Claude’s source code got leaked — and Anthropic is scrambling to contain it
The accidental release of Claude Code’s source files has opened a window into how Anthropic directs its AI, and how others might now follow
Anthropic has built its reputation on being "the careful AI company". It publishes detailed safety research, employs some of the sharpest minds in the field, and speaks at length about the responsibilities of building powerful technology.
This week, someone forgot to tick a box.
On Tuesday, when Anthropic pushed out version 2.1.88 of its Claude Code software package, it accidentally bundled in a file exposing nearly 2,000 source code files and over 512,000 lines of code.
Keep updated, follow The Business Standard's Google news channel
A security researcher named Chaofan Shou spotted it almost immediately and posted about it on X. Within hours, copies were multiplying across GitHub.
Anthropic's public response was measured, "This was a release packaging issue caused by human error, not a security breach."
However, the AI model itself was not leaked, and no customer data was exposed. The valuable mathematical weight underlying Claude's intelligence remains intact.
What did spill out was the software scaffolding around it — the proprietary techniques, tools, and instructions that tell Claude how to behave as a coding agent. In AI circles, this is called a harness. It is commercially sensitive, and competitors now have it.
By Wednesday morning, Anthropic had filed copyright takedown requests forcing the removal of more than 8,000 copies from GitHub.
That effort triggered its own reaction: a programmer promptly rewrote the leaked functionality in different programming languages to keep the information circulating.
Notably, this is the second leak in a week. Last Thursday, Fortune reported that Anthropic had accidentally exposed nearly 3,000 internal files.
