McDonald's McHire AI Hiring Platform Data Vulnerability (2024)

Incident Summary

In 2024, security researchers identified a critical access control vulnerability in McHire, an AI-powered hiring platform developed by Paradox.ai and deployed by McDonald’s for job applicant screening.^[1]

The researchers discovered that a test account secured with the trivially guessable password “123456” could provide access to a database containing up to 64 million applicant records. The researchers accessed only a small number of records to confirm the vulnerability’s scope and reported the finding through responsible disclosure channels. No evidence of prior unauthorized mass exfiltration was identified.^[2]

Paradox.ai subsequently patched the vulnerability after disclosure.

Key Facts

• Platform: McHire, an AI-driven hiring tool built by Paradox.ai for McDonald’s
• Vulnerability: Test account with password “123456” provided access to applicant database
• Potential scope: Up to 64 million applicant records were potentially accessible
• Actual access: Researchers accessed only a small sample to verify the vulnerability
• Exfiltration evidence: No confirmed mass data exfiltration by unauthorized parties
• Resolution: Vulnerability patched following responsible disclosure

Threat Patterns Involved

Primary: Model Inversion & Data Extraction — Inadequate access controls on an AI hiring platform exposed a large applicant dataset to potential unauthorized access.

Significance

This incident illustrates the data security risks inherent in AI-powered hiring platforms that process large volumes of personal information. Several aspects warrant attention:

1. Scale of potential exposure — The platform’s centralized database of up to 64 million applicant records created a high-value target, amplifying the consequences of any single access control failure.
2. Trivial authentication weakness — The use of a default test credential (“123456”) in a production-adjacent environment reflects fundamental deployment hygiene failures rather than sophisticated attack vectors.
3. Near-miss classification — While the vulnerability was confirmed and the data was technically accessible, the absence of evidence of mass exfiltration distinguishes this from a data breach. It remains a near-miss that could have resulted in significant harm.
4. Third-party AI vendor risk — Organizations deploying third-party AI hiring tools bear responsibility for ensuring those tools meet adequate security standards, particularly when processing sensitive personal data at scale.