Back to Blog
    Breach AnalysisInjected AI · Vendor exposure

    McHire: 64 Million Applicant Records Behind an AI Hiring Bot and a '123456' Password

    6 min readAndeavour Research
    123✓5ATTACK CHAINBroken by AgentGuard + AgentHunter
    64M
    applicant chat records potentially exposed

    Key takeaways

    • →Researchers accessed the admin backend of McHire, McDonald's AI hiring assistant built by Paradox.ai, using default credentials.
    • →An insecure API then exposed records for up to 64 million applicants.
    • →AgentGuard inventories vendor AI systems and their data reach. AgentHunter detects the enumeration pattern used to scrape records.
    Attack chain, and where it breaks
    1. 1

      Vendor AI chatbot deployed for hiring

    2. 2

      Test admin account with default password

    3. 3

      Researcher logs into admin backend

    4. 4

      Sequential IDs enumerated via API

      AgentHunter flags sequential record enumeration; AgentGuard surfaces the vendor AI's unmanaged access

    5. 5

      Millions of applicant records exposed

    Affected
    McDonald's job applicants (via vendor platform)
    When
    June–July 2025
    Impact
    Names, emails, phone numbers and chat histories exposed

    What happened

    Security researchers Ian Carroll and Sam Curry reported that the McHire platform's admin interface accepted the default credentials '123456'. From there, an insecure direct object reference in an API let them view applicant chat records just by changing an ID. Paradox.ai fixed the issues quickly after disclosure.

    Why it's an AI governance failure

    The chatbot ('Olivia') processed sensitive personal data at huge scale on behalf of a brand that didn't operate it. That AI system and its surrounding infrastructure sat outside the customer's security inventory.

    How Andeavour would have stopped it

    AgentGuard discovers vendor-operated AI systems handling your data and scores their exposure, including stale test accounts and admin surfaces. AgentHunter detects systematic, sequential enumeration at the edge and stops it long before millions of records are touched.

    The lesson

    Every incident in this series shares one root cause: an AI system acted with real authority, and nothing in the stack was checking whether that action made sense for its task, its Principal or its environment. That's the gap AgentGuard and AgentHunter was built to close, with zero integration and deployment measured in days.

    See AgentGuard + AgentHunter on your own environment

    Zero integration. Live in days. We'll show you what your current stack is missing.