When part of the source code for Claude Code, Anthropic’s famous AI agent, was published on GitHub on Tuesday, it was a wild giveaway.
Engineers from all walks of life absorbed it as quickly as possible in hopes of learning from it and perhaps leveraging it to improve their own projects.
Relying on content created by others to improve intelligence may sound familiar, but that’s exactly what major AI companies, including Anthropic, have dabbled in for years as they race to train large-scale language models.
So it was not without some irony that Anthropic was quick to issue a copyright takedown notice on the GitHub repository hosting the code to prevent engineers from accessing the leaked code.
“We have issued a DMCA takedown order against one repository hosting the leaked Claude code source code and forks of it,” an Anthropic spokesperson said, citing the Digital Millennium Copyright Act.
Anthropic, OpenAI, and Google are all facing lawsuits for using copyrighted materials, such as published books, articles, scientific journals, and other content found online, without explicit permission. In response, authors, artists, and publishers have used copyright law to demand accountability and, in some cases, payment.
In September, a court ordered Anthropic to pay $1.5 billion in damages in a class action lawsuit brought by authors and publishers, including lead plaintiffs Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, for allegedly using pirated books and a shadow library to train Claude.
Reddit sued Anthropic last June, accusing it of scraping vast amounts of user-generated content to train models without permission or compensation to users.
And last month, Universal Music Group, Concord, and ABKCO filed a lawsuit against Anthropic for illegally downloading more than 20,000 copyrighted songs and training models.
Now the tables have turned, and Anthropic is relying on copyright law to protect its creations. An Anthropic spokesperson said: “We are taking steps to ensure this never happens again.”
Fortunately for the company, the breach may not be as serious as some thought.
Paul Price, a cybersecurity expert and founder of ethical hacking firm Code Wall, which recently discovered vulnerabilities in McKinsey’s internal chatbot Lili, said the Anthropic leak did not reveal any material information.
“This is more embarrassing than harmful. Most of the really important stuff was in the internal source model, and that didn’t leak,” he told Business Insider.
He said the company had mistakenly exposed its “harness” – the software infrastructure typically used to connect a large language model to the broader context in which it will be used.
“Claude Code is one of the best-designed agent harnesses out there, and this shows you how they approach difficult problems,” Price added, noting that it could also be useful information for competitors.
The leak also highlighted the contradictions in the AI hype cycle. The same tools that allow products to be built and shipped faster than ever before also make information, sensitive or otherwise, vulnerable to leaking, duplication, and dissemination in an instant.