Recently, a high-profile scandal occurred in the world of artificial intelligence: Anthropic, the creator of the well-known AI Claude, accidentally made part of the internal source code for its tool Claude Code publicly available due to an unfortunate “human error.” What began as a technical glitch quickly turned into a massive legal battle and a philosophical discussion about who owns the rights to code created by AI itself.
What exactly “leaked” onto the network?
Security researcher Chaofan Shou was the first to notice the leak of a gigantic file, nearly 60 MB in size, containing over half a million lines of code. Because the data landed in a public repository, GitHub users instantly began creating “forks”—their own copies of this code.
It is important to understand: the leak concerned the software that manages requests. This is the very “secret ingredient” that makes Claude Code one of the best assistants for programmers. The leak revealed exactly how the system processes context and transforms a user’s simple question into a sequence of autonomous actions. For competitors, this became a true gift, as they can now see mechanisms that were previously hidden.
Legal counterattack and the emergence of “Claw-Code”
Anthropic reacted instantly, sending over 8,100 copyright infringement notices (DMCA) demanding the deletion of copies. However, the programming community was not slow to respond. Almost overnight, a “clean” version called Claw-Code appeared.
Korean developer Sigrid Jin, with the help of another AI (OpenAI’s Codex), completely rewrote the architecture of the stolen code from TypeScript to Python. In a matter of days, Claw-Code became the fastest-growing repository in GitHub history, garnering 100,000 stars (bookmarks). The creators of Claw-Code claim that their project contains no proprietary Anthropic code because it was “reimagined” and written anew.
The paradox: can AI hold copyrights?
Here arises the main dilemma: is such code protected by copyright at all? It is said that 90% of Claude Code’s code was written by Claude Code itself. According to recent court decisions in the US, works created exclusively by AI without significant human involvement are not eligible for copyright protection.
If AI is considered the author, Anthropic may find it difficult to prove its rights in court. On the other hand, if rewriting code using another AI (as in the case of Claw-Code) is considered “fair use” or the creation of a new, independent product, Anthropic’s claims may prove futile.
Mass deletions and their consequences
Initially, Anthropic demanded the blocking of all 8,100 copies, leading to the deletion of even entirely legitimate projects. Later, the company narrowed its demands to a hundred specific repositories. This situation highlighted a problem with the DMCA mechanism: it allows for the removal of content without any evidence or court decision, which often harms innocent developers.
The community increasingly suggests introducing a system of “cash deposits” for those filing complaints to prevent abuses.
Instead of a conclusion
This leak has elevated the discussion about intellectual property to a new level. We find ourselves in a “chicken and egg” situation: AI helps write code, another AI helps break and rewrite it, and lawyers attempt to apply old laws to future technologies.
While corporations fight for control, developers ironize: “If Anthropic left a window to their digital Versailles open, don’t be surprised that people walked inside.” Attempting to solve this problem solely through copyright is akin to trying to extinguish a fire with paper documents—it is a tool too slow and unsuitable for the AI era.
Founder of Research & Patent group Intectica, author of patent algorithms for solving problems in the pharmaceutical industry, patent attorney certified in all intellectual property objects (Patents, Design, TM), with education in chemistry and law, chief expert of the patent institution of Ukraine UKRPATENT (1997-2004). Member of international organizations, including ECTA, PTMG, UAM, lecturer and blogger.
