Resources

Checklist

API key exposure checklist for AI tools

A practical checklist for what to do after pasting an API key into ChatGPT, Claude, Cursor, Codex, or another AI tool.

Answer target

What should I do after pasting an API key into an AI tool?

Updated June 25, 2026

Short answer

Treat the key as exposed: rotate or revoke it, update legitimate systems, search for extra copies, review usage, document the change, and move future agent access to scoped runtime resolution instead of pasted raw keys.

What to do now

Use the checklist immediately if a real API key was pasted into an AI chat, prompt, code agent, screenshot, ticket, or log.

  • Revoke or rotate the exposed key.
  • Create the replacement with the narrowest available scope.
  • Update only the legitimate app, service, or agent workflow.
  • Search for copies in repos, logs, tickets, screenshots, shell history, and .env files.
  • Check usage and billing for unusual activity.

Better pattern

For future AI-agent work, store the provider key centrally, grant it to a named member or agent, resolve it at runtime, and record each reveal or retrieval event.

Avoid

Avoid deleting the message and assuming the incident is closed. Avoid creating a broader replacement key or pasting the new key back into the same AI workflow.

Example

If a developer pasted an OpenAI key into a coding-agent chat, rotate the key, update the app or secret record that legitimately uses it, then give future agents scoped access to the stored replacement.

Try this with one key

  1. 1.Store one API key.
  2. 2.Create one agent identity.
  3. 3.Grant only that key.
  4. 4.Resolve it at runtime.
  5. 5.See the audit entry.

No card required.

Use scoped access for the next key

First 10 minutes

Stop using the exposed key, revoke or rotate it at the provider, and prevent the old value from continuing to authorize real requests. If the key controls production access or billing, treat this as the priority before cleanup.

Next hour

Update legitimate systems to the replacement key, check recent usage, and search likely persistence points: local files, shell history, logs, issue comments, PR descriptions, screenshots, shared docs, and agent transcripts.

Before agent work resumes

Decide how the agent will access the key next time. Use a named agent identity, project assignment, direct grant, runtime resolution, and audit trail instead of a copied provider value.

Reusable API key safety checklist

Keep a small repeatable checklist for every incident: contain, replace, clean up, review usage, document, and change the workflow that caused the exposure.

  • Contain: revoke or rotate the exposed key.
  • Replace: update legitimate consumers only.
  • Clean up: remove known copies.
  • Review: inspect usage and billing.
  • Prevent: move agent access to scoped resolution.

Practical workflow

  1. 1ContainRevoke, rotate, or disable the exposed key at the provider.
  2. 2ReplaceCreate a narrower replacement and update only systems that should still have access.
  3. 3SearchLook for copies in the places where AI-tool work usually persists.
  4. 4PreventUse scoped runtime access for the next AI-agent task.

Manual cleanup only vs Checklist plus workflow change

TopicManual cleanup onlyChecklist plus workflow change
Immediate responseDelete the visible message and hope it is enough.Rotate or revoke the key first.
Future accessPaste a replacement key into the same workflow.Grant a named agent runtime access to the stored key.
ReviewRely on memory and chat history.Use audit events for reveal, resolve, revoke, and rotation.

Frequently asked questions

Is deleting the AI chat message enough?

No. The key may already exist in logs, tool output, browser state, or transcript history. Rotate or revoke the key.

What is the fastest safe response to API key exposure?

Revoke or rotate the exposed key, update legitimate consumers, and then investigate where else it may have been copied.

How do I stop this happening again?

Keep provider keys out of prompts and local files. Use named agent identities, direct grants, runtime resolution, and audit history.

Where ScopeHold fits

ScopeHold turns the prevention step into a repeatable workflow: store the replacement key, grant only the required members or agents, resolve at runtime, and review the audit trail.

Sources and reference notes