← All Labs
🛡 AI HACKING INTERMEDIATE +95 XP · +38 no-hint bonus

AI Chat Share-Link IDOR — Reading Other Users' Conversations

ChatNova lets you share a conversation with a link. The share page serves any conversation by ID — shared or not — and the IDs aren't random. Conversations with an AI assistant often hold the most sensitive things people type all day.

https://bookshop.local/search

What is a chat share-link IDOR?

IDOR means insecure direct object reference. The application hands out an ID for a record and returns whichever record matches the ID in the request, without checking that the person asking is allowed to see it.

In a chat product the record is a conversation. If share links use guessable IDs and the server does not check that a conversation was actually shared, changing the number shows other people's chats, which often contain names, documents and credentials.

A real case

McDonald's hiring chatbot exposed applicant chatsJuly 2025

Researchers Ian Carroll and Sam Curry found that the McHire platform, built by Paradox.ai, had a test admin account protected by the password 123456. An internal API then let them read other applicants' chat records by changing an ID number, covering up to 64 million applications.

This lab reproduces the second half of that finding: an endpoint that returns any conversation whose number you ask for.

Source: Ian Carroll

How to fix it

  • Check on every request that the conversation is marked as shared, or belongs to the person asking.
  • Use long random identifiers for share links, never a counter.
  • Let users see and revoke their shared links, and expire them by default.
  • Rate-limit and alert on someone requesting many IDs in sequence.

Questions people ask

What is an IDOR?

An insecure direct object reference. It is an access-control bug where changing an identifier in a request returns data belonging to someone else.

Does a random ID fix it?

It makes guessing harder but does not replace the access check. A link that leaks can still be opened by anyone unless the server checks permission.

Why does this matter more for AI chat apps?

People paste sensitive material into chats: contracts, code, health questions, passwords. A leaked conversation is often worse than a leaked profile.

Read the full OWASP LLM Top 10 guide · All free labs · AI red team certification