Disclosure up front: I make Outbound Veil, a Mac menu-bar app that checks what you type for sensitive text. It appears in step 5. I read the provider pages below on November 18, 2026. They change, so check them for your account.
1. Rotate the key or change the password first
Replace the secret before you do anything else, because deleting the chat does not make a pasted key safe and the old key keeps working until you turn it off.
- OpenAI API key. OpenAI says to rotate a key you believe has leaked, right away, from the API Keys page. Create a replacement, update your apps, then revoke the old key (Best practices for API key safety).
- AWS access key. Create a second key, move your applications to it, check when the old key was last used, then deactivate it and delete it. AWS says that if you lose a secret access key you must delete the key and create a new one (Update access keys).
- GitHub token. Open your profile picture, then Settings, Developer settings, Personal access tokens. Choose Fine-grained tokens or Tokens (classic) and select Delete beside the token. Deleting a token that created a deploy key also deletes that key (Managing your personal access tokens).
- Password. Change it at the service, and change it everywhere else you used the same one.
2. Delete the chat
Delete the chat. OpenAI says it is removed from your view at once and scheduled for permanent deletion within 30 days. The exceptions are data already de-identified or held for security or legal reasons (Deleting and archiving chats). Treat the secret as exposed anyway, which is why step 1 comes before this one.
3. Unshare any link to that chat
Anyone with a shared link to a personal-account chat can read it. OpenAI says deleting a conversation removes its shared link. To check, go to Settings, then Data controls, then Shared links, then Manage. Select Delete shared link for one chat, or Delete all shared links under More actions. Deleting a link stops new access but does not remove a copy someone already saved (Sharing conversations).
4. Check the access logs for use of the key
- OpenAI. Review the Usage page for activity that does not match your own work (same page as step 1).
- AWS. CloudTrail event history is a record of the past 90 days of management events in a Region, from the console, the command line, the SDKs and the APIs. You can search it with
aws cloudtrail lookup-events(Working with CloudTrail event history). The old key's last-used information, from step 1, is a quick first look. - GitHub. Your account's security log lists the actions on it from the last 90 days. Find it under Settings, then Archives, then Security log (Reviewing your security log).
If you see use you do not recognize, treat the account as compromised and follow the provider's own steps, which I have not covered here.
5. Check before you paste
Most pasted secrets sit inside a config file or a log that you copied for another reason. A checker that runs where you type can flag them before you press send. Outbound Veil runs on your Mac, and the checking is done locally. It will miss some things, as every tool does, so read the list below before you rely on it.
On November 18, 2026, I ran the detection code from release 1.6.0 on made-up values only (plus AWS's documented example key ID). I did not use a real key. This tested the detector on text. I did not launch the app or type into a text field, so it says nothing about how the app reads a particular field. The compatibility page covers the fields I tested earlier.
- Flagged as a secret key: AWS access key IDs, including temporary ones, GitHub tokens (classic and fine-grained), OpenAI keys, and bearer tokens. These were caught alone, in a sentence and after a name such as
GITHUB_TOKEN=. - Flagged as a password: text written as
password=...orDB_PASSWORD=..., and a password inside a database address. - Flagged only with a name beside it: an AWS secret access key. It was caught after
AWS_SECRET_ACCESS_KEY=, in a credentials file and in JSON. On its own, or in a sentence, it was not. - Missed: a password written in a plain sentence such as "my password is ...", and a password shorter than eight characters.
I left out other providers, other key formats, and private keys, because I did not test them.
The video AWS access key in a config you paste into ChatGPT on Mac shows how to find a key in a config file with grep and hide it with sed before you paste. Outbound Veil has its own product page, and the wider checklist, including what ChatGPT and Claude say they keep, is in the client-notes guide.
Related guides
- Before you paste client notes into ChatGPT or Claude: a checklist for Mac users: account type, training settings and what to remove first.
- Best tools to redact personal information before you use ChatGPT and other AI chats (2026): the tools compared.
- Where Outbound Veil works: every app and field tested, and the ones that do not work.
- Outbound Veil, all products and all guides.