Anthropic Claude source code leak explained: Techie reveals how a 4 am update exposed 512,000 lines of code

In the fast-moving world of artificial intelligence, even the biggest players can have an off day. That's exactly what happened to Anthropic, the company behind the popular Claude AI models, when a routine software update turned into one of the most talked-about tech stories of the week. A techie broke down how it all started at 4 AM on March 31.

As per the techie, Anthropic pushed out a fresh version of their "Claude Code" tool - an AI-powered helper that developers use to write and manage code more efficiently - to the npm registry, a popular platform where coders share and download software packages. But buried inside this update was a massive 60 MB debugging file, known as a .map file.

Also Read: Oracle Layoffs 2026: Here's what the early morning termination email read

What was meant to be a simple support tool accidentally included something far bigger: the complete source code of Claude Code itself. That's over 512,000 lines of code, covering everything from how the tool works behind the scenes to its plugins and features. Within minutes, a researcher named Chaofan Shou spotted the unusual file while checking the update. He downloaded it, zipped it up, and shared the link on X (formerly Twitter).

His post quickly caught fire. By the time most people in the US were waking up, the news had spread like wildfire. The leaked code was downloaded thousands of times and forked, copied and hosted on GitHub more than 41,000 times. Anthropic's team scrambled to issue takedown notices under copyright law, but it was too late. The damage was done.

What happened next was even more remarkable. Sigrid Jin, a developer from Korea known as one of the heaviest users of Claude Code (reports say he racked up a whopping 25 billion tokens of usage last year alone), woke up to a flood of notifications. Worried about legal trouble just for having the code on his computer, he decided to take action. In just eight hours, he rewrote the entire tool from scratch in Python, creating a new version called "claw-code." His GitHub repository shot up to 30,000 stars - faster than any project in the platform's history, according to observers.

Also Read: Gmail users update: Google now allows changing old, embarrassing usernames: CEO Sundar Pichai shares step-by-step guide

Not stopping there, Jin then rebuilt it again, this time in the faster Rust programming language. That version has already crossed 49,000 stars. Meanwhile, someone else took the original leaked code and mirrored it on a decentralized storage platform, adding a simple note: "will never be taken down." The code is now out there for good, beyond any single company's control.

The story has sparked a wave of reactions online, with many pointing out the delicious irony. Anthropic had actually built a special feature called "Undercover Mode" into their products - designed specifically to stop their AI from accidentally spilling internal secrets. Yet here they were, leaking their own codebase through a basic packaging mistake. As one user put it in the comments, "They shipped an entire anti-leak system... then leaked their own source code in a .map file. The irony is beautiful."

Others were amazed at the speed of the global developer community. "The real story is the speed here," wrote one commenter. "Not that code leaked, but that the community had it forked, ported to Python, then Rust, and running before Anthropic's PR team finished their coffee."

Another highlighted how this shows a bigger shift: once something is out, it's instantly copied, understood, and re-built - no going back. Some developers dug into the code out of curiosity, while a few questioned the rewrite timelines or wondered if the leak was truly accidental. But the overwhelming buzz on X revolves around one thing: how quickly closed-source tech can become public knowledge in today's connected world.

Claude Code isn't the AI model itself but the command-line interface that helps users interact with it for coding tasks. For everyday Indians following the AI boom, from young engineers in Bengaluru to students in smaller cities dreaming of tech careers, this episode is a reminder of how rapidly innovation moves. One packaging error at 4 AM, and suddenly proprietary secrets are in the hands of thousands. As the dust settles, the code remains widely available, and Anthropic's anti-leak efforts couldn't stop the spread.