Amazon's AI Recruitment Tool: A Case Study in Defensibility Failure

Amazon spent three years building an AI tool to score every job application 1 to 5 stars.

In early 2017, they scrapped it entirely.

Not because it was slow. Not because it was expensive. Because they could not explain why it gave any individual applicant their score.

Without an explanation, there was no audit trail. Without an audit trail, there was no defensible record. Without a defensible record, the tool could not be used.

That was 2017. Amazon had the resources to walk away quietly.

The regulatory environment in 2025 is different. The EU AI Act classifies AI hiring tools as high-risk systems. The EEOC's first AI hiring discrimination settlement was $365,000. A single applicant discovered the mechanism by submitting two identical applications with different birth dates.

The question for every TA team running automated screening today is not whether the tool is fast or accurate.

It is whether the tool can explain each decision it produces.

If it cannot, neither can you.