McDonald's McHire chatbot admin login used password 123456, exposing applicant data

McHire is McDonald's AI-driven hiring platform, built by Paradox.ai, in which a chatbot named Olivia screens and engages job applicants. According to a CSO Online report summarised by The Economic Times, security researchers Ian Carroll and Sam Curry looked into it in late June 2025 after Reddit users complained that Olivia was giving strange or nonsensical responses. They noticed a login option labelled 'Paradox team members' on McHire's admin interface and tried the default username and password '123456'. It worked.

The login gave them access not only to a test environment but to real administrative dashboards with live data. They then found an internal API endpoint that fetched applicant records using a predictable parameter, an insecure direct object reference. Through it they could view applicants' chat transcripts with Olivia, names, email addresses, phone numbers, application details and even tokens that could be used to impersonate a candidate. The reporting said the exposure could reach as many as 64 million job seekers.

After disclosure on 30 June, Paradox.ai and McDonald's acknowledged the flaw quickly. By 1 July the default credentials were disabled and the endpoint secured, and Paradox said it would run further security audits. Paradox also said, based on its records, that the test account had not been accessed by anyone other than the researchers, that no candidate information was leaked online, and that only five candidates' information was viewed.

Records of as many as 64M applicants were reachable; Paradox says only five candidates' data was viewed, by the researchers. No fine is reported.

A default administrator account with the password '123456' was still live and reachable from a public login page.

The account was not limited to a test environment: it opened real dashboards with live applicant data.

An API endpoint returned applicant records for predictable identifiers without checking who was asking.

The hiring chatbot was deployed at scale with elementary security gaps, which analysts quoted in the report tied to rushing AI tools into production.

An AI front end doesn't change the basics: delete default test accounts and check object-level access, or the chatbot's database is one guess away.

The credentials were disabled and the endpoint secured by 1 July 2025, within a day of disclosure. Paradox says the exposure affected one organization and that only the researchers viewed data. The Economic Times report mentions no fine or lawsuit; experts quoted warned the data could fuel phishing and social engineering.