If you pasted something you shouldn’t have into ChatGPT, Claude or Gemini, deal with what you pasted first, then the chat. Change a password or key straight away. Call your bank about card details. Tell your employer at once if it was work material or someone else’s personal data.
Then delete the chat and anything the app saved to memory. Deleting hides the chat at once and removes it from the company’s systems within about 30 days, but it cannot undo training that has already happened.
- Treat a pasted secret as known. Change the password or key. Deleting the chat is not enough.
- Work or other people’s data: tell your manager or IT the same day. Whether it must be reported is their call, and the clock may already be running.
- Delete three things: the chat, any saved memory, and any shared link to it.
- Deleting has limits: about 30 days to leave the company’s systems, longer for chats a reviewer has seen, and nothing comes out of a model already trained.
- Next time: run the text through the safe-to-paste checker first.
It is like sending an email to the wrong person. You cannot unsend it. You can make what it contained useless (change the password), warn the people who need to know, and ask for the copy to be deleted.
What should I do first? It depends on what you pasted
The plain point: fix what was exposed before you tidy up the chat. A changed password is safe whether or not the chat is ever deleted.
| What you pasted | Do this now | Then |
|---|---|---|
| A password or login code | Change it, and change it anywhere you reused it | Turn on two-step sign-in; delete the chat |
| An API key or access token | Revoke or rotate the key in the service that issued it | Tell whoever owns the account; check its usage |
| Card or bank details | Call your bank on the number on your card and ask whether to replace the card | Watch your statements |
| Someone else’s personal data | If it was for work: tell your manager or IT today | Delete the chat and any memory |
| Work or client files | Tell your manager or IT today, even if no names were in it | Follow their instructions before deleting anything |
| Your own health details | Delete the chat and any memory | Turn off training; keep future questions general |
Passwords. For a hacked account, the UK’s National Cyber Security Centre (NCSC) says to change the password and also to “change the password for any accounts that are using the same password”, because criminals try a leaked password everywhere. It also recommends two-step verification, which stops someone who knows your password from getting in.1 A pasted password has not been stolen, but this site’s advice is to act as if it had: you cannot know who will see the chat.Reasoning, 13 Oct 2026 — a password typed into a chat is stored by the company and may be read by staff in some cases; changing it costs minutes, so it is treated here like a leaked one.
Keys and tokens. An API key is a password a program uses. OpenAI’s advice for its own keys: “If you believe your key has been leaked, rotate your key immediately”, meaning replace it with a new one so the old one stops working.2 Do the same for any other service’s key.
Card and bank details. The NCSC says to watch for unauthorised purchases and that “You can contact your bank directly for further support”, using the bank’s official contact details, not a number from a message.1
Health details. These are yours, so nobody needs telling. Asking AI about your health shows how to ask without them next time.
Make the secret useless first: a new password, a new key, a call to the bank. Then deal with the chat.
How do I delete the chat in ChatGPT, Claude and Gemini?
The plain point: each app deletes one chat in a few clicks. In ChatGPT and Claude, also check memory, because a saved memory can outlive the chat.
ChatGPT. In your chat history, select the ••• menu next to the chat, then Delete, and confirm.3 Archiving is not deleting: an archived chat stays in your account. Then open Memory in Settings. OpenAI says “Deleting a chat alone does not necessarily delete a separate saved memory created from that chat”, so delete the saved memory too.4
Claude. Hover over the chat in the sidebar, click the ⋮ button, choose Delete, then Delete again.5 Anthropic says that when a chat is deleted, “related memory entries generated from it won’t be removed”, so open Settings, then Memory, find the topic and delete it there.6
Gemini. On gemini.google.com, open Settings & help, then Activity, and click Delete activity item next to the chat.7
Shared links. If you shared the chat, delete the link as well. Google says deleting a shared Gemini chat does not delete its public link (source above). In ChatGPT, anyone who has a link from a personal account can view the chat; delete it under Shared links in Settings.8
If the chat sits in a work account, ask IT before you delete it. Your employer may need it to judge what happened.Reasoning, 13 Oct 2026 — an organisation deciding whether a breach must be reported needs to know what was shared; deleting the only record first can make that harder.
Delete the chat, the memory and any shared link. In a work account, ask IT first.
What does deleting a chat actually do?
The plain point: the chat vanishes from your screen at once and from the company’s systems later. Some copies last longer, and training cannot be undone.
- ChatGPT: removed from view at once, and OpenAI “schedules it for permanent deletion from its systems within 30 days”, unless it was already stripped of details that identify you, or must be kept for security or legal reasons.9 A file you uploaded may also sit in Library; delete it there too.
- Claude: “Removed from your chat history immediately” and “Deleted from our back-end storage systems within 30 days”. A chat flagged by Anthropic’s safety systems can be kept for up to two years.10
- Gemini: Google removes it from view and starts deleting it. But “Chats reviewed by human reviewers … are not deleted when you delete your activity”; they are kept for up to three years, separated from your account.11
Training is one-way. Anthropic says a deleted chat will not be used for future training, but data stays in “models that have been trained” (source above). So deleting quickly helps; deleting late helps less.
Deleting limits the damage. It does not rewind it, which is why changing the password comes first.
How do I stop it being used for training?
The plain point: switch training off now, so the rest of your chats are not used. It does not delete anything.
In ChatGPT: Settings, Data controls, then turn off “Improve the model for everyone”. OpenAI says this setting does not delete chats.12 In Claude: Settings, Privacy, then switch off “Help Improve our AI models”.13 In Gemini: Settings & help, Activity, then turn off Keep Activity. Chats are still kept for up to 72 hours.7 The turn off AI training cheat sheet has every switch on one printable page.
Should I tell my employer or IT?
The plain point: yes, if it was work material or anyone else’s personal data, and the same day. Telling early is what lets your employer meet its own deadlines.
Data protection guidance expects exactly this. The UK regulator, the ICO, asks organisations to make sure staff “know how to escalate a security incident to the appropriate person or team”.14 Your manager, IT team or data protection officer decides what happens next; owning up early beats being found out, as using AI at work explains. Copy and adapt this message:
Report it the same day, say exactly what went in and which account, and let the people responsible decide.
Is it a data breach?
The plain point: it can be. Under the GDPR rules used in the EU and UK, an accidental disclosure of personal data counts as a breach, and some breaches must be reported within 72 hours.
The European Data Protection Board (EDPB), which brings together the EU’s data protection regulators, defines a personal data breach as “a breach of security leading to the accidental or unlawful destruction, loss, alteration, unauthorised disclosure of, or access to, personal data”. Its examples of accidents include “sending an email to the wrong recipient”.15 Pasting customers’ or colleagues’ details into an AI tool your employer has not approved looks much like that, in this site’s reading.Reasoning, 13 Oct 2026 — an unapproved chatbot is an outside company the data was not meant to reach, like a wrong email recipient; whether it is a reportable breach is a judgement for the organisation, not for this page. The Dutch data protection authority, the Autoriteit Persoonsgegevens, says so directly: it warned in August 2024, after several breach reports, that when staff use an AI chatbot against their employer’s rules and enter personal data, that is a data breach.16
The organisation, not you, must report it to the regulator within 72 hours when there is a risk to people, and tell the people affected when the risk is high. The UK’s ICO gives the same 72-hour rule under UK GDPR.14 The EDPB says the clock starts when the organisation is reasonably sure personal data was compromised, which is why your early report matters.15 Outside the EU and UK the rules differ by country and state; your employer’s IT or legal team will know which apply.Reasoning, 13 Oct 2026 — this page cites only EU and UK guidance; other places have their own breach laws, which it does not summarise.
You do not have to decide if it is a breach. You have to tell the people who do, quickly.
How do I avoid it next time?
The plain point: a quick check before you press send is the habit that stops this happening again.
- Check before you paste. The safe-to-paste checker marks emails, phone numbers, card numbers, passwords and keys, in your own browser.
- Swap details for placeholders, such as [CLIENT] or [AMOUNT]. The answer is usually just as useful.
- Use the approved work account for work, never a personal one. Shadow AI explains why the account matters.
- Use a temporary chat for one-off personal questions. In ChatGPT it does not create or update memories.4
Questions people ask
Can I ask the company to delete my data?
Deleting the chat is the quickest route. Where GDPR-style law applies, you can also ask: the ICO says you “can ask an organisation that holds data about you to delete that data”, and in some circumstances it must.17
Should I tell the person whose details I pasted?
At work, leave that to your employer: telling affected people is part of its breach process. Outside work, this site’s advice is to tell them if the details could harm them, such as health or money information, so they can watch for misuse.
Where to go next
- “I use AI tools built into my work apps.” Read AI features in work apps: what they can see.
- “I want the switches on one page.” Print the turn off AI training cheat sheet.
- “Where does a chat actually go?” Read what happens to what you type.
- “What are my company’s rules?” See a team AI policy and AI at work for beginners.
- “I want the whole safety picture.” Read how to use AI safely, or follow the use AI at work safely path.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. You pasted your email password into a chatbot. What does the page say to do first?
- Delete the chat and wait 30 days
- Ask the chatbot to forget it
- Change the password, and anywhere you reused it
Show the answer
Answer: Change the password, and anywhere you reused it. The page says to make the secret useless first: change the password everywhere you used it, then deal with the chat.
2. In ChatGPT, does deleting a chat also delete a memory saved from it?
- Yes, the memory always goes with the chat
- Not necessarily, so delete the saved memory too
- Only when training is switched off
Show the answer
Answer: Not necessarily, so delete the saved memory too. The page quotes OpenAI: deleting a chat alone does not necessarily delete a saved memory created from it.
3. You pasted a client list into an AI tool at work. Who decides if it must be reported as a breach?
- Your employer, so tell your manager or IT that day
- You, after reading the GDPR yourself
- The AI company, once it notices
Show the answer
Answer: Your employer, so tell your manager or IT that day. The page says you do not have to decide; you tell the people responsible quickly, because the organisation reports within 72 hours when needed.
- 1National Cyber Security Centre (UK), Recovering a hacked account, read at source 13 Oct 2026“change the password for any accounts that are using the same password”; “cyber criminals know that many people use the same password for different accounts, and so will try the same ‘hacked’ password across multiple accounts”; “even if a criminal knows your password, they won’t be able to access your accounts” ↑a“Keep a look-out for unauthorised purchases, or check your bank accounts for any unusual transactions. You can contact your bank directly for further support.” ↑b
- 2OpenAI Help Center, Best Practices for API Key Safety, read at source 13 Oct 2026“If you believe your key has been leaked, rotate your key immediately from the API Keys page.”; “A compromised API key allows a person to gain access to your account quota, without your consent.” ↑
- 3OpenAI Help Center, Deleting and archiving chats in ChatGPT, read at source 13 Oct 2026“Select the more options menu (•••) for the chat you want to delete. Select Delete. Confirm the deletion.”; “Archiving removes the chat from the regular sidebar but does not delete it.” ↑
- 4OpenAI Help Center, Memory in ChatGPT, read at source 13 Oct 2026“Deleting a chat alone does not necessarily delete a separate saved memory created from that chat.”; “To remove information saved as a memory, delete the saved memory and the chat where you first shared it.”; “OpenAI may retain logs of deleted saved memories for up to 30 days for safety and debugging purposes.” ↑a“Temporary chats do not create or update memories, even with personalization on.” ↑b
- 5Claude Help Center, Delete or rename a conversation, updated 10 Aug 2026, read at source 13 Oct 2026“Hover over the conversation and click the ‘⋮’ button on the right side.”; “If deleting, click ‘Delete’ again in the confirmation prompt.” ↑
- 6Claude Help Center, Use Claude’s chat search and memory to build on previous context, read at source 13 Oct 2026“When a conversation expires or is deleted, related memory entries generated from it won’t be removed, but you can delete individual memories at any time.”; “See exactly what Claude remembers about you in Settings > Memory.” ↑
- 7Gemini Apps Help, Manage & delete your activity in Gemini Apps, read at source 13 Oct 2026“Click Settings & help Activity”; “A specific activity: Next to the activity, click Delete activity item”; “If a shared chat is deleted, it doesn’t delete the related public link.” ↑a“Near the top, click On Turn off or Turn off and delete activity.”; “Even when Keep Activity is off, your conversations will be saved with your account for up to 72 hours to allow Google to provide the service and process any feedback.” ↑b
- 8OpenAI Help Center, Sharing conversations and scheduled tasks in ChatGPT, read at source 13 Oct 2026“Anyone who has a link created from a personal account can view the shared conversation or scheduled task.”; “Deleting a shared link stops future access to that link but does not delete the original conversation or task. It also does not remove a conversation or task another person already saved.” ↑
- 9OpenAI Help Center, Chat and file retention in ChatGPT, read at source 13 Oct 2026“OpenAI schedules it for permanent deletion from its systems within 30 days, unless the chat was already de-identified and disassociated from your account or OpenAI must retain it longer for security or legal obligations.”; “Deleting a chat does not delete a file that remains saved in Library.” ↑
- 10Anthropic Privacy Center, How long do you store my data?, dated 1 Jul 2026, read at source 13 Oct 2026“Removed from your chat history immediately”; “Deleted from our back-end storage systems within 30 days”; “We retain inputs and outputs for up to 2 years and trust and safety classification scores for up to 7 years if your chat or session is flagged by our automated trust and safety systems as violating our Usage Policy.”; “Your data will still be included in model training runs that are already in progress, or in models that have been trained.” ↑
- 11Gemini Apps Help, Gemini Apps Privacy Hub, read at source 13 Oct 2026“Chats reviewed by human reviewers (and related data like your language, device type, location info, or feedback) are not deleted when you delete your activity. Instead, they are retained for up to three years.”; “Please don’t enter confidential information that you wouldn’t want a reviewer to see” ↑
- 12OpenAI Help Center, Data controls in ChatGPT, read at source 13 Oct 2026“When Improve the model for everyone is off, your new conversations won’t be used to train OpenAI models.”; “Does turning off model training delete my chats? No. Delete a chat or your account separately when you want to remove content.” ↑
- 13Anthropic Privacy Center, How do I change my model improvement privacy settings?, read at source 13 Oct 2026“Select Privacy”; “Under Help Improve our AI models, click the button to toggle it off/on” ↑
- 14Information Commissioner’s Office (UK), Personal data breaches: a guide, read at source 13 Oct 2026“Our staff know how to escalate a security incident to the appropriate person or team in our organisation to determine whether a breach has occurred.” ↑a“You must do this within 72 hours of becoming aware of the breach, where feasible.”; “If the breach is likely to result in a high risk of adversely affecting individuals’ rights and freedoms, you must also inform those individuals without undue delay.” ↑b
- 15European Data Protection Board, Data breaches | Data protection guide for small business, read at source 13 Oct 2026“a breach of security leading to the accidental or unlawful destruction, loss, alteration, unauthorised disclosure of, or access to, personal data”; “personal data breaches include security incidents that are the result of both accidents (such as sending an email to the wrong recipient, losing a USB key containing customer data, or accidentally deleting medical data for which no backup is available)”; “notification of any personal data breach to the relevant data protection authority (DPA) within 72 hours, unless it is unlikely to result in a risk to individuals” ↑a“An organisation should be regarded as having become ‘aware’ when there is a reasonable degree of certainty that a security incident has occurred and compromised personal data.” ↑b
- 16Autoriteit Persoonsgegevens (Netherlands), Let op: gebruik AI-chatbot kan leiden tot datalekken (“Watch out: using an AI chatbot can lead to data breaches”), 6 Aug 2024, read at source 13 Oct 2026 in a browser, because the site refuses automated fetches: “als daarbij persoonsgegevens zijn ingevoerd, dan is sprake van een datalek” (if personal data was entered in doing so, there is a data breach).↑
- 17Information Commissioner’s Office (UK), Your right to get your data deleted, read at source 13 Oct 2026“You can ask an organisation that holds data about you to delete that data. In some circumstances, they must then do so.” ↑
Read at source 13 Oct 2026: OpenAI Help Center (deleting chats, retention, memory, shared links, data controls, API key safety); Anthropic Privacy Center and Claude Help Center (retention, model improvement setting, deleting a conversation, memory); Gemini Apps Help (activity, privacy hub).
Read at source 13 Oct 2026: National Cyber Security Centre, Recovering a hacked account; Information Commissioner’s Office, Personal data breaches: a guide and Your right to get your data deleted; European Data Protection Board, Data breaches.
Menus move: if a step does not match your screen, follow the vendor’s help page. This page is general information, not legal advice. The wrong-email comparison, the order of steps and the message template are this site’s own, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK