A materially lower selection rate for a protected group: Under nondiscrimination law, which conclusion is most

A large adverse selection gap can trigger disparate-impact liability without intent, so the employer must validate job-relatedness and seek less-discriminatory alternatives.

The question

Preparing for AIGP? Take the free 5-min readiness quiz →

1. Removing protected attributes guarantees compliance, so the disparity is legally irrelevant and the employer may keep deploying the tool while it waits for the vendor to improve explainability.

   Plausible because dropping protected fields feels protective, but proxies can reproduce bias and a significant adverse selection gap can still create disparate-impact liability.
2. Liability depends only on proving the vendor intended to discriminate, so absent evidence of intent the employer bears no legal risk regardless of how uneven the selection outcomes turn out to be.

   Almost right on the disparate-treatment theory, but nondiscrimination law also recognizes disparate impact, which does not require proof of discriminatory intent.
3. A materially lower selection rate for a protected group can create disparate-impact exposure even without intent, so the employer should validate the tool for job-relatedness and less-discriminatory alternatives. ✓

   Correct: nondiscrimination law reaches facially neutral practices that produce adverse impact, and the defense turns on demonstrated job-relatedness and consideration of less-discriminatory alternatives.
4. Because a third-party vendor built and scored the model, all responsibility shifts entirely to that vendor, and the employer may rely on the contract's indemnity clause alone to discharge its own statutory duties.

   Plausible but an employer generally remains legally responsible for its hiring decisions, and a private indemnity does not transfer statutory nondiscrimination obligations to the vendor.

How to remember it

A large adverse selection gap can trigger disparate-impact liability without intent, so the employer must validate job-relatedness and seek less-discriminatory alternatives.

More Understanding How Laws, Standards and Frameworks Apply to AI questions

• Copyright can restrict use of the scraped works: Considering intellectual property law and the EU AI Act →
• Unsupported performance claims: Under consumer protection law against unfair or deceptive practices, which →
• Harm caused by design or manufacturing defects: In the context of AI-enabled products, what does product →
• All 394 Understanding How Laws, Standards and Frameworks Apply to AI questions →