The hiring software rejected them for being older. Nobody noticed until someone lied about their birthday.

iTutorGroup recruited tutors to teach English to students in China, hiring through an online application. The EEOC alleged the company had programmed that application software to automatically reject women aged 55 or older and men aged 60 or older [1][2].

More than 200 qualified applicants were rejected this way. No human read those applications; there was no decision to appeal, and nothing in the rejection revealed why it had happened [1][2].

It surfaced almost by accident. An applicant submitted two applications that were identical except for the date of birth — and was offered an interview only on the one with the more recent date [3].

The EEOC sued in 2022 and settled in 2023. iTutorGroup agreed to pay $365,000 to the rejected applicants without admitting wrongdoing, and to a consent decree including anti-discrimination policies, recruiter training, an injunction against requesting birth dates, and years of EEOC monitoring [1][3].

The EEOC's own framing is the line worth keeping: "Age discrimination is unjust and unlawful. Even when technology automates the discrimination, the employer is still responsible" [2].

Screening automation exists for a real reason: a posting can draw thousands of applications, and a human reading each one is not neutral either — tired reviewers are inconsistent, and inconsistency is its own kind of unfairness. A filter applied identically to everyone can be fairer than a hurried person, and it can be audited in a way a person's afternoon cannot. That is the genuine promise here, and it is not what happened. The distinction is not automated versus human. It is whether anyone can see what the automation did [1][3].

This case is a useful corrective to the assumption that these failures are exotic. The public record describes software that rejected people over an age — closer to a line of business logic than to anything a model learned. It was not too complicated to understand; it was simply never examined. That is the common case. Most automated decisions that harm people are not inscrutable neural networks: they are ordinary rules, written once, applied to everyone, and never looked at again [1][2].