though? How many don't know your Timezone?

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.

It's insane that it's becoming a concern now. It should've ended the discussion from the very beginning.

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.

After they uploaded their code to private repositories on GitHub, Bitbucket etc since forever?. They trust GitHub not to read their code but they don't trust an AI from Microsoft not to read it? It would be schizophrenia

A bit too late for that, most of them have already dumped most of their codebase and IP into cloud models.

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

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.

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.

The US government deemed it necessary to snoop on foreign governments and citizens decades ago and is doing it on a continuous basis. Also on their own government and citizens.

There was a big thread about it here the other day. https://news.ycombinator.com/item?id=48734373

Wow and very websote on earth practically, collects locationvdata

2025 taught me that my employer would replace me with a slave if they could get away with it.

The confusing part to me is why these companies believed the "AGI" hype, I.E. that OpenAI or Claude's LLM is the ideal white collar slave.

I suppose I can understand that the executive class resents labor enough to make irrational business decisions for the purpose of insulting the workers who design and operate their companies.

That being said, the 2025 AI binge feels like a murder-suicide done by the executives of many of these companies.

Alibaba has engineers in Hongkong, Singapore, North America. It’s a global corporation

https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...

Same way every ban is evaded, smurfing

There is a reason Singapore tops the rank on Claude usage

Does Alibaba only have developers in the China?

Cc can be used with non Anthropic models.

The same way they buy "banned" and "sanctioned" NVIDIA GPUs.

> 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.

one possibility:

iproyal.com Oxylabs.io

https://krebsonsecurity.com/2025/10/aisuru-botnet-shifts-fro...

That’s not what a backdoor is…

I think most websites transmit general locationbto the server.

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.

Whatever happened to honor among theives? What is this world coming to..

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.

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...

The outcome is that we get either free or cheaper model. Good work.

[citation needed]

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.

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.

Yes, 100%. GLM 5.2 is capable of RSI. It's too late to stop.

Depends on a lab, but they do have plenty of compute and engineering. So this would only slow down the progress.

Of course, it is like any other kind of weapon system, eventually the knowledge gets acquired.

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.

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 I should care about it since I can't access it.

Probably yes.

More than a year ago, when Anthropic and OpenAI started to hide 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.