Claude Code Leak: 8100 Takedown Requests and the Birth of Claw-Code

A moment of carelessness was enough to tear down the painstakingly erected walls around one of the most valuable secrets in the AI industry. What began as a “human error” at Anthropic has, within a few weeks, developed into a full-blown scandal that is shaking the foundations of digital copyright.

At the center is the source code of Claude Code: a tool that assists programmers in software development. Observers recently reported über the alleged theft of intellectual property through model extraction (“distillation”). But that was just the beginning of the AI startup's worries. An accidentally published source map file revealed over half a million lines of source code, triggering a chain reaction that sent the developer community into an uproar.

Security researcher Chaofan Shou was one of the first to notice the file in a public directory at Anthropic. This allowed the compiled code to be traced back to the original source files. Soon after, the original source code appeared online.

In the world of open source, mirroring content and creating forks is common practice. The data package spread on GitHub within hours before Anthropic could notice the error and delete the original. The company's reaction was draconian: Over 8100 takedown requests under the Digital Millennium Copyright Act (DMCA) were issued, according to the blog IPKat, to stop the spread of the copies. But the ghosts Anthropic summoned could not be easily tamed.

The Architecture of Autonomous Agents in the Spotlight

What makes the leak so explosive, according to expert analysis, is the nature of the published material. It wasn't details about the function of language models like Opus or Sonnet, but the software architecture that controls their interaction. Claude Code is so valuable because it represents a very efficient way for AI to write software. The leak revealed how the system assembles contexts for user requests and how the so-called agentic framework works.

Such autonomous agents are considered the next step in AI evolution: They no longer just follow predefined paths. Instead, they make independent decisions, use tools, and correct their instructions during the process. For Anthropic's competitors, this insight was a gold mine for improving their own systems.

At the same time, the incident raises new legal questions. It becomes particularly sensitive since, according to reports, Claude Code was written by Claude itself to about 90 percent. If an AI writes its source code, what about authorship? US courts have repeatedly clarified that fully autonomous AI creations do not enjoy copyright protection. The human must be the decisive element of creation. If Anthropic now tries to enforce rights on code that is largely machine-generated, the company is treading on thin ice.

Claw-Code: The Community's Agile Counter-Attack

The developer community reacted to the wave of takedowns with a proactive move. Within a very short time, Claw-Code appeared, a Python rewrite of the core architecture. It is attributed to Korean developer Sigrid Jin with the help of another AI – OpenAI's Codex. Claw-Code became the fastest-growing repository in GitHub's history.

The Claw-Code creators emphasize that their project does not contain any proprietary Anthropic files. It is an independent new development. This leaves Anthropic with the dilemma that its AI-powered legal departments must examine whether an AI-translated version of its AI-generated code constitutes a copyright infringement.

The sheer volume of takedown requests also highlights a systemic problem with the DMCA process. Anthropic initially had thousands of repositories blocked, only to quickly narrow down the claims to just under a hundred specifically targeted copies. Critics have long called for reforms, as the current system forces platform operators to remove content immediately without judicial review or evidence. For the AI industry, copyright is increasingly becoming outdated: When AI assistants are used to analyze and replicate other AI systems, the classic concepts of copy and original reach their limits.

The irony of the story: Anthropic is trying to defend its digital Versailles through legal means. Developers worldwide are simultaneously using the very tools that Anthropic and its competitors have perfected to blur legal boundaries.

Claw-Code continues to exist and is gratefully accepted, for example, by Elon Musk's xAI thanks to it. The case shows that the era in which intellectual property could be protected merely by keeping source code secret is likely over. This could overturn the power dynamics in the software world.

(nen)