Anthropic accidentally released the source code for Claude Code. For those not in the know, Claude Code is a tool that many of us are using to generate code, troubleshoot issues, and act as high-level project managers rather than writing every line manually. Some find this to be a positive development, while others do not; however, that is a discussion for another day.
The Register is where I first heard the news, and they explained the situation quite well. To make a long story short, Anthropic provided the following statement to the publication:
“Earlier today, a Claude Code release included some internal source code,” an Anthropic spokesperson told us in an email, adding that no customer data or credentials were involved or exposed. “This was a release packaging issue caused by human error, not a security breach. We’re rolling out measures to prevent this from happening again.”
We won’t delve into the technical details here, as thousands of others have already done so. To my knowledge, the first person to post the code to GitHub was ultraworkers/claw-code, which now has over 101k forks. This means a significant number of people now possess the source code for that specific version of Claude Code.
According to The Register, it appears Claude Code 2.1.90 was released without notice on the official Anthropic GitHub repository. Looking at that repo, however, there is no mention of a potential software leak or any recommendation to apply a patch.
While this is a significant mistake—and I am sure the team is as embarrassed as anyone would be—we are all human, and to err is human. However, that doesn’t mean we shouldn’t strive to double-check our work. One thing I have learned while “vibe coding” is to tell the Agent you are working with: “Perform a security audit,” “Find holes in the code,” or “Conduct a security posture audit.”
Interestingly, on March 6, 2026, Anthropic announced they were partnering with Mozilla to improve Firefox’s security. In that instance, Claude Code identified approximately 22 CVEs within Mozilla’s code. My advice? Use the tools that Anthropic deems great for others on your own internal projects. If it works for Mozilla, Anthropic should give it a try. They should practice what they preach.
Currently, if you search for “Claude Code” on GitHub, you will see a flood of results ranging from “Clawd Code” to various users who managed to fork the project quickly. I highly advise anyone downloading these forks to inspect the code for malware—or simply avoid using them altogether. Either way, the choice is yours. While there are likely many “clean” copies available, I personally do not want to take that risk.
So, is Anthropic going to pretend this never happened? It certainly seems that way.