Anthropic states that the code for Claude Code, which is proprietary, was accidentally made public. They assure that no client information or login details were compromised.

Despite supporting open-source initiatives, Anthropic has kept Claude Code private, until recently. An update inadvertently included internal code.

Anthropic acknowledged the leak to BleepingComputer and stated that no private or confidential data was disclosed.

“Earlier, a Claude Code release had some internal code. No private client information or login details were involved or exposed. This was a mistake in how the release was packaged, caused by human error, not a security breach. We are implementing safeguards to prevent this from happening again,” Anthropic told Bleepingcomputer.

The leaked code was first discovered by Chaofan Shou (@Fried_rice) and has been widely shared on GitHub and other platforms.

The code was mistakenly leaked when Anthropic briefly published Claude Code version 2.1.88 on NPM.

This version had a 60 MB file called cli.js.map that contained all of the code for the latest version.

A source map file aids in debugging by connecting compiled JavaScript back to its original form.

If the map files contain a field called “sourcesContent,” which includes the full original code within the map, the entire codebase can be rebuilt from the file.

Therefore, including a large .map file in a public package can expose a significant amount of code.

The reconstructed code consists of about 1,900 files, 500,000 lines of code, and details of various Claude-exclusive functions.

While the code is being shared online, Anthropic has started issuing DMCA takedown notices to remove it where possible.

Developers have already started analyzing the code for hidden features and understanding the application’s inner workings.

According to Alex Finn, Anthropic is testing a new “Proactive mode” where Claude codes for you continuously. This was found in the Claude Code code.

Another interesting feature was also noted.

It’s called “Dream” mode, where Claude can constantly think, develop ideas, improve plans, and solve problems in the background.

Separately, users are reporting that Claude’s usage limits have been reduced. This means Pro or Max plan users are reaching usage limits much faster.

The reporting party notes they saw this on their Claude Personal account, which costs $20. After a few messages in a Claude Code terminal, usage increased to 30% and reached 100% after only a few minutes.

That usage was unexpected, especially because the context was small, as the interaction with Claude had just started.

The issue is widespread, and Anthropic has confirmed they are looking into a bug causing limits to be reached quicker.

“We are aware that people are reaching usage limits in Claude Code faster than expected. We are actively investigating and will share more when we have an update,” Anthropic’s Lydia Hallie posted on X.

As of March 31, 2:00 PM ET, the issue is still unresolved, and Anthropic has shared the following update:

“[We are] still working on this. It is the team’s top priority. We know this is affecting a lot of you. More information will be provided as soon as we have it.”

Some users believe it could be an intentional action by Anthropic because Claude’s popularity has increased recently, but confirmation requires more information from the company.