Anthropic exposes Claude: code ends up online due to human error

Anthropic mistakenly released the entire code for its AI model, Claude Code, and part of its software source map online, which could be used to reconstruct the proprietary source code.
The release was attributed to a human error, not a security breach and no sensitive data or login credentials were leaked.
The exposed code dates from version 2.1.88 of the software, released on March 31st.
This incident followed a previous leak that revealed Anthropic is working on an even more powerful AI model named Mythos, which also raised cybersecurity concerns.
Despite the accidental leak, the exposed code could potentially allow competitors to learn about the model’s work and architecture, leveraging this knowledge to craft rival versions.
Anthropic’s spokesperson at first claimed the leaked code was legitimate because it was open source, but later clarified that this was incorrect and it was actually proprietary code.
The files that were leaked were promptly removed from online platforms. Sole24Ore retrieved those files and analyzed them using Anthropic’s own newest model.
Despite Anthropic’s reassurance that the data didn’t contain sensitive details, it raised the question of the possible exposure of confidential project Mythos.

You may have missed