What the McHire breach should change about how you buy hiring AI

In the summer of 2025, two security researchers logged into the admin console of McHire — the AI hiring platform used by McDonald's franchisees, built by Paradox.ai — using a test account whose username and password were both "123456." The account had been sitting unused since 2019. From there, a second flaw (sequential applicant IDs in an internal API) let them page through the chat records of roughly 64 million job applicants: names, email addresses, phone numbers, and IP addresses, along with everything those applicants had typed to "Olivia," the hiring chatbot. Social Security numbers were not exposed.

The facts, because they matter more than the headline: researchers Ian Carroll and Sam Curry disclosed the flaws on June 30, 2025. Paradox and McDonald's acknowledged the report within hours, the default credential was disabled and the endpoint secured within about a day, and Paradox subsequently launched a bug-bounty program. There is no public evidence anyone malicious accessed the data first. (Reporting: Krebs on Security, CSO Online.)

So this isn't a story about a villain. It's a story about what 64 million people's data was protected by until someone checked: a default password on a forgotten account.

Why this lands on recruiters, not just CISOs

Every AI screening tool — ours included — sits on top of exactly the data that was exposed here: contact details, employment answers, sometimes resumes and transcripts of everything a candidate said. Candidates hand it over because they trust you, not your vendor. When it leaks, it's your agency's name in the candidate's inbox.

And candidates are already primed to walk. In Greenhouse's 2026 Candidate AI Interview Report (April 2026; 2,950 job seekers across five countries), 38% of US candidates had already withdrawn from a hiring process because it included an AI interview, and lack of disclosure about how AI is used was one of the top reasons. Trust is the scarce resource. A breach spends it for everyone.

Most recruiting teams buying these tools don't have a security engineer to run diligence. So here's the checklist we'd want every buyer to run — on any vendor, including us.

The 8-question vendor security checklist

1. "Have you had a breach or a disclosed vulnerability? What happened next?" A vendor with a disclosure story and a fast fix (as Paradox ultimately demonstrated) is more credible than one claiming a spotless record it can't evidence. You're listening for specifics, not adjectives.
2. "Do you run a bug-bounty or vulnerability-disclosure program?" McHire's flaw was found by outside researchers acting in good faith. Vendors that invite that scrutiny find out about default passwords before the press does.
3. "What candidate data do you store, and for how long?" Exact fields, exact retention. The McHire exposure was so large partly because years of applicant chats were retained and reachable. Less stored = less to lose.
4. "Who can access candidate data, and how is that access controlled?" Listen for: no shared accounts, no default credentials, MFA on admin access, audit logs. A test account from 2019 with "123456" fails every one of these.
5. "Are candidate-facing records protected against enumeration?" The second McHire flaw was sequential IDs — change a number, see another person's file. Ask whether record identifiers are guessable.
6. "What did your last independent security review cover, and when?" Pentest or audit, with a date. You don't need to read the report; you need to know one exists and is recent.
7. "How would you notify us — and our candidates — if something happened?" Concrete commitment with a timeframe, in the contract, not a values statement.
8. "What happens to candidate data when we leave?" Deletion on offboarding, in writing.

Print it, paste it into your next vendor call, and use it on us too. Holding vendors to the standard is the point — we'd rather win on the answers than on the absence of questions.

What we do with this at Vettika

We're not going to claim invulnerability — the lesson of McHire is that nobody should. What we commit to in design: candidates always know they're talking to an AI; every interview produces a transcript the recruiter can see in full; and we evaluate only the content of candidates' answers. On security specifics, we hold ourselves to the checklist above and answer all eight questions for any prospect who asks.

All facts as of June 11, 2026. Litigation and incident details reported from public sources linked above.