AI coding assistants might unintentionally expose thousands of secrets to public GitHub repositories, turning seemingly normal developer interactions into a significant enterprise security risk This article explores secrets public github. . Security firm Apiiro uncovered 2,539 credentials spread across nearly 2,000 public GitHub repositories, including sensitive information like database passwords, cloud keys, API tokens, and SSH root logins.
According to a recent report, Apiiro traced some of the leaked data to Claude Code, which inadvertently stored secrets in a hidden file alongside the developer's code.
Specific secrets leaked included a database vendor's admin-level cloud key, a production Azure Postgres password, a Square production token, and the owner password of a medical-records database. Many developers click it without realizing the file exists. Following the publication, Boris Cherny, Anthropic's head of Claude Code, stated in public comments that while Claude Code historically has not performed secrets scrubbing, the company is investigating ways to prevent the secrets exposure described by Apiiro.
This leakage highlights how Claude Code's protection heavily relies on a developer-specific Git setting rather than built-in safeguards, revealing a significant vulnerability that can turn a simple command approval into a persistent credential leak. As AI agents grow integrated into software development, organizations must acknowledge that beyond just the source code, their security protocols must also safeguard agent configuration files, transcripts, and execution histories, which may hold sensitive information.












