Incident Overview
On July 9, 2025, researchers discovered critical security vulnerabilities in McDonald’s McHire AI hiring platform, exposing the personal data of over 64 million job applicants. The platform, developed by Paradox.ai, uses an AI chatbot named Olivia to screen candidates, collect personal details, and administer personality tests. The vulnerabilities allowed unauthorized access to applicant records through a combination of weak default credentials and an insecure direct object reference (IDOR) flaw.
The researchers initiated their investigation after noticing complaints on Reddit about the chatbot providing nonsensical responses. They began by applying for a job through the consumer-facing site, jobs.mchire.com, where they interacted with Olivia, submitted personal information, and completed a personality test hosted by Traitify.com. The test involved selecting responses to statements like 'enjoys overtime,' which appeared to be designed to align with employer preferences.
During their testing, the researchers observed that restaurant owners could access applicant data through a separate login portal at mchire.com/signin. They discovered that the system accepted default credentials—username and password both set to '123456'—granting them administrative access to a test restaurant account. This account was linked to Paradox.ai employees, indicating it had not been properly decommissioned despite being inactive since 2019.
Upon gaining access, the researchers identified an API endpoint, PUT /api/lead/cem-xhr, which allowed them to retrieve candidate data by manipulating the lead_id parameter. By decrementing the lead_id from their test application (approximately 64,185,742), they accessed the personal details of other applicants, including unmasked contact information, chat histories, and auth tokens. This confirmed the presence of an IDOR vulnerability, enabling full enumeration of applicant records across the platform.
The exposed data included candidate states, shift preferences, and raw chat interactions, along with authentication tokens that could be used to impersonate applicants within the consumer UI. The researchers estimated that the vulnerability affected over 64 million records, though Paradox.ai later clarified that only a fraction of accessed records contained personal information. Five of the seven records reviewed by the researchers contained identifiable data, including names, email addresses, phone numbers, and application dates.
The combination of weak authentication and improper API access controls created a pathway for unauthorized data exposure, posing threats such as phishing attacks, identity fraud, and reputational harm to affected applicants. McDonald’s and Paradox.ai confirmed the issues and resolved them within days of disclosure, though the vulnerabilities underscored broader concerns about security practices in AI-powered recruitment systems.
Possible Root Cause
The primary root cause of the data exposure was the acceptance of default credentials—username and password both set to '123456'—for administrative access to the McHire platform. This practice violated basic security principles by allowing unauthorized users to bypass authentication controls.
A secondary and more critical vulnerability was the insecure direct object reference (IDOR) flaw in the API endpoint PUT /api/lead/cem-xhr. This endpoint relied on the lead_id parameter to retrieve candidate data without implementing proper authorization checks. By manipulating this parameter, attackers could access any applicant’s records, effectively bypassing access controls entirely.
The researchers noted that the test restaurant account they accessed had not been logged into since 2019, suggesting it had been overlooked during routine security reviews or system maintenance. This oversight allowed the default credentials to remain active, providing an entry point for exploitation. Additionally, the lack of multifactor authentication (MFA) on the Paradox.ai login page compounded the risk, as it did not require additional verification beyond the weak credentials.
The design of the McHire platform also contributed to the vulnerability. The system’s architecture permitted franchise owners to view applicant data without stringent access controls, and the API’s reliance on predictable lead_id values enabled full data enumeration.
Paradox.ai’s security documentation further indicated a lack of proactive measures, as their public security page suggested that users did not need to worry about security issues. The incident revealed systemic failures in authentication management, API security, and vulnerability monitoring, all of which contributed to the widespread data exposure.
Takeaways
- Weak Authentication Controls: The acceptance of default credentials ('123456') demonstrates a fundamental failure in authentication security. Organizations must enforce strong password policies, disable default accounts, and implement MFA to prevent unauthorized access. Test or staging environments should be regularly audited and decommissioned to eliminate residual risks.
- Insecure API Design: The IDOR vulnerability in the PUT /api/lead/cem-xhr endpoint highlights the dangers of improper access controls in API design. Developers must implement strict authorization checks, input validation, and least-privilege principles to prevent unauthorized data access.
- Third-Party Risk Management: The incident underscores the importance of rigorous vendor security assessments, particularly when third-party providers handle sensitive data.
- Data Minimization and Encryption: The exposure of 64 million applicant records emphasizes the need for data minimization—collecting only essential information—and encryption of sensitive data at rest and in transit.
- Proactive Vulnerability Disclosure: The researchers’ disclosure process was hindered by the lack of publicly available contacts for reporting security issues. Paradox.ai’s initial response—stating that users did not need to worry about security—further complicated responsible disclosure. Organizations should establish clear, accessible channels for reporting vulnerabilities and foster a culture of transparency to encourage ethical security research.
- Phishing and Fraud Risks: The exposed applicant data, including contact information and employment intentions, created opportunities for phishing attacks and fraud. Attackers could impersonate recruiters to request sensitive financial details or manipulate applicants into disclosing additional personal information.
- Proactive security measures and rapid incident response are critical to mitigating both legal and reputational fallout.
- AI System Security Challenges: The use of AI chatbots in hiring introduces unique security challenges, including prompt injection risks and data handling vulnerabilities. Organizations deploying AI-driven tools must conduct regular security assessments, red-team exercises, and dependency reviews to identify and remediate flaws. The McHire incident serves as a case study in how poorly secured AI systems can lead to large-scale data exposures.