Could You Explain a Rejection From Three Years Ago?

California's rules on AI in hiring turned one this week. Since October 1, 2025, the state's regulations on automated-decision systems have spelled out that a screening tool which discriminates can break the Fair Employment and Housing Act just as a person can, and they stretched the time employers must keep hiring records, including the data those systems produce, to four years [1]. Most of the commentary last year was about bias testing. The part I'd worry about is the four years.

Here's why. Take a candidate your team rejected for a warehouse role in the spring. Say they file a complaint in 2028 and someone asks a simple question: why were they turned down? At most companies I've seen, nobody could answer that honestly. There's a status in the ATS, maybe a knockout question they got wrong, maybe a note that says "not a fit." The recruiter who made the call left a year ago.

What the rules actually reach

The regulations define an automated-decision system broadly: a computational process that makes a decision about employment or helps a person make one. Resume scoring and AI interviews are the obvious examples. It also plausibly covers plainer things most teams don't think of as AI, like the knockout questions in an application form that auto-reject anyone who says no to weekend shifts. We've written about how rarely those knockout questions get audited. Under California's rules they're exactly the sort of step you might be asked to explain.

Two other details matter. A vendor acting on an employer's behalf can be treated as its agent, so "the tool did it" doesn't get you far. And whether you tested a system for bias, what you found and what you did about it can be weighed as evidence [1]. I'm not a lawyer, and you should talk to one about your own setup, but the direction is clear enough. When a question comes, it will be "show us," and the answer will need to exist in writing.

The phone screen is the hard part to defend

The odd thing is that the AI tools in a hiring stack are often the easiest part to document, because software logs things by default. The steps that are hard to defend are the human ones sitting between the automated ones.

A recruiter phones forty applicants on a Thursday, has a seven-minute chat with each and moves twelve forward. What did they ask? Did everyone get the same questions? Why did candidate 23 get a no when candidate 31, who said much the same thing, got a yes? The notes say "good communicator" for one and "not sure about availability" for the other.

That is the record you'd be defending in 2029. It was written for the recruiter's own memory, not for anyone outside the team, and certainly not for a lawyer. Two recruiters screening differently isn't evidence of bias on its own. But with notes that thin you also can't show it wasn't. We've asked before whether two recruiters would screen the same candidate the same way. The legal version of that question is whether you can prove they did.

What a record you can stand behind looks like

It doesn't have to be elaborate. You need four things you could hand to someone years later: the questions, written down and tied to the job; the candidate's actual answers, not the recruiter's impression of them; the rule that turned answers into a decision; and who or what applied that rule, and when.

Most teams can get most of the way there without buying anything. Freeze a question set for each role and keep the old version when you change it. Have recruiters record answers in fields instead of paragraphs. Write the pass rule down, even if it's only "can work Saturdays and lift 50 pounds." Then once a quarter, look at pass rates by stage, broken out by group wherever you have self-reported data, and look for any step where one group falls away much faster than the rest. If there's a problem in there, you want to be the one who found it, with a note saying what you changed.

How Asendia AI handles this

Asendia is a voice-first AI recruiter. It phones every applicant, usually within hours of them applying and at any hour, 24/7, and runs the same structured screen for the role every time, asking again when an answer is vague. Every conversation is transcribed, so if someone asks in three years what candidate 23 was asked and what they said, it's there word for word, along with the version of the questions in use that week.

The answers go into the ATS you already use as separate fields on the candidate's record, with a short summary next to them, so they sit with the rest of the file under the retention policy you already run. Your team sets the questions and the pass criteria. Most of the agencies using Asendia came to us for volume, to get through hundreds of applicants a week without hiring more recruiters. The ones with California clients tend to find the paper trail is worth almost as much.

Final Word

California didn't invent a duty to hire fairly. That was always there. What changed is the expectation that you can show your work for any candidate, four years later, including the steps you never thought of as automated.

A test for this week: pick one candidate your team rejected about eighteen months ago and try to work out why, using only what's in your systems. Give yourself half an hour. If you can't do it, fix the record before you start worrying about the AI.

Ready to transform your hiring strategy? Schedule a Demo with our founders today!: https://asendia.ai/talk-to-founders