Skip to content
News

GUIDES · October 29, 2026 · 7 min read

Someone Pasted Customer Data Into ChatGPT. Now What?

It has happened, somebody has told you, and the instinct is to delete the conversation. That is the one step that does almost nothing. Here is a sequence that does.

First: establish what actually went

Before anything else, get the text. Not a summary of it, and not the person's recollection of it — the actual content, from the conversation if it still exists.

  • •What fields were in it, and how many people do they identify?
  • •Which account was it — corporate, or personal? This decides almost everything that follows.
  • •Which tool, on which plan? Training settings and retention differ enormously between a free account and an enterprise one.
  • •When, and has it been used since in the same conversation? Context carries forward.

Do not delete the conversation until you have preserved what was in it. You need it for the assessment and possibly for a regulator, and once it is gone you are reconstructing from memory.

Second: understand what deleting does

Delete removes your copy. The disclosure already happened.

Removing a chat clears it from the user's history. The provider's retention runs on its own schedule — abuse-monitoring copies, backups, logs — and none of those are addressed by your delete button. If the account had training enabled, the content may already have been used.

So deletion is a reasonable step, not a remedy. Treat it as tidying rather than as containment, and do not let it end the conversation internally.

Third: the notification question

Under GDPR the test is risk to the people in the data, not embarrassment to you. A personal-account paste of a customer list is a personal data breach even though nothing was hacked: the data went somewhere it was not authorised to go.

Seventy-two hours runs from awareness, not from certainty. If you are still investigating at hour sixty, notify with what you have. Regulators are markedly more forgiving of an incomplete early report than of a late complete one.

Whether the individuals need telling is a separate and higher bar — high risk to them rather than any risk. Take advice; this is the part where instinct is least reliable.

Fourth: the part everybody skips

Ask how often this is happening rather than treating it as an incident. The overwhelming majority of pasting into AI tools happens on personal accounts, which means the one you heard about is the one somebody reported.82% — the share of those pastes made from unmanaged, personal accounts — outside any enterprise oversight. LayerX, Enterprise AI & SaaS Data Security Report

If the response is disciplinary, you have guaranteed that the next one is not reported. The people who tell you are the ones keeping you informed.

Fifth: make the next one harmless

The durable fix is not a stronger rule. It is removing the identifying detail before anything is sent, so that the same mistake next month sends a document about somebody who does not exist.

That is what this product does, and it is the honest reason it exists: people will paste, the instruction not to has been tried, and the only control that survives a human in a hurry is one that does not require them to remember.

Try Secure AI free

Frontier AI chat — encrypted by default, with your identity stripped before any request reaches the provider.

Common questions

Is pasting customer data into ChatGPT a reportable breach?

Under GDPR it can be, and often is: a personal data breach includes unauthorised disclosure, not only hacking. The test is risk to the individuals. Sending customer records to a third party you have no agreement with is a disclosure, and a personal account makes it harder to argue otherwise.

Does deleting the conversation fix it?

No. It removes your copy. The provider's retention schedule — logs, backups, abuse-monitoring — is unaffected, and if training was enabled on that account the content may already have been used. Preserve the text before deleting anything; you will need it for the assessment.

How long do we have to report?

Under GDPR, 72 hours from becoming aware, not from establishing the facts. If the investigation is unfinished, notify with what you have and supplement later. An incomplete early report is treated far better than a late complete one.

Should the employee be disciplined?

Consider what that buys. The person who told you is the reason you know. A disciplinary response to a self-reported paste reliably produces a workplace where the next one is not reported, and the underlying behaviour is near-universal rather than exceptional.

Related

More from News
Secure AI

Use any AI. Stay anonymous.

© 2026 Secure Artificial Intelligence, Inc.Private details never reach the AI.
Someone Pasted Customer Data Into ChatGPT. Now What? · Secure AI