The AI Did It” Is Not a Defense

[When AI hiring tools discriminate, employers bear the liability—here’s how to audit and protect your process.]

Tags: Compliance

---

Who’s really on the hook?

Many talent teams assume that if an AI screening tool makes a biased call, the vendor absorbs the fallout. The reality, reinforced by a recent federal ruling, is different: **the organization that deploys the tool is responsible for its outcomes**. If the algorithm disproportionately screens out older workers, candidates of a certain race, or people with disabilities, the employer—not the software builder—faces legal exposure.

The Mobley ruling and its ripple effect

Derek Mobley applied to over 100 positions processed through Workday’s AI‑driven screen and received rapid, repetitive rejections. He alleged the system discriminated based on his age, race, and disability. In March 2026, the court rejected Workday’s argument that age‑discrimination protections apply only to current employees, confirming that job applicants are equally covered.

More striking, the AI had processed **1.1 billion applications** across **11,000+ companies**. Every employer that relied on the tool without validating its fairness now sits in the potential line of fire. The decision makes clear that “the vendor did it” is no longer a shield.

A 15‑minute audit you can run today

You don’t need a data‑science team to spot red flags. Follow these steps:

1. **Know the trigger** – If any protected group is rejected at a markedly higher rate than others, liability rests with you.

2. **Pull the data** – Export all applicants your AI rejected in the past year, alongside those you hired.

3. **Run a quick pattern check** – Load the two lists into a spreadsheet and ask an AI assistant (Claude, ChatGPT, etc.) to compare rejection rates by age, race, gender, or disability flag. A consistent over‑representation of one group at 80 % or higher is a warning sign.

4. **Question the model** – Ask your vendor:

- What inputs drive the score or rejection decision?

- Was the model trained on historical hiring data from your organization?

- Does it rely on proxies for age, such as graduation year or years of experience?

If the answers reveal biased inputs or opaque logic, it’s time to adjust thresholds, retrain on more inclusive data, or add a human‑review step before final decisions.

Building guardrails, not just hoping for the best

Compliance isn’t a one‑off checklist; it’s an ongoing feedback loop. Establish a routine audit cadence (quarterly works for most teams), document your findings, and treat any disparity as a trigger for remediation. Pair this process with transparent candidate communication—let applicants know how decisions are made and offer a path to contest outcomes.

By embedding these checks into your workflow, you turn a potential legal liability into a competitive advantage: fairer hiring, richer talent pools, and confidence that your AI‑powered recruitment stands up to scrutiny.

---

**