The McHire Morning-After: Hospitality CIOs Scramble to Audit AI Vendors

By 7:14 a.m. yesterday I was on the phone with a CIO at a 600-unit casual-dining group who had not slept. “We don’t use Paradox,” she said, “but I have no idea what my franchisees use.” Then she asked me to hold while she pulled a spreadsheet her HR director had emailed her at 3 a.m. — a list of every recruiting tool any region had ever piped applicant data into. It was forty-one rows long. She had personally approved four of them.

That call is the story. The McDonald’s–Paradox breach broke yesterday, and Paradox patched the McHire endpoint within hours. But the actually revealing thing — the part that should keep hospitality boards up at night — is how few operators ever did meaningful due diligence on the AI tools their HR teams signed up for. McHire wasn’t an exotic shadow-IT install. It was a flagship deployment at the largest restaurant company in the world. And the password was 123456.

The 24-hour scramble nobody planned for

Three CIOs and two VPs of People Ops walked me through what their morning looked like. The pattern was identical: a frantic vendor-risk audit assembled on the fly, mostly by forwarding the TechCrunch piece to a legal counsel who hadn’t reviewed the recruiting stack since procurement signed it. One VP told me her company’s SOC 2 questionnaire for their AI screening vendor came back in 2024 with a single sentence: “Reviewed and acceptable.” No attestation, no penetration test summary, no incident response SLA. The vendor is still in production for 14,000 weekly applicants.

What’s surfacing in these audits isn’t sophisticated supply-chain risk. It’s the basics. Default credentials left on admin panels. Customer-data buckets without object-level encryption. Conversational agents whose transcripts — including Social Security number fragments, work-authorization documents, and disability disclosures — sit in plaintext logs at third-party observability vendors nobody at the chain has ever heard of. The McHire exposure was lawsuit-ready data: applicant PII adjacent to SSNs, the kind of payload that converts a breach into a class action in a week.

Mark interpretation: the operators who get sued in the next twelve months won’t be the ones who picked a bad vendor. They’ll be the ones whose contracts can’t prove they ever asked.

Why the timing is brutal

This story didn’t just break in a vacuum. It broke the same week the EU AI Act’s GPAI Guidelines landed in draft form, sharpening what “deployer” obligations look like for any chain operating in or hiring into the EU. My read — and a forthcoming May piece walked through the enterprise compliance shape of the Act in detail — is that “we trusted the vendor” is about to stop being a defense. The deployer side of the obligation chain is real, and “we didn’t ask” reads in court the same as “we didn’t care.”

It’s also breaking into a market that just told itself a different story. PolyAI’s Series D close earlier this summer — $86M at a billion-dollar valuation for enterprise voice AI — was framed as a maturity moment for hospitality conversational AI. Maturity, apparently, didn’t include making sure the admin login wasn’t 123456. Operators who read that funding round as a green light to accelerate procurement are now reading McHire as a flashing red one.

The chains I spoke with are quietly accelerating in two directions at once: tightening intake on new AI vendors (every CIO I called mentioned moving from a questionnaire to a live penetration test requirement), and pulling forward consolidation onto stacks they can actually audit. Chipotle’s approach — an upcoming May piece detailed their in-house orchestration layer over best-of-breed AI — is suddenly being cited in rooms where six months ago it was dismissed as overbuilt.

The companion piece I filed yesterday on the breach itself is the box score. This one is the standings. By next Friday, half a dozen chains will have rewritten their AI vendor onboarding policy. The other half will find out why they should have when their applicant database shows up on a forum.

The password was 123456. The lesson is that nobody checked.

— Maya covers restaurant tech for TableTransfers. Tips: [email protected].