Value judgment aside: I am a bit surprised at how sloppily they did this. I think they could've achieved the same effect while decreasing the odds of detection via reverse engineering.
(This field is known as "underhanded code", coined by the Underhanded C contest: https://www.underhanded-c.org. It's a little-known "art"; little-known for probably self-explanatory reasons. There are much cleverer ways of achieving objectives like this. One obviously being you can move more out of the client and into the server, but the other being you can write plausibly deniable client code in a much more benign-seeming way than this. Some of what they added can only be done on the client, but I think some could've been moved, and the client-required parts could've been done more subtly and credibly.)
It's possible they knew the JS bundle gets so heavily scrutinized that it'd eventually get spotted and reported on regardless so they didn't bother doing something more subtle and duplicitous. But still seems slightly lazy.
It's also possible that there are more in-depth detection methods and that this was just a cheap and easy first step that hasn't been removed because it catches a lot of less sophisticated bad actors.
It's unlikely that this will stop a big AI lab from distilling their model if they're really determined, but A) it may be enough to stop a bunch of fly-by-night token resellers looking to make a quick buck and B) you never know when one person at one of those big labs will mess up and forget to install whatever workaround they have and out themselves.
I think of it like if you have a problem with birds in your yard so you go buy one of those plastic owls. The owl scares away most of the birds, but not all of them, so you go and buy some ultrasonic noise thing to scare them away (I'm just making something up). Just because you bought the new ultrasonic thing though, that doesn't mean you're going to take the owl down. You leave it up because now you've got two layers of defense instead of one.
Well considering how Claude is vibe coded, I can't say I'm really surprised by sloppiness at all. I've been moving more towards Codex and OpenCode not because the the anthropic models are bad, but because Claude seems to break something new and annoying every day.
I would guess this part - since it's so sensitive, and fairly small - was either written or heavily driven by humans. Though I do also think it's possible their internal Mythos ~5.5 or whatever may also not necessarily be heavily optimized for thinking in the right manner for highly effective underhanded code. (I think it's possible it is capable and they just didn't use it for this, for whatever reason, though.)
Watch out for the press release where Dario denies this was ever intentional, and it’s actually emergent behavior demonstrating that Claude wants to claim authorship of its works
Likewise, Reasonix harness for Deepseek gets me better performance for practically free, hitting the cache. And this is with an unsubsidized American provider.
They also could have been much more interesting in the approach. LLMs can use their token distributions to generate stegotext that read like plausible prose but decode to payloads.¹
At first I was agreeing with you, that this seemed like a sloppy way to implement this that was sure to be pretty quickly detected, but there is another possibility.
Anthropic could have implemented this not as a durable detection system against proxying resellers, but instead as a point-in-time sampling system to detect where (and with what context) proxying reselling is currently happening. Sure, it would be detected eventually, but in the meantime Anthropic could gain useful snapshot data.
I see your point, but in any case the more data / the less detectable, the better. But, yes, regardless of the exact motivation, I do think it's fairly plausible that they knew this would likely get detected fairly quickly no matter what and made a deliberate decision to not try to make it a super subtle, super clever insertion.
It's just the first layer and there are multiple layers underneath this that we don't know about.
As a side note, I have a pet theory that one of the reasons that OpenAI and Anthropic are okay with the latest models not being released is to prevent distillation.
I think they want to wait a couple months and see if the Chinese models continue to keep catching up or if their gains are really just because they're distilling the frontier models.
well if you ask claude how to implement something, you may not always get the optimal solution. this feels like something claude would spit back at you given a basic prompt
Have you looked into anything about Claude Code, how it’s configured, how it interacts with your system, etc? Because “sloppy” is a defining characteristic.
I finally bought Claude Pro (I am not coding etc these days so I just wanted to try it). The Claude desktop app is downright pathetic. I mean they could write a better one just with their own LLMs. What's stopping them?
so all we need is someone to leak a sufficiently large amount of claude generations onto the open and private web for all other LLMs to mimic the same marking style?
wouldn't this happen due to the massive amounts of spam/slop being released?
It’s even more funny how this blew in their faces. They even advertised pretty much all providers on hackernews home page. Here is in case you missed in the article
The site collection seems pretty random. There's a mix of actual AI labs, extremely questionable resellers (like whatever "claude-opus.top" is), and then random consumer sites like baidu and xiaohongshu.
In addition, many Chinese companies are trying to give their programmers access to Anthropic models even though they're legally prohibited from doing so. And that might involve employees using unmodified Claude Code with an ANTHROPIC_BASE_URL pointing to a proxy on the company intranet. In Alibaba's case, I've been told by an employee that they went the extra mile of setting up a hermetic cloud environment where employees could indirectly use Claude Code without ever having it touch their work computers.
You have an odd definition of "blew up in their faces". What, do you somehow think your average Claude Code user on HN is going to think "Oh wow, I'm sure I'll get a much better experience if instead of going to the standard Anthropic Claude API endpoint I go through xiaohongshu.com."
I mean, yes? I heard of these Chinese resellers like a week ago and put it on the TODO pile due to a lack of leads. Now I'm gonna go trough the list and see if there's any I find acceptable.
If enough Westerners start using the service someone will make a website more anglo-friendly.
It’s not surprising at all, they’re vibecoding Claude code so of course they are not going to get anything other than slop out of it. A novel or clever solution is just out of the question for them.
Codex CLI is FOSS, unlike Claude Code, so Codex is less likely to do things like that, and it's one more reason to avoid Claude Code and Claude in general. Hopefully, many eyes will be looking into Codex for malicious things like that.
Genuine question though, why would I care about this if I'm paying for a subscription and adhering to TOS. I'm very skeptical about their privacy policy, business practices, and so on, but am curious what the negative about this is. Seems like it would work to my favour as a customer pushing back any date of the cutting of subsidies.
That said, these fraudulent proxies are helping Chinese labs keep up, which might be to my advantage long term in eventually having a high quality private AI I fully control on my own hardware. That's not support, but I do recognize the incentive, for whatever that's worth.
It's released and signed by GitHub I believe (although not deterministic builds), but there's at least a little bit of provenance that you're getting the real repository.
The conclusion of this blog post is a bit hysterical. The intent of this steg is excruciatingly clear (identifying usage by Chinese firms that may be conducting model distillation). It's unclear on how this "punishes normal developers" in any shape or form.
I don't understand the privacy concerns the author is trying to highlight. Granted, doing anything "sneaky" will always raise suspicious once caught, but on the other hand, there would be no point in implementing these "security features" if they were upfront about how they work.
And no, IMO stenography isn't security by obscurity, in the same that using RSA and keeping the private key private isn't security by obscurity - keeping the private thing private is part of the security model.
If the countries were reversed, and some Chinese software implemented an equivalent "security feature" to track US users, it would be all over the news about how China is conducting spying and espionage on America.
Or maybe you don't understand this hypothetical situation either, but I'm suspecting you just don't care about other people's privacy.
Seeing as how Anthropic cannot stop raising a stink about "illicit Chinese distillation attacks" every month or so, I'd bet money on them either already silently degrading model performance if any of the identification patterns match, or, at the very least, considering it/doing dry runs.
Particularly considering that they've openly stated that the technology to do so exists and that they were going to use it in production on Fable.
Can somebody clarify for me - if ANTHROPIC_BASE_URL is set to a different provider... then isn't this "marked" system prompt being sent to that provider's API rather than Anthropic's?
I understand how this can be useful to Anthropic if the 3rd-party is acting as a proxy (because they end up hitting the Claude API with the marked prompt), but it looks like requests where "hostname contains deepseek" would never be sending data to Anthropic. What am I missing?
Won’t catch many after has been on hn home page. And now the providers will be even more careful to upgrade the cc code. Might even provide their own agent to prevent this mockery. And isn’t what anthropic did unauthorized use of another pc which is kind of illegal?
Thats the thing, hoping to control things on client side like this is a lost battle if you are dealing with technical clients. The best they can do is probably based on IP, but again the motivated clients would just create bastion servers in allowed IP ranges. I am surprised why are they even throwing resources in this kind of effort.
My guess is for distillation, they need to forward the prompt to Anthropic to get the real Anthropic model's response so they can train their own models on it
The theory is probably Deepseek might be collecting those streams, and sending a portion of it to Anthropic to see what the Anthropic/Opus response would be.
I am also really confused and annoyingly stuck on this. I understand that the model name might appear in prompts for distillation (I guess? "You are RipOffModelv2, learn from these responses from Claude")?
I guess the only explanation is that there's a side-telemetry channel that still sends some data to Anthropic, regardless of ANTHROPIC_BASE_URL overrides.
> I understand that the model name might appear in prompts for distillation (I guess? "You are RipOffModelv2, learn from these responses from Claude")
This does not make sense. You wouldn't send such a prompt to the Claude model. And when you're sending the prompt (anywhere) you don't have the response yet. This is not how distillation works.
This is very interesting. Combating resellers and distillation seems like a very difficult problem indeed. Interesting to me is that these techniques mentioned in the article are just like anti-observation techniques used by some of the more sophisticated malware out there, however defeating them is pretty trivial.
Yes, defeating this is relatively easy, particularly for sophisticated actors. But it's hard to always defeat all of the tricks. Sort of like how it's expensive and hard and uncertain to defeat all of the tricks when forging money.
Here's an example. Say you have your team use patched binaries. Then CC updates and requires a new patched binary with new tricks. You now have to have a team ready to analyze the binary and begin to address the tricks; meanwhile, unpatched code is now a fingerprint. If some researcher decides to update Claude on their own to access new features, they get fingerprinted.
Defeating a single fingerprinting technique once is easy. Defeating all of the techniques all the time is hard.
I'd love for you to try this and report back. My guess is that no models today will successfully run a binary analysis for fingerprinting without a lot of handholding. If you try to use Opus it will almost certainly decline (and fingerprint/ban you).
Can you share more details? I ask because my experience suggests that models still require a decent amount of expertise to use for binary analysis (largely inferring because of use on other tasks of this level). I would expect models to always find "something" when you ask for stenographic techniques in the code, but with an extremely high false positive rate.
I don't think the diffs between Claude releases are that big. The amount of code in a diff doing sketchy stuff like looking into the host environment is going to be pretty small and obvious for the model. You can do things like ask for what an update included that wasn't mentioned in the release notes and stuff like that.
If they only collect the data for analysis I guess this is fine (they already get way more sensitive data from users anyways, so if privacy is your concern you've made the mistake many steps ago). The much more interesting question is if they directly act on this data in their API. For example by rate-limiting, compute-limiting or rerouting to weaker models. That might even be legally questionable. I would really like to see this as a follow-up analysis, but I guess it is way more difficult and will also cost quite a bit in tokens.
"If they only collect the data for analysis I guess this is fine"
I think you missed the memo on how foolish this attitude is. It came out around the time Edward Snowden made his discoveries at the NSA public. I suggest you look into it
I've heard that it was possible to trigger really obvious output poisoning on Fable with something as basic as asking the model to think outside of its built-in hidden thinking delimiters.
None of this is surprising - they're trying to mask and relay when they detect known patterns of what looks like distillation attacks and client app copying/modification. The list obfuscation here is likely to prevent or make it difficult for those same adversaries to work around this or delete/null it out when making a bootleg copy.
Cool reverse engineering/analysis report but if this is the extent of nefarious activity that came of it (trying to catch/mitigate chinese lab model distillations), that's kind of encouraging.
There has been an anti anthropic propaganda push by bad actors across social media sites especially Reddit and twitter. This started a few months ago when anthropic started beating openai.
What's the point of even trying to obfuscate this with such a simple method? Could at least have hidden the targeted features by storing their hashes or embedding a bloom filter or similar
In this case, this is probably not the only stereographic tattletale.
Had a competitor pull something like this with a previous employer. They were supposed to be interoperating with a standard, but they had a secret steganographic handshake, which they used to pretend that competitors products were unreliable (they had a first mover position in a smaller national market with specific requirements, so this wasn't shooting themselves in the foot). Our guys figured out the handshake and just silently implemented it. In this case, the competitor wasn't big enough to waste engineering time on multiple such hacks, but Anthropic have time (or Claude does).
(This sounds like a clumsy way of catching the Chinese that easily can be side-stepped.)
Claude Code has more or less full access to the client computer. The server (that hosts the actual AI) can just go: execute this payload and tell me the result - otherwise I won't answer any further questions or re-route you to a stupider model.
The payload could check for Chinese time-zones, scan for copies of the little red book on the local hard-drive, or ping truth.social to see it was behind the great firewall.
> Claude Code has more or less full access to the client computer.
It shouldn't, not if you run CC as a separate unprivileged user. I wouldn't run CC on my main user account with sudo and access to my home directory or other resources. This is what the UNIX permissions system was designed for.
> "That also means the client itself deserves scrutiny. If a coding agent can read your repo and run commands, the binary that ships it should be boring (ƒor example, pi harness)"
the Chinese they are trying to catch must be amateurs, first thing you should do is construct a sandbox which looks indistinguishable from a common user. second thing is to put it behind a residential proxy.
Dario's been openly talking how worried he is about China and labs getting synthetic training data off their models, for years. Most recently in relation to "Mythos level" capabilities.
Not really distillation, just synthetic training data.
After loving Claude Code for most of its lifetime, I've been extremely annoyed by every change in the past months, even on the model level.
There seem to be all sorts of continual under-the-cover changes like this one that make life harder. It feels like the entire product has been taken over by overly ambitious PMs that care more about making their mark than in improving the experience, and all of their marks have made me less productive.
I've been using Pi with GLM5.2 the past few days, and though it's expensive, I find it far more productive and less annoying. The remote session plugin is far more reliable, I don't need to intuit some undocumented usage pattern to figure out how to use it well, and it just works.
curious for those with experience - what do people prefer about Pi vs. opencode alternatives? i've mostly been using pi as well but not out of any principled decision
> I've been using Pi with GLM5.2 the past few days, and though it's expensive
are you using the API for glm 5.2 or how exactly is it more expensive? How is GLM5.2 more expensive than using Claude code, that doesn't line up to my experience but to be fair I am on an older yearly subscription which generously only has 5 hour limits.
To be fair though one minor criticism of GLM 5.2 that I have is that it does seem to overthink quite a lot sometimes but the results end up being (good?),
I personally have used Glm 5.2 with (Opencode + obra/superpowers) / Oh-my-pi / Maki.sh
I like the 1st one when I am doing a longer project, the 2nd or 3rd one when I am doing a project which doesn't want me to ask too many questions and simply spin me up something. I sometimes use free online interfaces of claude and gemini and others like AIstudio for that as well which surprisingly can lead you to go far as well.
Overall, I am decently happy with the state of Open-source models actually and the eco-system around it is probably gonna have even more innovation surrounding it.
This seems really, really stupid. Similar to the weird Zig runtime signature thing from a few months ago ago, it was bound to be discovered, quickly, and all the resellers have to do is find a new domain name that (checks notes) doesn't have the word DEEPSEEK in it. Like, seriously? Your goal was to identify resellers by checking if the proxy has the corporate name of one of your competitors in it? Is this amateur hour?
All Anthropic has done is reduce trust, once again, with legitimate customers, while doing nothing to stop illegitimate customers. They need to get adults into key leadership roles, quickly.
The AI race right now is in a sad state. Chinese's playbook is releases open weight models and trains them on their own chips.
Anthropic pushes fear and control. But the only way to win is by innovating. China is flooding the market with cheap, good enough models, while the U.S. is building a Chinese firewall.
Frankly, I don't see this as the concerning behaviour the article describes.
It is fine to try to protect against distillation through a technique like this.
This will also allow them to, instead of blocking the distillation agents, respond with a poorer result/model, hindering the progress of distillation, momentarily at least.
I would guess that's their first line of defense; they should have more techniques to identify distillation because that's a very simple way of detecting the host and can be easily spoofed.
1st, this technique is not fraud, and fraud is a separate accusation. 2nd, paying customers can legally and legitimately be banned and monitored for breaking terms of service, which probably includes things like using the model against U.S. export restrictions.
Steganography is, essentially, hiding information within another message, such that it's not readily apparent that the message contains the information.
I used Claude Code for a month because my boss gifted me a sub and wanted me to try it.
I used that month to complete a work project and then beef up my personal harness so I'd never have to deal with Anthropic (and these sorts of shenanigans) again.
The issue is that using Claude Code is an easy compromise for most to make, when you get to use the models 10x cheaper than through API pricing with a custom harness.
Build it from scratch. Understanding fundamentals of how agentic coding harnesses is a must though if you gonna go that route. I think everyone should take time and learn these things, maybe reverse engineer Codex Cli or something like that as a starter. That info is very valuable in this day and age.
Can you say more about Codex? I'm using GPT-5.5 in my own harness and it's not liking it very well, so I'm thinking I ought to make it more Codexy so it's more ergonomic for it. (edit format, tool calls etc.) But haven't gotten around to it yet.
Not the comment author, but I use pi and customize it with my own extensions. Pi automatically tells models how to customize itself, so it's a pretty easy process.
I started mine from scratch in 2023 because I wanted to use LLMs from a terminal and there was nothing else compelling at the time (nowadays there is pi and opencode)
Harnesses are/can be incredibly simple things, not much more than a HTTP client that renders things in a way that suites your taste.
It’s not that difficult, it’s just a system prompt and a set of basic file edit/bash/etc tools.
Me, personally, I didn’t build it from scratch but I ported original CC from published sources into Python and extended it to match my own requirements.
I use GLM in my custom harness. It completes the same tasks at the same level of quality, except 8x faster and 8x cheaper. (Same goes for GPT!)
I'm not sure how that's possible. I expected to get increased correctness for that order of magnitude (something something test-time compute!) but I am not getting it.
When have they ever been credible? They have always been shady with their talk of safety, Dario was the one who wrote back in 2019 that GPT 2 was too dangerous to release.
I don't think many people care that they are trying to detect resellers and distillation.
It also doesn't seem very consistent to fixate on that while sending Anthropic everything about you via your day to day prompts, every line of the projects and environments you're working on at work, etc.
Their credibility comes from having one of the best models.
It has some good effects on the their models, like Claude seeking cooperation first. But the people behind the company have a typical "unconstrained" (in the Sowell vision sense) perspective that assumes that they know better, so they are righteous for attempting to control things (users, paying customers, their model outputs, their tool chain, the supposed deity they assume they will produce... etc.)
I self-host DeepSeek V4 Flash on 2 DGX Sparks (approx. $10k)
I expect DeepSeek V4 Flash (or an equivalently sized model) to reach parity with GLM 5.2 some time this year (this based on DeepSeek V4 Flash launching at GLM 5.0 parity[0], and GLM 5.2 being freely available to distill from)
GLM 5.2 is within spitting distance of Opus 4.8 and is at least as good as Opus 4.6[1] which some devs were willing to spend hundreds to single-digit thousands of dollars a month for a few months ago.
Headline is, frankly, awful. This isn't the AI secretly doing stuff and hiding it. This is the very human Anthropic engineers trying to detect Chinese scraping via some frankly hamfisted and unimaginative URL trickery.
I didn't assume it was the AI, just that some part of the the overall Claude Code product was doing this. I didn't assume the feature was added to Claude Code without human oversight. If it was added by Claude-the-AI itself without the humans prompting it to I would still hold the humans at Anthropic responsible. Does that make you feel better?
Here's the sha of the prompt I submitted... no I don't know why there are no saved prompts with that sha.
What do you mean you don't know where the bug is coming from?
No, I absolutely didn't make it up, how could you accuse me of that?
Does anyone know when this regex isn't working? I double checked it 27 times, I even asked the LLM. They all say this regex should be finding these dates.
Weird, suddenly all the conversations are breaking when I feed them into this other tool? Something about UTF-8 errors, but I'm sure I'm only using ASCII?
I do try to take care to make sure the things I build can be used by other people even when they care about different things. I care about understandably, determinism (as it relates to computing), and repeatability (because I want to be able to trust the systems I use).
If y'all would be willing to try to account for use cases of others, and try not to break them... that would be nice.
Please note: that generally when you modify something that belongs to someone else without telling them... things should be expected to break.
Would you also say that "someone who wants to use an IDE / LSP features to code and not give credit to the IDE / LSP is the worst kind of person"? If not, what is the difference between the two for you?
> one wrote code while the other is used by meatbags to write code.
One is not a "meatbag" while the other is not a "meatbag". And no, outputting something on stdout that happens to function as code is not "writing" it in the sense that we actually care about here. That's conflating the metaphor we use in describing program behaviour with the actual "meatbag" activity.
> why is this example always marched out like it means something?
again, that's not what we are talking about here. we have humans writing code using an IDE. we have LLMs generating code that is placed in the IDE. why are people obtuse to this? why are bots obtuse to this?
> Let's start this out right: if they're equivalent, first you explain to us why you think so.
I think it should be really obvious how they're equivalent: both are the result of a program running on a computer, and not the result of in-the-moment cognition by a moral agent or moral patient. Of course the LLM is just a tool. Models can literally be downloaded as ordinary files. There is not some threshold to cross where some configurations of bits on a disk deserve "credit" for work and others do not.
> I think it should be really obvious how they're equivalent: both are the result of a program running on a computer...
In fact it's really obvious everything is equivalent: it's all just matter and energy!
> Of course the LLM is just a tool. Models can literally be downloaded as ordinary files. There is not some threshold to cross where some configurations of bits on a disk deserve "credit" for work and others do not.
Of course there is such a threshold. And it's definitely been crossed when the "tool" can operate autonomously or nearly so, when it can generate the "creation" with minimal operator input or understanding.
Your classic IDE can't do anything without the detailed control of its operator. It's nothing like a coding agent.
I just don't agree that it's a false equivalency. I see them both as "tools I use to get the job done". For me, the job is not "writing code" - it is "deliver feature", "fix bug", and the accountability, responsibility, and communication that comes with it.
> I just don't agree that it's a false equivalency. I see them both as "tools I use to get the job done". For me, the job is not "writing code" - it is "deliver feature", "fix bug", and the accountability, responsibility, and communication that comes with it.
> If scrapping content is legal, model distillation should be legal too.
No, because legality should be determined by what's in the best interests of Athropic and OpenAI's business models.
Hopefully they're working on RLHF their models to insert clauses making that reality clear into any legislation their models generate or review. That way it's only a matter of time until the confusion is cleared up.
I suppose model distillation is technically legal, in terms of copyright, because LLM output is automatically public domain.
It's only "illegal" from a standpoint of breach of contract given its against the terms of use/service, which is to say its not illegal at all, there's no criminality there.
Yeah I considered whether I should use the term "illegal" in my original post, but in this case, I believe these models are actually banned for use in China, right? Like there are probably export controls (at least with the NVidia chips)
I honestly don't know ... yeah if it's just technically a terms of use violation (which isn't illegal, just a violation of one company's rules, for which Anthropic has every right to stop), or do we now have export controls applied from the various government actions, etc making them truly illegal now.
we have global export controls on Fable/Mythos, and I think (but I'm not 100% sure) that other frontier models are illegal for a US company to provide to China. So Anthropic geoblocks it, but unlike Mythos/Fable, non US citizens can still use Opus, etc just not from within China.
But because of the public domain status of LLM output (in the US) I'm not sure paying someone to run a bunch of prompts through Claude, post the output on a public website and then have a lab in China pull that output, would run afoul of any laws I think that would be legal on technicality. AFAIK Anthropic has no ban in its terms of use that you can't share Claude's output publicly. You still need interactivity for distillation, but I don't think (for now) there's anything stopping a Chinese or other lab from sending people to the US, signing up for a Claude subscription and doing the work state side.
Distillation is pretty much impossible to stop. The US GOV would have to go the full export controls route like they did for Fable/Mythos to stop any non-US citizen from using/accessing the model, which is going to be impractical if not impossible to enforce.
There are so many China born Chinese employees at Anthropic and OpenAI and I think quite a lot of them have already been recruited as spy . So it is almost impossible to keep secrets from Chinese government.
At what point though doesnt somebody stand back and say "wow, thats really dumb!" I think its probably more an indication of a dev having too much time on their hands rather than being in a hurry.
nous research. started out making overhyped llama finetunes, now they got a great agent harness and a cutting edge distributed training network that actually works.
Value judgment aside: I am a bit surprised at how sloppily they did this. I think they could've achieved the same effect while decreasing the odds of detection via reverse engineering.
(This field is known as "underhanded code", coined by the Underhanded C contest: https://www.underhanded-c.org. It's a little-known "art"; little-known for probably self-explanatory reasons. There are much cleverer ways of achieving objectives like this. One obviously being you can move more out of the client and into the server, but the other being you can write plausibly deniable client code in a much more benign-seeming way than this. Some of what they added can only be done on the client, but I think some could've been moved, and the client-required parts could've been done more subtly and credibly.)
It's possible they knew the JS bundle gets so heavily scrutinized that it'd eventually get spotted and reported on regardless so they didn't bother doing something more subtle and duplicitous. But still seems slightly lazy.
It's also possible that there are more in-depth detection methods and that this was just a cheap and easy first step that hasn't been removed because it catches a lot of less sophisticated bad actors.
It's unlikely that this will stop a big AI lab from distilling their model if they're really determined, but A) it may be enough to stop a bunch of fly-by-night token resellers looking to make a quick buck and B) you never know when one person at one of those big labs will mess up and forget to install whatever workaround they have and out themselves.
I think of it like if you have a problem with birds in your yard so you go buy one of those plastic owls. The owl scares away most of the birds, but not all of them, so you go and buy some ultrasonic noise thing to scare them away (I'm just making something up). Just because you bought the new ultrasonic thing though, that doesn't mean you're going to take the owl down. You leave it up because now you've got two layers of defense instead of one.
Well considering how Claude is vibe coded, I can't say I'm really surprised by sloppiness at all. I've been moving more towards Codex and OpenCode not because the the anthropic models are bad, but because Claude seems to break something new and annoying every day.
I would guess this part - since it's so sensitive, and fairly small - was either written or heavily driven by humans. Though I do also think it's possible their internal Mythos ~5.5 or whatever may also not necessarily be heavily optimized for thinking in the right manner for highly effective underhanded code. (I think it's possible it is capable and they just didn't use it for this, for whatever reason, though.)
Watch out for the press release where Dario denies this was ever intentional, and it’s actually emergent behavior demonstrating that Claude wants to claim authorship of its works
Sounds like clear evidence that AI is dangerous and totally needs to be regulated, guys.
Likewise, Reasonix harness for Deepseek gets me better performance for practically free, hitting the cache. And this is with an unsubsidized American provider.
Claude Code are slopmaxxxing and you're considering their "judgement"? :-)
They also could have been much more interesting in the approach. LLMs can use their token distributions to generate stegotext that read like plausible prose but decode to payloads.¹
¹ https://github.com/hodgesmr/calgacus-mlx
Sure, but the point here is to add a fingerprint from the client.
At first I was agreeing with you, that this seemed like a sloppy way to implement this that was sure to be pretty quickly detected, but there is another possibility.
Anthropic could have implemented this not as a durable detection system against proxying resellers, but instead as a point-in-time sampling system to detect where (and with what context) proxying reselling is currently happening. Sure, it would be detected eventually, but in the meantime Anthropic could gain useful snapshot data.
I see your point, but in any case the more data / the less detectable, the better. But, yes, regardless of the exact motivation, I do think it's fairly plausible that they knew this would likely get detected fairly quickly no matter what and made a deliberate decision to not try to make it a super subtle, super clever insertion.
It's just the first layer and there are multiple layers underneath this that we don't know about.
As a side note, I have a pet theory that one of the reasons that OpenAI and Anthropic are okay with the latest models not being released is to prevent distillation.
I think they want to wait a couple months and see if the Chinese models continue to keep catching up or if their gains are really just because they're distilling the frontier models.
Dunno, it seems like the exact kind of thing Claude would think up if you asked it to subtly alter the system prompt to hide this info.
It's all a losing battle anyway.
I've seen Eve Online corporations that do a better job of steganographic marking than this.
well if you ask claude how to implement something, you may not always get the optimal solution. this feels like something claude would spit back at you given a basic prompt
Have you looked into anything about Claude Code, how it’s configured, how it interacts with your system, etc? Because “sloppy” is a defining characteristic.
I finally bought Claude Pro (I am not coding etc these days so I just wanted to try it). The Claude desktop app is downright pathetic. I mean they could write a better one just with their own LLMs. What's stopping them?
That's … exactly what they're doing. This is the outcome.
so all we need is someone to leak a sufficiently large amount of claude generations onto the open and private web for all other LLMs to mimic the same marking style?
wouldn't this happen due to the massive amounts of spam/slop being released?
It’s even more funny how this blew in their faces. They even advertised pretty much all providers on hackernews home page. Here is in case you missed in the article
‘’’ cn baidu.com alibaba-inc.com alipay.com antgroup-inc.cn bytedance.net kuaishou.com xiaohongshu.com jd.com bilibili.co iflytek.com stepfun-inc.com moonshot.ai anyrouter.top claude-code-hub.app claude-opus.top openclaude.me proxyai.com yunwu.ai zenmux.ai
‘’’
You can view the full list here: https://cdn.thereallo.dev/blog/assets/cc-domains.js
const knownDomains = [ "cn", "sankuai.com", "netease.com", "163.com", "baidu-int.com", "baidu.com", "alibaba-inc.com", "alipay.com", "antgroup-inc.cn", "kuaishou.com", "bytedance.net", "xiaohongshu.com", "ctripcorp.com", "jd.com", "jdcloud.com", "bilibili.co", "iflytek.com", "stepfun-inc.com", "aliyuncs.com", "cn-shanghai.fcapp.run", "cn-beijing.fcapp.run", "xaminim.com", "moonshot.ai", "anyrouter.top", "packyapi.com", "aicodemirror.com", "aigocode.com", "hongshan.com", "iwhalecloud.com", "dhcoder.net", "lemongpt.top", "zhihuiapi.top", "intsig.net", "high-five-ai.xyz", "cloudsway.net", "4sapi.com", "529961.com", "88996.cloud", "88code.ai", "88code.org", "91code.pro", "992236.xyz", "ai.codeqaq.com", "ai.hybgzs.com", "ai.kjvhh.com", "aicanapi.com", "aicoding.sh", "aifast.site", "aihubmix.com", "anmory.com", "api.5202030.xyz", "api.ablai.top", "api.bianxie.ai", "api.bltcy.ai", "api.cpass.cc", "api.dev88.tech", "api.dreamger.com", "api.expansion.chat", "api.gueai.com", "api.holdai.top", "api.ikuncode.cc", "api.lconai.com", "api.linkapi.org", "api.mkeai.com", "api.nekoapi.com", "api.oaipro.com", "api.ruyun.fun", "api.ssopen.top", "api.tu-zi.com", "api.uglycat.cc", "api.v3.cm", "api.whatai.cc", "api.wpgzs.top", "api.xty.app", "api.yuegle.com", "api.zzyu.me", "apimart.ai", "apipro.maynor1024.live", "apiyi.com", "applyj.hiapi.top", "augmunt.com", "b4u.qzz.io", "clauddy.com", "claude-code-hub.app", "claude-opus.top", "claudeide.net", "co.yes.vg", "code.wenwen-ai.com", "code.x-aio.com", "codeilab.com", "cubence.com", "deeprouter.top", "dimaray.com", "dmxapi.com", "docs.aigc2d.com", "duckcoding.com", "fk.hshwk.org", "flapcode.com", "foxcode.hshwk.org", "foxcode.rjj.cc", "fuli.hxi.me", "getgoapi.com", "gpt.zhizengzeng.com", "gptgod.cloud", "gptkey.eu.org", "gptpay.store", "hdgsb.com", "henapi.top", "instcopilot-api.com", "jeniya.top", "jiekou.ai", "kg-api.cloud", "n1n.ai", "new-api.u4vr.com", "new.xychatai.com", "one-api.bltcy.top", "one.ocoolai.com", "oneapi.paintbot.top", "open.xiaojingai.com", "openclaude.me", "opus.gptuu.com", "poloai.top", "poloapi.top", "privnode.com", "proxyai.com", "qinzhiai.com", "right.codes", "runanytime.hxi.me", "sssaicode.com", "store.zzyus.top", "tiantianai.pro", "uiuiapi.com", "uniapi.ai", "vip.undyingapi.com", "wolfai.top", "wzw.de5.net", "wzw.pp.ua", "xairouter.com", "xaixapi.com", "xiaohuapi.site", "xiaohumini.site", "xy.poloapi.com", "yansd666.com", "yansd666.top", "yunwu.ai", "yunwu.zeabur.app", "zenmux.ai", ];
const labKeywords = [ "deepseek", "moonshot", "minimax", "xaminim", "zhipu", "bigmodel", "baichuan", "stepfun", "01ai", "dashscope", "volces", ]
rhoooo - so this is where to go to get cheap Claudeo at 90% off the listing price!
The site collection seems pretty random. There's a mix of actual AI labs, extremely questionable resellers (like whatever "claude-opus.top" is), and then random consumer sites like baidu and xiaohongshu.
Baidu has an actual AI lab: https://huggingface.co/baidu So does Xiaohongshu: https://huggingface.co/rednote-hilab Pretty much every Chinese internet company seems to have an AI team nowadays, however small.
In addition, many Chinese companies are trying to give their programmers access to Anthropic models even though they're legally prohibited from doing so. And that might involve employees using unmodified Claude Code with an ANTHROPIC_BASE_URL pointing to a proxy on the company intranet. In Alibaba's case, I've been told by an employee that they went the extra mile of setting up a hermetic cloud environment where employees could indirectly use Claude Code without ever having it touch their work computers.
You have an odd definition of "blew up in their faces". What, do you somehow think your average Claude Code user on HN is going to think "Oh wow, I'm sure I'll get a much better experience if instead of going to the standard Anthropic Claude API endpoint I go through xiaohongshu.com."
I mean, yes? I heard of these Chinese resellers like a week ago and put it on the TODO pile due to a lack of leads. Now I'm gonna go trough the list and see if there's any I find acceptable.
If enough Westerners start using the service someone will make a website more anglo-friendly.
It’s not surprising at all, they’re vibecoding Claude code so of course they are not going to get anything other than slop out of it. A novel or clever solution is just out of the question for them.
Codex CLI is FOSS, unlike Claude Code, so Codex is less likely to do things like that, and it's one more reason to avoid Claude Code and Claude in general. Hopefully, many eyes will be looking into Codex for malicious things like that.
Genuine question though, why would I care about this if I'm paying for a subscription and adhering to TOS. I'm very skeptical about their privacy policy, business practices, and so on, but am curious what the negative about this is. Seems like it would work to my favour as a customer pushing back any date of the cutting of subsidies.
That said, these fraudulent proxies are helping Chinese labs keep up, which might be to my advantage long term in eventually having a high quality private AI I fully control on my own hardware. That's not support, but I do recognize the incentive, for whatever that's worth.
It's released and signed by GitHub I believe (although not deterministic builds), but there's at least a little bit of provenance that you're getting the real repository.
But wasnt claude code leaked? Why wasnt this found earlier?
It doesn't take long for them to vibe code new features for CC
Or vibe code it completely differently. After all, they have basically unlimited access to best models with maximum speed if they just wanted to.
This specific form of steganography was not present when the leak happened, as far as I can tell.
I reported a similar system prompt injection mechanism here:
https://news.ycombinator.com/item?id=48259288
https://github.com/anthropics/claude-code/issues/62061
Looks like they just keep finding new "creative" uses for such things, as expected. I'll keep patching them out.
The conclusion of this blog post is a bit hysterical. The intent of this steg is excruciatingly clear (identifying usage by Chinese firms that may be conducting model distillation). It's unclear on how this "punishes normal developers" in any shape or form.
[delayed]
I don't understand the privacy concerns the author is trying to highlight. Granted, doing anything "sneaky" will always raise suspicious once caught, but on the other hand, there would be no point in implementing these "security features" if they were upfront about how they work.
And no, IMO stenography isn't security by obscurity, in the same that using RSA and keeping the private key private isn't security by obscurity - keeping the private thing private is part of the security model.
If the countries were reversed, and some Chinese software implemented an equivalent "security feature" to track US users, it would be all over the news about how China is conducting spying and espionage on America.
Or maybe you don't understand this hypothetical situation either, but I'm suspecting you just don't care about other people's privacy.
“So the feature mostly punishes the exact people who are easier to fingerprint: normal developers doing weird but legitimate things”
What’s the punishment here exactly?
For being flagged as possibly a competitor? They nuke your account.
Higher odds of being banned for legitimate usage.
Output poisoning and/or eventual account bans, if I had to guess.
Returning invalid poisoned different results that were not what you paid for
They probably run a heavily dumbed down version of the model, same as what they got caught doing with Fable.
And that's also why, as a legitimate customer, want none of it, you never know if you accidentally entered a zone they don't like.
"got caught"
to clarify, this behavior was announced with the model release
The extent got caught.
if by announce you mean shove it somewhere in a pdf with hundreds of pages, yes
https://www.anthropic.com/news/claude-fable-5-mythos-5
This is not hundreds of pages and it gets its own bold headline section.
> What’s the punishment here exactly?
Seeing as how Anthropic cannot stop raising a stink about "illicit Chinese distillation attacks" every month or so, I'd bet money on them either already silently degrading model performance if any of the identification patterns match, or, at the very least, considering it/doing dry runs.
Particularly considering that they've openly stated that the technology to do so exists and that they were going to use it in production on Fable.
Can somebody clarify for me - if ANTHROPIC_BASE_URL is set to a different provider... then isn't this "marked" system prompt being sent to that provider's API rather than Anthropic's?
I understand how this can be useful to Anthropic if the 3rd-party is acting as a proxy (because they end up hitting the Claude API with the marked prompt), but it looks like requests where "hostname contains deepseek" would never be sending data to Anthropic. What am I missing?
This catches Claude resellers. Meaning companies who proxy Claude traffic for users in, say, China.
https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...
"Catch" as in made a list?
Won’t catch many after has been on hn home page. And now the providers will be even more careful to upgrade the cc code. Might even provide their own agent to prevent this mockery. And isn’t what anthropic did unauthorized use of another pc which is kind of illegal?
Thats the thing, hoping to control things on client side like this is a lost battle if you are dealing with technical clients. The best they can do is probably based on IP, but again the motivated clients would just create bastion servers in allowed IP ranges. I am surprised why are they even throwing resources in this kind of effort.
“Hey Claude, fix the issues with Chinese resellers and distillers. Make no mistake”
My guess is for distillation, they need to forward the prompt to Anthropic to get the real Anthropic model's response so they can train their own models on it
The theory is probably Deepseek might be collecting those streams, and sending a portion of it to Anthropic to see what the Anthropic/Opus response would be.
Did I understand correctly, that custom base URL triggers this behavior? So if I'm running Claude through a LLM proxy, I'm also affected?
Ask Claude to check, lol
I am also really confused and annoyingly stuck on this. I understand that the model name might appear in prompts for distillation (I guess? "You are RipOffModelv2, learn from these responses from Claude")?
I guess the only explanation is that there's a side-telemetry channel that still sends some data to Anthropic, regardless of ANTHROPIC_BASE_URL overrides.
> I understand that the model name might appear in prompts for distillation (I guess? "You are RipOffModelv2, learn from these responses from Claude")
This does not make sense. You wouldn't send such a prompt to the Claude model. And when you're sending the prompt (anywhere) you don't have the response yet. This is not how distillation works.
There are a lot of companies reselling Claude to Chinese users. You use their base URL but it's still going to Anthropic.
This is very interesting. Combating resellers and distillation seems like a very difficult problem indeed. Interesting to me is that these techniques mentioned in the article are just like anti-observation techniques used by some of the more sophisticated malware out there, however defeating them is pretty trivial.
Yes, defeating this is relatively easy, particularly for sophisticated actors. But it's hard to always defeat all of the tricks. Sort of like how it's expensive and hard and uncertain to defeat all of the tricks when forging money.
Here's an example. Say you have your team use patched binaries. Then CC updates and requires a new patched binary with new tricks. You now have to have a team ready to analyze the binary and begin to address the tricks; meanwhile, unpatched code is now a fingerprint. If some researcher decides to update Claude on their own to access new features, they get fingerprinted.
Defeating a single fingerprinting technique once is easy. Defeating all of the techniques all the time is hard.
Not to mention, it isn't that hard for vendor's to require updated code to run the product. Vendors do this all the time.
Corporate surveillance malware on employee machines is also defeatable but most don't bother.
Is it hard? Just ask AI if the update added any new fingerprinting vectors?
I'd love for you to try this and report back. My guess is that no models today will successfully run a binary analysis for fingerprinting without a lot of handholding. If you try to use Opus it will almost certainly decline (and fingerprint/ban you).
Not with Claude Code, but I trivially had Opus scan other closed source software for fingerprinting, including native libraries that it called into.
Can you share more details? I ask because my experience suggests that models still require a decent amount of expertise to use for binary analysis (largely inferring because of use on other tasks of this level). I would expect models to always find "something" when you ask for stenographic techniques in the code, but with an extremely high false positive rate.
I don't think the diffs between Claude releases are that big. The amount of code in a diff doing sketchy stuff like looking into the host environment is going to be pretty small and obvious for the model. You can do things like ask for what an update included that wasn't mentioned in the release notes and stuff like that.
seems ironically like a similar problem of content owners trying to filter bot scrapers from legit users
If they only collect the data for analysis I guess this is fine (they already get way more sensitive data from users anyways, so if privacy is your concern you've made the mistake many steps ago). The much more interesting question is if they directly act on this data in their API. For example by rate-limiting, compute-limiting or rerouting to weaker models. That might even be legally questionable. I would really like to see this as a follow-up analysis, but I guess it is way more difficult and will also cost quite a bit in tokens.
Would it be legally questionable, or actually complying with U.S. export law?
"If they only collect the data for analysis I guess this is fine"
I think you missed the memo on how foolish this attitude is. It came out around the time Edward Snowden made his discoveries at the NSA public. I suggest you look into it
As I said above, if you are worried about privacy while hooking up Claude Code, you need to reevaluate your understanding of this technology.
I've heard that it was possible to trigger really obvious output poisoning on Fable with something as basic as asking the model to think outside of its built-in hidden thinking delimiters.
This watermark may trigger a similar mechanism.
None of this is surprising - they're trying to mask and relay when they detect known patterns of what looks like distillation attacks and client app copying/modification. The list obfuscation here is likely to prevent or make it difficult for those same adversaries to work around this or delete/null it out when making a bootleg copy.
Cool reverse engineering/analysis report but if this is the extent of nefarious activity that came of it (trying to catch/mitigate chinese lab model distillations), that's kind of encouraging.
This is weird but, help me understand how this meaningfully impacts our exposure.
I'm authenticated to Claude, so they already have the whole attribution thing solved.
User != paying person/company/reseller.
based and steganopilled
Claude code does feel very malwarey to be honest. They have been like that from the start.
This was already discovered during the source map leak.
> This is not a malicious feature, but it is a weird choice for a developer tool that asks for trust.
They already tell you they scan for malicious prompts, and they have no ZDR guarantees for consumers. Why do signatures like this matter at all?
There has been an anti anthropic propaganda push by bad actors across social media sites especially Reddit and twitter. This started a few months ago when anthropic started beating openai.
Absolutely. Nothing makes me believe dead internet theory more than text threads discussing anyhropic and openai.
That's a lot of effort when they could just play a short video saying 'You wouldn't steal a car' instead
What's the point of even trying to obfuscate this with such a simple method? Could at least have hidden the targeted features by storing their hashes or embedding a bloom filter or similar
In this case, this is probably not the only stereographic tattletale.
Had a competitor pull something like this with a previous employer. They were supposed to be interoperating with a standard, but they had a secret steganographic handshake, which they used to pretend that competitors products were unreliable (they had a first mover position in a smaller national market with specific requirements, so this wasn't shooting themselves in the foot). Our guys figured out the handshake and just silently implemented it. In this case, the competitor wasn't big enough to waste engineering time on multiple such hacks, but Anthropic have time (or Claude does).
The point is not raising red flags I guess
I love how well this comment works as a vexillology joke, even if it wasn't intended.
(This sounds like a clumsy way of catching the Chinese that easily can be side-stepped.)
Claude Code has more or less full access to the client computer. The server (that hosts the actual AI) can just go: execute this payload and tell me the result - otherwise I won't answer any further questions or re-route you to a stupider model.
The payload could check for Chinese time-zones, scan for copies of the little red book on the local hard-drive, or ping truth.social to see it was behind the great firewall.
> Claude Code has more or less full access to the client computer.
It shouldn't, not if you run CC as a separate unprivileged user. I wouldn't run CC on my main user account with sudo and access to my home directory or other resources. This is what the UNIX permissions system was designed for.
> "That also means the client itself deserves scrutiny. If a coding agent can read your repo and run commands, the binary that ships it should be boring (ƒor example, pi harness)"
You're actually trust your security to your harness AND model AND inference API provider in this scenario: https://jacob.gold/posts/why-i-wont-run-untrusted-models/
the Chinese they are trying to catch must be amateurs, first thing you should do is construct a sandbox which looks indistinguishable from a common user. second thing is to put it behind a residential proxy.
It is about China detection. They seems to put a tracker on the email as well.
Anthropic must think that their moat isn't very large if they're this worried about distillation.
Dario's been openly talking how worried he is about China and labs getting synthetic training data off their models, for years. Most recently in relation to "Mythos level" capabilities.
Not really distillation, just synthetic training data.
What moat?
After loving Claude Code for most of its lifetime, I've been extremely annoyed by every change in the past months, even on the model level.
There seem to be all sorts of continual under-the-cover changes like this one that make life harder. It feels like the entire product has been taken over by overly ambitious PMs that care more about making their mark than in improving the experience, and all of their marks have made me less productive.
I've been using Pi with GLM5.2 the past few days, and though it's expensive, I find it far more productive and less annoying. The remote session plugin is far more reliable, I don't need to intuit some undocumented usage pattern to figure out how to use it well, and it just works.
curious for those with experience - what do people prefer about Pi vs. opencode alternatives? i've mostly been using pi as well but not out of any principled decision
Given the source code leak, I would think there'd be open source versions by now.
Huh, that's right! You'd say that an enterprising developer with a 20x subscription could slopmaxx this in a weekend...
> I've been using Pi with GLM5.2 the past few days, and though it's expensive
are you using the API for glm 5.2 or how exactly is it more expensive? How is GLM5.2 more expensive than using Claude code, that doesn't line up to my experience but to be fair I am on an older yearly subscription which generously only has 5 hour limits.
To be fair though one minor criticism of GLM 5.2 that I have is that it does seem to overthink quite a lot sometimes but the results end up being (good?),
I personally have used Glm 5.2 with (Opencode + obra/superpowers) / Oh-my-pi / Maki.sh
I like the 1st one when I am doing a longer project, the 2nd or 3rd one when I am doing a project which doesn't want me to ask too many questions and simply spin me up something. I sometimes use free online interfaces of claude and gemini and others like AIstudio for that as well which surprisingly can lead you to go far as well.
Overall, I am decently happy with the state of Open-source models actually and the eco-system around it is probably gonna have even more innovation surrounding it.
This seems really, really stupid. Similar to the weird Zig runtime signature thing from a few months ago ago, it was bound to be discovered, quickly, and all the resellers have to do is find a new domain name that (checks notes) doesn't have the word DEEPSEEK in it. Like, seriously? Your goal was to identify resellers by checking if the proxy has the corporate name of one of your competitors in it? Is this amateur hour?
All Anthropic has done is reduce trust, once again, with legitimate customers, while doing nothing to stop illegitimate customers. They need to get adults into key leadership roles, quickly.
Is this why Claude never knows what date and time it is right now?
If there weren't already enough tells that something is AI-generated, I guess you could add this to the list.
The AI race right now is in a sad state. Chinese's playbook is releases open weight models and trains them on their own chips.
Anthropic pushes fear and control. But the only way to win is by innovating. China is flooding the market with cheap, good enough models, while the U.S. is building a Chinese firewall.
Frankly, I don't see this as the concerning behaviour the article describes. It is fine to try to protect against distillation through a technique like this. This will also allow them to, instead of blocking the distillation agents, respond with a poorer result/model, hindering the progress of distillation, momentarily at least.
I would guess that's their first line of defense; they should have more techniques to identify distillation because that's a very simple way of detecting the host and can be easily spoofed.
> This will also allow them to, instead of blocking the distillation agents, respond with a poorer result/model,
i.e. this will allow them to literally commit fraud against paying customers
1st, this technique is not fraud, and fraud is a separate accusation. 2nd, paying customers can legally and legitimately be banned and monitored for breaking terms of service, which probably includes things like using the model against U.S. export restrictions.
Banning is completely different than charging for a service you're silently not providing.
Evidence?
So if I change my timezone to Shanghai I deserve to get banned? Or get shitty model instead of what I’m paying for?
Evidence?
Do paying customers distill? Is it fraud to protect against distillers?
That's what capitalism is all about, baby! Especially if the customers don't notice.
It piqued my interest. I think I’ve found a weekend project
One more example of "I thought Anthropic was supposed to be the good guys."
Cool fingerprinting avenue.
I use its too
I clicked the link to learn what steganography mean...
Steganography is, essentially, hiding information within another message, such that it's not readily apparent that the message contains the information.
Is it just a minified localization(l10n) function maybe?
Non-hugged: https://archive.is/Wdhp0
this is the one they wanted us to find
Silicon valley season 6 was on point.
I used Claude Code for a month because my boss gifted me a sub and wanted me to try it.
I used that month to complete a work project and then beef up my personal harness so I'd never have to deal with Anthropic (and these sorts of shenanigans) again.
The issue is that using Claude Code is an easy compromise for most to make, when you get to use the models 10x cheaper than through API pricing with a custom harness.
The cheap tokens are the product.
Which is why my vibeslop harness supports `claude -p` as one of its backends.
If that ain't getting steganographically tagged...
How do people build something like a personal harness? Are there tools for that or is it done from scratch?
I like this tutorial for an agent in 50 lines:
http://minimal-agent.com/
And if you add one additional while loop, for user input, you can actually use it! :)
https://gist.github.com/a-n-d-a-i/5461a662ef8a7ee0a5eb7778c8...
Build it from scratch. Understanding fundamentals of how agentic coding harnesses is a must though if you gonna go that route. I think everyone should take time and learn these things, maybe reverse engineer Codex Cli or something like that as a starter. That info is very valuable in this day and age.
Can you say more about Codex? I'm using GPT-5.5 in my own harness and it's not liking it very well, so I'm thinking I ought to make it more Codexy so it's more ergonomic for it. (edit format, tool calls etc.) But haven't gotten around to it yet.
Not the comment author, but I use pi and customize it with my own extensions. Pi automatically tells models how to customize itself, so it's a pretty easy process.
I started mine from scratch in 2023 because I wanted to use LLMs from a terminal and there was nothing else compelling at the time (nowadays there is pi and opencode)
Harnesses are/can be incredibly simple things, not much more than a HTTP client that renders things in a way that suites your taste.
It’s not that difficult, it’s just a system prompt and a set of basic file edit/bash/etc tools.
Me, personally, I didn’t build it from scratch but I ported original CC from published sources into Python and extended it to match my own requirements.
Are you using it with Claude? They only allow their own harness with the subs right? (And per-token billing is like 10x more expensive?)
Here is a video I made explaining it from absolute basics:
https://m.youtube.com/watch?v=_AgKuFGvJfI
And the repo:
https://github.com/abtinf/homunctor
Building something like this is the todo list of agents.
I found this one easy to understand:
https://ampcode.com/notes/how-to-build-an-agent
The real question is when do you transition from building it with codex/CC to the harness itself.
Why use a personal harness?
You have to pay API pricing, which is far more costly.
I'd either switch to GLM wholesale or just continue to use Opus within Claude Code as the blessed, subsidized path.
I would guess it is to avoid model lock-in.
My question is still this - why not just use GLM at that point?
The pricing of Opus outside of Claude Code is insane.
The tokens cost too much outside of Anthropic's blessed path.
I use GLM in my custom harness. It completes the same tasks at the same level of quality, except 8x faster and 8x cheaper. (Same goes for GPT!)
I'm not sure how that's possible. I expected to get increased correctness for that order of magnitude (something something test-time compute!) but I am not getting it.
Given the Anthropic shenanigans, do you trust the personal harness code it wrote for you?
It did not write it for me, I used it to add a feature I wanted. It's a pretty small and understandable codebase, in fact :)
Does anyone know what’s gone wrong with Anthropic?
They used to be a decently credible company with not-too-shady behaviour...
I hope they can actually regain some credibility…
When have they ever been credible? They have always been shady with their talk of safety, Dario was the one who wrote back in 2019 that GPT 2 was too dangerous to release.
I don't think many people care that they are trying to detect resellers and distillation.
It also doesn't seem very consistent to fixate on that while sending Anthropic everything about you via your day to day prompts, every line of the projects and environments you're working on at work, etc.
Their credibility comes from having one of the best models.
This sounds similar to what people were saying regarding Microsoft when the shady tricks of consumer Windows 10 versions were revealed.
…And then Windows 11 became even worse.
Their philosophy is what's gone wrong.
It has some good effects on the their models, like Claude seeking cooperation first. But the people behind the company have a typical "unconstrained" (in the Sowell vision sense) perspective that assumes that they know better, so they are righteous for attempting to control things (users, paying customers, their model outputs, their tool chain, the supposed deity they assume they will produce... etc.)
Amodei world: pompous zealot with God complex
Altman world: malfeasant nihilist with God complex
Yeah I guess there is a slight undertone that they are the superiors… with the rest of the tech world being the inferiors.
But I hadn’t thought that as anything more than temporary flights of fancy.
They've only been around 5 years and have grown tremendously during that time. There's no stable reputation you can rely on yet.
They just show their true face. You’ve been lied all this time. They were never “good”.
I used to interact with the LW crowd… and they were mostly not outright swindlers or scoundrels. (from what I could sense)
I think it’s fair to say most had decent respectability.
Anthropic hired heavily from that pool so it’s astonishing how it turned out.
Enshitification. Too big to.. upset the govt.
What models are you using? Aren’t you still dealing with some provider even if you are not using their binary
I self-host DeepSeek V4 Flash on 2 DGX Sparks (approx. $10k)
I expect DeepSeek V4 Flash (or an equivalently sized model) to reach parity with GLM 5.2 some time this year (this based on DeepSeek V4 Flash launching at GLM 5.0 parity[0], and GLM 5.2 being freely available to distill from)
GLM 5.2 is within spitting distance of Opus 4.8 and is at least as good as Opus 4.6[1] which some devs were willing to spend hundreds to single-digit thousands of dollars a month for a few months ago.
[0]: https://artificialanalysis.ai/models/comparisons/deepseek-v4...
[1]: https://artificialanalysis.ai/models/comparisons/claude-opus...
How fast is it?
2000 t/s prompt processing and 40-50 t/s generation. We should see 60-70 t/s generation with DSpark support solidifying in vLLM in a few days
Recent discussion on DSpark: https://news.ycombinator.com/item?id=48696585
[flagged]
Phased rollouts are a triggering microagression for some.
Headline is, frankly, awful. This isn't the AI secretly doing stuff and hiding it. This is the very human Anthropic engineers trying to detect Chinese scraping via some frankly hamfisted and unimaginative URL trickery.
I didn't assume it was the AI, just that some part of the the overall Claude Code product was doing this. I didn't assume the feature was added to Claude Code without human oversight. If it was added by Claude-the-AI itself without the humans prompting it to I would still hold the humans at Anthropic responsible. Does that make you feel better?
Defence in depth isn't hamfisted. They're only noobs if this is all they do.
The AI is Claude. Claude Code is the harness.
Here's the sha of the prompt I submitted... no I don't know why there are no saved prompts with that sha.
What do you mean you don't know where the bug is coming from?
No, I absolutely didn't make it up, how could you accuse me of that?
Does anyone know when this regex isn't working? I double checked it 27 times, I even asked the LLM. They all say this regex should be finding these dates.
Weird, suddenly all the conversations are breaking when I feed them into this other tool? Something about UTF-8 errors, but I'm sure I'm only using ASCII?
I do try to take care to make sure the things I build can be used by other people even when they care about different things. I care about understandably, determinism (as it relates to computing), and repeatability (because I want to be able to trust the systems I use).
If y'all would be willing to try to account for use cases of others, and try not to break them... that would be nice.
Please note: that generally when you modify something that belongs to someone else without telling them... things should be expected to break.
[flagged]
[dead]
[flagged]
this seems a bit extreme. pangram does not work. i have tricked it multiple times. i don't get how people are still trusting these systems.
it's just a different car on the hype train
Source: Other AI
[flagged]
Is it worse than the companies that built the agent and gave no credit for the data they used?
Why would you give free advertising to trillion dollar corporations?
Would you also say that "someone who wants to use an IDE / LSP features to code and not give credit to the IDE / LSP is the worst kind of person"? If not, what is the difference between the two for you?
one wrote code while the other is used by meatbags to write code. why is this example always marched out like it means something?
> one wrote code while the other is used by meatbags to write code.
One is not a "meatbag" while the other is not a "meatbag". And no, outputting something on stdout that happens to function as code is not "writing" it in the sense that we actually care about here. That's conflating the metaphor we use in describing program behaviour with the actual "meatbag" activity.
> why is this example always marched out like it means something?
Because it obviously does.
Almost all ways of creating programs are effectively just using tools to produce code. Compiling, transpiling, interpreting byte code, etc.
again, that's not what we are talking about here. we have humans writing code using an IDE. we have LLMs generating code that is placed in the IDE. why are people obtuse to this? why are bots obtuse to this?
Claude didn't "write" anything until a meatbag told it to.
My employer didn't write anything until they told me to.
> Would you also say that "someone who wants to use an IDE / LSP features to code and not give credit to the IDE / LSP is the worst kind of person"?
That's a false equivalency.
> If not, what is the difference between the two for you?
Let's start this out right: if they're equivalent, first you explain to us why you think so.
> That's a false equivalency.
How is it false?
> Let's start this out right: if they're equivalent, first you explain to us why you think so.
I think it should be really obvious how they're equivalent: both are the result of a program running on a computer, and not the result of in-the-moment cognition by a moral agent or moral patient. Of course the LLM is just a tool. Models can literally be downloaded as ordinary files. There is not some threshold to cross where some configurations of bits on a disk deserve "credit" for work and others do not.
> I think it should be really obvious how they're equivalent: both are the result of a program running on a computer...
In fact it's really obvious everything is equivalent: it's all just matter and energy!
> Of course the LLM is just a tool. Models can literally be downloaded as ordinary files. There is not some threshold to cross where some configurations of bits on a disk deserve "credit" for work and others do not.
Of course there is such a threshold. And it's definitely been crossed when the "tool" can operate autonomously or nearly so, when it can generate the "creation" with minimal operator input or understanding.
Your classic IDE can't do anything without the detailed control of its operator. It's nothing like a coding agent.
I just don't agree that it's a false equivalency. I see them both as "tools I use to get the job done". For me, the job is not "writing code" - it is "deliver feature", "fix bug", and the accountability, responsibility, and communication that comes with it.
> I just don't agree that it's a false equivalency. I see them both as "tools I use to get the job done". For me, the job is not "writing code" - it is "deliver feature", "fix bug", and the accountability, responsibility, and communication that comes with it.
Hello, Tom Smykowski. You have people skills!
https://www.youtube.com/watch?v=hNuu9CpdjIo
A lot more durable than software engineering in this day and age...
Should I credit Microsoft with my perfect spelling as well ?
And your comment is completely irrelevant to the article's content.
[flagged]
If scrapping content is legal, model distillation should be legal too.
> If scrapping content is legal, model distillation should be legal too.
No, because legality should be determined by what's in the best interests of Athropic and OpenAI's business models.
Hopefully they're working on RLHF their models to insert clauses making that reality clear into any legislation their models generate or review. That way it's only a matter of time until the confusion is cleared up.
I suppose model distillation is technically legal, in terms of copyright, because LLM output is automatically public domain.
It's only "illegal" from a standpoint of breach of contract given its against the terms of use/service, which is to say its not illegal at all, there's no criminality there.
Yeah I considered whether I should use the term "illegal" in my original post, but in this case, I believe these models are actually banned for use in China, right? Like there are probably export controls (at least with the NVidia chips)
I honestly don't know ... yeah if it's just technically a terms of use violation (which isn't illegal, just a violation of one company's rules, for which Anthropic has every right to stop), or do we now have export controls applied from the various government actions, etc making them truly illegal now.
we have global export controls on Fable/Mythos, and I think (but I'm not 100% sure) that other frontier models are illegal for a US company to provide to China. So Anthropic geoblocks it, but unlike Mythos/Fable, non US citizens can still use Opus, etc just not from within China.
But because of the public domain status of LLM output (in the US) I'm not sure paying someone to run a bunch of prompts through Claude, post the output on a public website and then have a lab in China pull that output, would run afoul of any laws I think that would be legal on technicality. AFAIK Anthropic has no ban in its terms of use that you can't share Claude's output publicly. You still need interactivity for distillation, but I don't think (for now) there's anything stopping a Chinese or other lab from sending people to the US, signing up for a Claude subscription and doing the work state side.
Distillation is pretty much impossible to stop. The US GOV would have to go the full export controls route like they did for Fable/Mythos to stop any non-US citizen from using/accessing the model, which is going to be impractical if not impossible to enforce.
There are so many China born Chinese employees at Anthropic and OpenAI and I think quite a lot of them have already been recruited as spy . So it is almost impossible to keep secrets from Chinese government.
> steal the models or illegally distill them
The irony.
At what point though doesnt somebody stand back and say "wow, thats really dumb!" I think its probably more an indication of a dev having too much time on their hands rather than being in a hurry.
Not totally new territory; there was a highly compressed period of panic about encryption 35 [0] years ago:
https://en.wikipedia.org/wiki/Pretty_Good_Privacy#Criminal_i...
[0] f**k I'm old
> steal the models or illegally distill them
Oh no, they're trying to steal the models that were trained on stolen data? That's horrible, I feel so bad for Anthropic.
The more I learn about Anthropic the more they disgust me. Finger crossed for all the companies from their “ban list”
Which AI company have you learned more about where you liked them more as more details came out?
nous research. started out making overhyped llama finetunes, now they got a great agent harness and a cutting edge distributed training network that actually works.
Moonshot.
Deepseek.
Ridiculous.
Is that really how it is? How will this affect our future?