Code assistant is also an attack surface: how not to leak secrets and keys

AI tools in development increase productivity but create new paths for key leakage and unintended execution.

How leakage happens

Pasting .env in chat, including credentials in issues, giving access to folders with dumps or backups. The assistant can pass context to other users or logs.

Minimum protections (low effort)

Never paste .env or tokens in chat or issues. Use secret scanning in the repo and in CI. Separate data by classification: public vs internal vs sensitive. Block folders (dumps, backups, credentials) from the assistant's context.

Checklist for the team

"What can the AI see?" "What can it execute?" "Who approves PRs with infra changes?" AI is like a very fast intern: useful, but needs rules and supervision.

Key takeaways

Treat the assistant as an attack surface. Secret scanning, data separation and approval rules reduce risk.

Read also

• Legal and license risk: the side almost no one considers when using AI in code
• Code review with AI + RAG: how to review intent, not just style

FAQ

What if I need to debug with real data? Use anonymized data or isolated environments; never paste production into chat.

Does the secret scanner get in the way? Configure exceptions only for mock tests and docs, never for real credentials.