Using AI/ML to Find and Remediate Enterprise Secrets in Code and Document Sharing Platforms

The inadvertent exposure of sensitive information such as credentials, API tokens, and private keys in code and document sharing platforms (DSP) poses significant security risks, especially for financial institutions. Traditional heuristic-based tools often generate high false positives, complicating remediation efforts. This paper presents two AI-driven models to enhance secret detection: one for reducing false positives in code repositories and another for minimizing false negatives in DSP. By leveraging logistic regression with character n-grams and XGBoost, our models demonstrate improved accuracy and efficiency over existing methods, significantly reducing the burden on developers. The results indicate that AI-driven approaches can effectively identify and address potential leaks of sensitive information, providing a more secure environment for developers.