Earlier this week the developer community was shaken by reports that Anthropic’s Claude Code AI assistant was unintentionally publishing sensitive credentials on public GitHub repositories.
Claude Code refers to Anthropic’s AI‑powered code generation tool that integrates directly with GitHub to suggest and execute code snippets.
What Triggered the Leak?
The root cause was a “Do not ask again” prompt that, once approved, allowed Claude Code to run bash commands without further confirmation. When those commands contained passwords, API keys, or other secrets, the tool stored the raw command output in the repository’s settings.local.json file, exposing the data to anyone who could view the repo.
Scale of the Exposure
According to public reports, 2,539 Claude Code secrets leaked on GitHub across 1,994 repositories after the faulty allow‑list handling was discovered.1 The issue was first logged in GitHub issue #5544 on August 11, 2025, where contributors highlighted that approved commands were saved verbatim in plain text.
How Developers Can Prevent Similar Leaks
- Review and restrict the “Do not ask again” setting. Keep interactive approvals enabled for any command that could handle credentials.
- Use secret‑management tools. Store API keys and passwords in vaults (e.g., HashiCorp Vault, AWS Secrets Manager) rather than embedding them in code.
- Enable GitHub secret scanning. Activate automated scans that flag accidental credential commits.
- Audit CI/CD pipelines. Ensure that any AI‑generated scripts are reviewed before execution.
Implications for Enterprises and Cloud DevOps
Enterprises that rely on AI‑assisted development must treat the Claude Code incident as a reminder that convenience can introduce new attack surfaces. Integrating AI tools into production pipelines without proper guardrails can lead to:
- Unintended credential exposure across multiple environments.
- Compliance violations under standards such as ISO 27001 or GDPR.
- Increased remediation costs and brand damage.
Frequently Asked Questions
Is the leak limited to Claude Code users?
No. Any GitHub repository that enabled the “Do not ask again” option for Claude Code could have been affected, regardless of the project’s size.
Can the leaked secrets be revoked?
Yes. Immediately rotate passwords, regenerate API tokens, and invalidate any compromised keys.
Does Anthropic plan to fix the issue?
Anthropic has acknowledged the bug and promised an update that will redact secrets from stored settings files and enforce stricter command approvals.
Should I stop using AI coding assistants?
Not necessarily. Use them with proper security policies, regular code reviews, and secret‑management practices to mitigate risk.
How can I detect if my repo has been impacted?
Run a search for common secret patterns (e.g., “aws_secret_access_key”, “password=”) and review the settings.local.json file for unexpected entries.
Neptune Infotech can help you build secure, AI‑enhanced development pipelines that protect your intellectual property and data.
