Post

This is either brilliant or scary: Anthropic accidentally leaked the TS source code of Claude Code (which is closed source). Repos sharing the source are taken down with DMCA. BUT this repo rewrote the code using Python, and so it violates no copyright & cannot be taken down!

533 reposts2.3K likes

Isn't me or, I've seen last year that stuff written by machine learning and LLMs can't be copyrighted? If it is the case and they rewrote parts of FOSS software to claim copyright violations, I'm not sure it has force of law...

This is why you need hardware. Software can be spun up by any random person in the world. Or you can leak your whole company's work overnight? Ouch. You need something to be the backbone of the software company. HARDWARE!

The deeper story: the leak also exposed unreleased model families (Capybara/Mythos 5) and internal agentic features (KAIROS with nightly memory dreaming). The DMCA fight is about the code — but the real news was in the comments.

Prose and Khans@temujin9.t9productions.com

i've yet to see convincing legal theory that it's illegal especially given the apparent non-copyrightability of ai output model owners might dislike it but them's the breaks: live by the copyright exemption, die by the copyright exemption

If a general leaked battle plans, he must expect to lose. That's life. Lock up your treasures or someone else will grab them, one way or another.

I would trust something written in Chinese over something written by American oligarchs any day. USA is a fascist shithole regime worse than Russia and China combined

This is a subjective opinion, and this text is a complete pile of spam If you translate this, it will be like "free Claude Code app"& other nonsense.). This is confirmed by the fact that the author has disabled the issue reporting option. I would recommend using a translator and not being so rude.

I just found out 10 minutes ago that Claude code was a thing. It’s really difficult for us old folks to not only understand the ever-changing technology, but to try and understand the lingo that goes with it. Gen Z is: “Hey Boomer! Empty your head of 70+ years of information; it’s useless!”

This is the best summary of the whole circus. DMCA the code, someone rewrites it in Python, the DMCA is meaningless, and now the ideas are free while the implementation is protected. Copyright was designed for a world where copying was hard. We don't live there anymore.

This sounds like something I should be concerned about, but am ignorant. Is this showing me that someone thinks it’s cool to steal our money while we’re not looking?

I would guess this is intentional. OpenSource AI is the only way we can hold these guys accountable. We did it with other Open Source tools....we'll do it again. Yay!

Anthropic recently also leaked system prompt. As expected climate hoaxing is real. AI has to be trained on pristine data otherwise it cannot win on leader board tests and then system prompt has to have something in there to tell it to lie and propagate climate hoax.🤣

And it will create useless stupid so-called "engineers" that have no clue how anything actually works. A travesty of technology that is literally destroying the planet. Oligarch dystopian bullshit.

Fair, and Anthropic is the least of the evils, but I’m a big open source guy. Use it against them, we can’t uninvent it, so we have to figure out risk mitigation. As far as useless stupid so-called engineers go: look at Musk. He’s a rich idiot with some smart sounding catch phrases.

I will NEVER download or use any of the AI garbage EVER. It is utterly useless shit and has no place whatsoever except in highly-trained, very specific, applications. Generalize AI and using it for code is a fucking joke that leads to stupidity, misinformation, and cookie cutter plagiarized apps.

I think you need to take a moment and realize that it exists. As a code monkey I also wish that it didn’t, just like I wish we never invented guns. But the genie is out of the bottle and now it’s regulation time. How can it be positive? Without knee jerk responses. Unless you can magic it away.

The DMCA angle is fascinating but what actually got me was the 44 hidden feature flags in there, a fully built autonomous background agent that just hasn't shipped yet. That's the part worth a longer conversation.

Right now is a crucial pivot in tech. It isn’t the time for leaks. I hope whoever did this was fired and blacklisted. It’s not just about how it could cripple a company - it’s taking down all the work your peers put in. Claude will still be ahead, but now demands rise on the people building it.

So-called "AI" is a total sham! What is reported here + copyright issues confirm. It's starting to look like a huuuuge Ponzi scheme!

There's no claim on this particular repository to have performed a clean-room reimplementation though. (I have seen another repository, reimplementing Claude Code in Rust, which explicitly makes this claim.)

That said I think courts are going to be very skeptical of a single person using an automated tool to perform both halves of the clean-room reimplementation process, I'd be very surprised if they found that to provide sufficient separation to not count as a derivative work.

Given how much of Claude Code is ostensibly the product of Claude Code, and the current uncopyrightable status of LLM-generated works, well... the assemblage is still *probably* copyrightable, but somebody on the receiving end of a DMCA takedown has a fun potential test case if they want (fsvo fun)

I am not offering any advice ha. I'm observing this repo, based off of Claude Code's original source is up, and very popular, and becoming even more so. The workaround claimed: We don't even know if fully AI-generated code has copyright btw (like Claude Code's). All open questions, no?

it's obfuscated; you can deobfuscate it, which is basically what happened here -- the source map contains everything you need to map the obfuscated code to the actual source.

Gotcha, that's what I figured happened with my limited knowledge of js/ts. The posts are making it sound like something bigger happened.

Your take is wrong and misinformed. One of the fundamental bundles of copyright protection relate to derivative works. That gives Anthropic the ability to control creation and distribution of derivative works.

The Python rewrite sidestep is a good illustration of why copyright is a weird fit for software architecture. The ideas in the leaked code, the patterns and tradeoffs, those are now public. DMCA can suppress the syntax. It cannot suppress understanding.

The Python rewrite angle is clever. But the more interesting thing is what the leak revealed: the internals were not that exotic. A lot of scaffolding, context injection, and careful prompting. No magic. Which is either reassuring or unsettling depending on what you expected.

The Python rewrite angle is interesting. Functionally equivalent, legally distinct. That kind of creative compliance will become more common as AI tooling sits in this weird IP gray zone.

I highly doubt that the courts would find the code to be uncopyrightable. Courts have found that translations are infringement, and there are lengthy law journal articles you can read arguing the same is true for refactored code.

He was sued for copyright infringment. DMCA basically took those laws digitally and made it illegal for digital copies. The IP is protected by copyright

so wait, Anthropic accidentally exposed source code to the public. somebody used AI to pirate it in a different codebase, and it is now public software? 🫨 somebody's in trouble. US govt should not let that stand. If true, Open AI, others have already stolen it.

Copyright absolutely does protect derivative works, like translations of a book to another human language. But yes, if you change things enough, it's considered a new work (e.g. GNU reimplementations of proprietary UNIX programs).

It depends. A "non-exclusive licensee" (Opensource) doesn't have standing to sue alone and thus needs to get the original author to join any action as an "indispensable party". It's a huge oversight of the whole open source ethos; partly because many never really understood copyright law properly.

This resonates. The tooling for multi-agent coding workflows is still in its early days — lots of room for better patterns around isolation, testing, and task coordination.

No Python port. It's firmly in the TypeScript/Bun ecosystem. Given the scale and the tight coupling to Ink and Bun-specific features like bundle-time dead code elimination, a port would be a major undertaking. Most (all?) of the "rewrote" posts are jokes.

My understanding is that AI-generated content does not have copyright protection, at least in the United States. That is, it is not considered a derivative work, in a legal sense - because of the lack of human authorship. Happy to be corrected.

That paragraph that begins "The brilliance" is completely, utterly, insanely wrong. That it was rewritten using genAI means that the output cannot (currently) be copyrighted, but that still doesn't mean it's legal.

Well… yeah, morally, absolutely—but this a real hoisted on their own petard bit isn’t it? The whole project relies on being able to steal other people’s data and that when you run it through an AI, somehow it’s no longer infringing.

no but a few things: 1. can AI-written code (that has no copyright) have copyright? Almost all of Claude Code was AI-written 2. The implementation claims a "clean-room Python rewrite" trying to position as not an exact derivative. LLMs themselves have some level of derivative characteristics also

i'm not asserting that it would apply in this case, but you made a statement about copyright and derivative works which is just simply wrong i don't think any court cases have tackled generated code but i'm guessing the same rules will apply if the judges don't get snowed by "ooh so technical"

Oracle vs. Google effectively established that even an API specification is copyrightable; while Google eventually got a ruling that *their particular* use was fair use, that was a much better defense in that case than it sounds like this would be. I highly doubt their "clean room" is clean enough.

This is not correct. Copyright absolutely covers derivative works. Copyright may not cover *sufficiently transformative* works, but this almost certainly does not apply in this case.

Also another angle to consider: Fully AI-generated code has no copyright Almost all of Claude Code is AI generated (as shared by Anthropic) So... does that AI-generated code itself have copyright, to start with? bsky.app/profile/amyh...

Given how much natural language text was stolen to feed these chatbots, I don't think the public in general should care about Anthropics or OpenAIs or any other AI giants code.

Got a website idea? 👀 Describe it and Freebuff’s AI coder can build it for you — completely free. I’ve used it for 2 weeks and hit ~5B tokens. I’m sharing my referral because referrals give me more daily usage 😭 Try it: freebuff.com?ref=ref-e0a3... GitHub signup required. You can build anything.

freebuff.comFreebuff — the free coding agent (free Claude Code, Codex, Cursor & Lovable alternative)Freebuff is the free coding agent: a free CLI coding agent and Freebuff Web, the free way to build full-stack apps. No subscription, no setup, no lock-in. The free alternative to Claude Code, Cursor, ...

A "point of attachment" in © law signifies a connection between a creative work or its author & a country that is a member of an international © convention. This connection is *crucial* ...it establishes the work's eligibility for © protection under that specific treaty. *TDLR: No author = no ©*

IMO- not legal advice, based on info for social media - huge if true- if ANY AI gen system uses opensource or AI "vibe coding" then due to a lack of author - the firm that produced such stuff *has made an "overt act" to place such stuff in the public domain*. Lol!

> Fully AI-generated code has no copyright I would disagree if this was in the UK. There is copyright; but there's no other defendable IP. But transmogrification of that code does indeed clear the copyright issue because it's not a copy and you can't patent generic use software in the UK.