Redbot Labs / Threat Intelligence
McHire AI Hiring Platform Exposes Applicant Data Through Weak Access Controls
- Incident ID
- RBT-TI-0082
- Organization
- McDonald's / Paradox
- Reported Location
- Chicago, Illinois, United States
- Severity
- High
- Category
- AI Application Data Exposure
- Record Date
- 2025-07-09
Incident Summary
Security researchers discovered serious authentication and authorization flaws in McHire, the AI-powered hiring platform used by McDonald's franchisees.
The platform's Olivia conversational AI assistant collects applicant information, hiring preferences, application responses, and chat history as candidates move through the recruiting process.
Researchers found a legacy administrative account protected by the credentials 123456 / 123456 and then identified an insecure direct object reference in an internal API.
By changing an applicant identifier, the researchers were able to retrieve another applicant's personal information and raw Olivia conversation data from the live production environment.
Paradox, which operates the McHire platform, revoked the legacy credentials and patched the API after disclosure. The company later stated that seven applicant records were accessed during the researchers' validation, five of which contained U.S. candidate information, and that its logs showed no access by anyone other than the researchers.
What Happened
Security researchers Ian Carroll and Sam Curry examined the McHire recruitment platform used by McDonald’s franchise locations.
McHire uses Paradox’s Olivia conversational AI assistant to interact with job applicants, collect information, answer hiring questions, screen candidates, and coordinate interviews.
The researchers discovered an administrative interface containing a legacy account configured with the username 123456 and password 123456.
After authenticating to the administrative environment, they identified an authorization flaw in an internal API.
Applicant records were referenced using numerical identifiers, and the API did not properly verify that an authenticated user was authorized to access the requested applicant record.
By changing the applicant identifier, the researchers retrieved a real applicant’s personally identifiable information and raw conversation history with the Olivia AI assistant.
The vulnerable identifier space extended across tens of millions of applicant and conversation records, demonstrating the potential scale of the authorization failure.
The researchers disclosed the issue to Paradox on June 30, 2025.
Paradox revoked the legacy administrative credentials and patched the authorization vulnerability within hours.
Paradox later reported that the researchers accessed seven records during validation, including five records containing U.S. candidate information, and that its logs found no evidence of unauthorized access by anyone else.
Affected Systems
McHire / Olivia AI Hiring Assistant / Applicant Records / Candidate PII / AI Conversation Histories / Internal Recruiting APIs
Attack Vector
Weak Administrative Credentials / Broken Object-Level Authorization / Applicant Data Exposure
Business Impact
The vulnerability exposed real applicant information stored inside an AI-driven hiring workflow.
Paradox said the five U.S. candidate records accessed during validation contained names, email addresses, phone numbers, and IP addresses.
Researchers were also able to access raw conversation content between applicants and the Olivia AI assistant.
Because conversational recruiting platforms collect applicant responses, availability, preferences, employment information, and other context during interactive hiring sessions, compromise of these systems can expose significantly more information than a traditional contact database.
The vulnerable applicant identifier space extended beyond 64 million records, indicating a potentially broad exposure path. However, that number represents the reachable identifier space rather than 64 million confirmed compromised individuals.
Paradox reported no evidence that anyone other than the security researchers exploited the vulnerability.
Redbot Analysis
The McHire incident demonstrates that AI application security still depends heavily on conventional identity, authentication, and API authorization controls.
The affected system was an AI conversational hiring platform, but the security failures were familiar: weak credentials and broken object-level authorization.
What makes the incident AI-specific is the sensitivity and context of the affected data.
Conversational AI systems routinely collect and retain information that would historically have been spread across application forms, emails, recruiter notes, scheduling systems, and direct conversations.
When those interactions are centralized inside an AI platform, compromise can expose both traditional personal information and the conversational context accumulated by the AI assistant.
This creates a broader security requirement than simply protecting the underlying model.
Organizations deploying AI-assisted hiring, customer support, healthcare, financial, or other conversational workflows should treat AI conversation histories as sensitive application data and enforce strict authorization at every API and object boundary.
The incident also highlights the importance of distinguishing theoretical scale from confirmed impact.
The vulnerable record space was extremely large, but Paradox’s logs indicate that only the researchers accessed the affected records.
Redbot Labs therefore assesses this as a confirmed production authorization failure and real applicant-data exposure rather than a mass malicious breach.
