McHire AI hiring chatbot exposes McDonald's applicant data

McHire.com, the hiring site many McDonald's franchisees used to handle job applications, ran an AI screening chatbot named Olivia that was built by AI software firm Paradox.ai.

Security researchers Ian Carroll and Sam Curry revealed that simple vulnerabilities in the platform could have let virtually any hacker access the records of every chat Olivia had ever had with McDonald's applicants.

What happened

Security researchers Ian Carroll and Sam Curry revealed that they found simple methods to hack into the backend of the AI chatbot platform on McHire.com.

They gained administrator access to a Paradox.ai account by guessing the username and password 123456, and there appeared to be no multifactor authentication for that Paradox.ai login page.

With that access they could query the company's databases that held every McHire user's chats with Olivia.

They also found that they could increment an applicant ID number for their application down to a smaller number and see someone else's chat logs and contact information.

Impact

The data appears to include as many as 64 million records, including applicants' names, email addresses, and phone numbers.

Paradox.ai says that the researchers accessed seven records in total, and five contained personal information of people who had interacted with the McHire site.

Curry said the exposed data could support payroll scams, noting if you wanted to do some sort of payroll scam, this is a good approach.

Response

A spokesperson for Paradox.ai shared a blog post confirming the findings and said the company is instituting a bug bounty program to better catch security vulnerabilities in the future.

Paradox.ai said it had verified that the administrator account with the 123456 password that exposed the information was not accessed by any third party other than the researchers.

McDonald's said it mandated Paradox.ai to remediate the issue immediately once it learned of it, and that it was resolved on the same day it was reported.

Lessons for security leaders

Organizations that outsource hiring to AI chatbot vendors should treat those platforms as part of their own attack surface, including requiring multifactor authentication on every administrator account the vendor operates. Dormant test or support accounts need the same decommissioning discipline as production systems, since a single forgotten login tied to a 2019 test environment opened the door to this exposure. Applicant-facing AI systems that use sequential or guessable record identifiers should be reviewed for insecure direct object reference flaws before launch, not after independent researchers find them.