AI prompts leaked: what it means and what to do

Risk level: medium. Can you change it: no. Found in 2 breaches in this directory.

What this data is

The text you typed into an AI assistant, along with the replies, stored as conversation history by the provider. People paste far more into these boxes than they realise.

The risk on its own

Prompts are a diary written without an audience in mind. They routinely contain health worries, relationship problems, job applications, legal questions, business plans and pasted documents. Even with your name stripped off, the content itself often identifies you within a few lines.

The risk combined with other data

Attached to your email address or account name it becomes a private window into your life that anyone can read. That is extortion material and also a research tool: an attacker learns your employer, your projects, your worries and your writing style. Pasted credentials or keys in old prompts are worse again, because those still work.

How criminals use it

  • An extortion email quotes a genuinely private prompt back to you and demands payment to keep it quiet.
  • Business secrets pasted for summarising, contracts, salary lists, customer data, end up readable by a competitor.
  • An attacker harvests passwords and keys that people pasted into prompts while debugging.
  • Your writing style and personal detail are used to craft a message that sounds exactly like you to your colleagues.

What to do now

  1. Open the assistant settings and delete your conversation history, then switch off history or training retention if the product offers it.
  2. Rotate any password, key or account number you ever pasted into a chat, treating it as public now.
  3. Tell your employer if work material was involved, because that may be a reportable incident on their side.
  4. Change the account password and turn on two-factor authentication with an authenticator app.
  5. From now on keep names, account numbers and credentials out of prompts, and use placeholders instead.

Frequently asked questions

I never put anything sensitive in a chatbot. Am I fine?

Check before you assume. Most people find old prompts containing an address, a medical question, a draft resignation letter or a pasted document.

Does deleting my history now help?

It does not recall what already leaked, but it shrinks what a future breach can expose, which is the part you still control.

Can leaked prompts be traced back to me?

Often yes. Prompts contain names, places and specifics, so even an anonymised export can usually be matched to a person by the content alone.

Breaches that exposed this data

  • Muah.AI: 2024-09-17, 1.9M accounts, AI prompts, Email addresses, Sexual fetishes
  • KomikoAI: 2026-02-25, 1.1M accounts, AI prompts, Email addresses, Forum posts, Names

Tools that help

Related guides

Breach data sourced from Have I Been Pwned, used under CC BY 4.0. Additional breach metadata provided by XposedOrNot. DataExposed is an independent service and is not affiliated with or endorsed by either.