"The software rejected them, not us."
Whether you build AI hiring tools or use one to screen candidates, that sentence should worry you. Because a federal court has already signalled it won't save either of you.
What Mobley v. Workday actually signalled
In Mobley v. Workday, the vendor argued it just provides software — that employers set the parameters and make the decisions. The court let the discrimination claims proceed anyway, treating a tool that performs the screening function as the employer's agent.
And the employers using those tools? Already being pulled into the case. Nobody in the chain got to point down the line and walk away.
An automated rejection is a legal event
Here is the implication most teams are missing: an automated rejection is a legal event regardless of who set the criteria — or who built the machine.
If an AI tool auto-rejects a candidate on something that correlates with age, race, or disability, it doesn't matter much whether the vendor coded it or the employer configured it. The company doing the hiring is on the hook. The vendor increasingly is too.
"We just deployed it" and "we just built it" are the same argument, and it's the one that's failing.
If you use AI to hire
You can't outsource the liability to the software. Using a tool doesn't move the responsibility for the outcome off your company.
Ask your vendor one question: when this says no, who actually made that decision?
If the answer is "the algorithm," that's your exposure, not theirs.
If you build AI to hire
Configurability is not a shield.
Letting clients auto-reject on any criteria they choose isn't a feature. It's a channel for discrimination running through your product, with your name on the mechanism.
How we built Eva around this
This is the principle we built aiAvenu around. Our AI interviewer, Eva, doesn't auto-reject.
She evaluates job-related competencies, surfaces the evidence, and a human makes every advance-or-decline call. Anything that could touch a protected characteristic routes to a person, not an automatic screen.
Because the goal was never to help anyone hire faster by handing the decision to a machine. It's to help them hire well — with a human owning the call, and a record that shows it.
Three things to hold onto
- The danger isn't AI in hiring. It's automated rejection with no human in the loop.
- Intent doesn't matter, impact does. A neutral-looking criterion that screens out a protected group is enough.
- "The software did it" is not a defense. Not for the builder, and not for the employer.
The question worth asking
If you're deploying AI in hiring, the question isn't whether the tool is fast or cheap.
It's whether, when it says no, a human owned that decision — and whether you can prove it.
Kaylynn Kim is Founder & CEO of aiAvenu.
