On June 30, 2025, two security researchers — Ian Carroll and Sam Curry — decided to investigate the McDonald’s job application system.
The system was called McHire. It was powered by Paradox.ai, a conversational AI hiring platform. The AI persona that greeted applicants was named Olivia. She would ask them questions, screen their qualifications, and route promising candidates to the appropriate McDonald’s franchise for follow-up.
Carroll and Curry, in a matter of hours, discovered that the administrative backend of the platform was protected by admin credentials of 123456/123456. Yes. Six ones. Twice.
They then discovered an Insecure Direct Object Reference (IDOR) vulnerability that allowed them to enumerate applicant records by sequentially incrementing an ID number in the URL.
By the end of the day, they had established that approximately 64 million applicant records — names, contact information, application data, screening transcripts — were accessible with essentially no security.
They responsibly disclosed the vulnerability. It was patched.
Paradox.ai’s Chief People Officer, Stephanie King, was widely quoted in the subsequent coverage with a three-word response: “We own this.”
The operational layer
This is the second of three pieces on the McDonald’s AI trilogy. Part one addressed the information layer. This piece addresses the operational layer: what happens when AI-mediated business processes are deployed at scale by teams that have not built the operational discipline required to support them.
The McHire breach was not a failure of AI capability. Olivia, the conversational AI, worked as designed. The failure was operational. Basic security hygiene — non-default admin credentials, protection against IDOR vulnerabilities, standard access controls — was absent.
The specific failure mode: an AI-powered product had been built and sold and deployed at massive scale before the operational discipline required to run it responsibly had been developed.
Why this pattern is generalizable
The McHire breach is a specific case study of a general pattern. AI capabilities are, in 2026, dramatically easier to deploy than the operational infrastructure required to run them responsibly at scale.
An AI-powered customer service system can be deployed by a small team in weeks. The operational infrastructure to run it — monitoring for prompt injection, tracking hallucination rates, managing conversation escalation, handling regulatory compliance across jurisdictions — requires substantial ongoing investment that most deploying organizations have not funded.
An AI-powered document processing system can be deployed by a small team in weeks. The operational infrastructure to audit its decisions, catch systematic errors, maintain data lineage, and defend against regulatory challenge requires ongoing investment that most deploying organizations have not budgeted.
An AI-powered hiring platform can be deployed by a small team in weeks. The operational infrastructure to protect 64 million applicant records requires security discipline that Paradox.ai apparently had not developed.
Three practical questions
One: for every AI capability your organization has deployed, has the operational infrastructure required to run it responsibly been funded at proportional scale? In most organizations, the answer is no. The capability is running at 10x the scale that the operational discipline was designed to support.
Two: what is your organization’s equivalent of the 123456 admin password? The specific operational shortcut that would embarrass you publicly if it were discovered by a security researcher tomorrow. It exists. Find it now, or wait for the researchers to find it.
Three: who owns operational responsibility for each of your AI deployments, and do they have the authority and budget to actually operate them responsibly? If ownership is unclear or under-resourced, the breach is a matter of when, not if.
The closing thought
Stephanie King’s three-word response — we own this — is the sharpest available executive framing for what accountability looks like in this moment. She did not deflect. She did not blame the researchers for finding the vulnerability. She did not claim the breach was sophisticated. She said her company owned it.
That level of executive accountability is the operational discipline that AI deployment at scale actually requires. The organizations that develop it will produce durable AI-enabled businesses. The organizations that do not will produce their own version of the McHire breach — and their own three-word response, which will hopefully be as direct as Stephanie King’s was.
Part three of this trilogy addresses the architectural layer: why the specific technical patterns underneath AI deployments create new categories of risk that traditional security frameworks cannot address.
The world has changed. The leaders who notice will be the ones the next decade is built around.
