Every business needs efficient and reliable shortcuts.
“Can we make hiring faster?”
“Can we screen better?”
“Can we stop reading 312 resumes from people who swear they’re ‘detail-oriented’ (while misspelling detail)?”
So we hand the keys to AI and let it sort through the pile. But outsourcing the load — even to a machine — doesn’t outsource the liability.
Now California has codified that point: if you use AI in hiring, you own the outcome.
You may be thinking “OK…but I’m in New Jersey, 3,000 miles away (and this week, 60 degrees colder) from The Golden State.”
Yes, but Sunny Cal calls a legal tune. And Jersey is humming along — witness the new rule, adopted in December, confirming that NJ will apply a “disparate impact” analysis to job discrimination claims, and calling out “automated decision-making tools” as a potential source of such discrimination. (In other words, if the stats look bad, it doesn’t matter if you had no intent to discriminate). The smart, baseline lessons are helpful anywhere you’re using hiring tech.
“The rules adopted today also explain[ ] how disparate impact liability applies to the use of artificial intelligence (AI) and other automated decision-making tools in the employment context. The rules adopted today mark the latest action by DCR as part of its Civil Rights and Technology Initiative, which seeks to address the risks of discrimination stemming from the use of AI and other advanced technologies. The rules explain how employers’ use of automated online application tools or facial analysis software in hiring employees may have a disparate impact based on protected characteristics under the LAD.”
Affect on Neurodiverse Candidates
Because neurodivergent candidates are often the first to get filtered out by “efficient” systems for reasons that have nothing to do with doing the job well:
- Communication styles that don’t match the model’s idea of “polished”
- Employment gaps that reflect life, not lack of talent
- Timed assessments that measure speed, not skill
- “Culture fit” proxies that quietly punish differences
- The built-in biases of historic datasets
I’m telling you, there’s a tidal wave of these private lawsuits coming, and they don’t need Washington’s blessing — so start moving to the higher ground now. If you’re using AI in recruiting (or anything that uses an algorithm to make these big decisions), here are some practical to-dos that are just good management.
The 8 To-Dos
- Inventory your tools to assess your risks. Resume filters, assessments, video interview scoring, ad targeting, ranking tools. If it touches candidate flow, examine it.
- Ask what it’s optimizing for. “Past success” can mean “past sameness.” That’s how bias gets automated without a second thought.
- Run bias checks and keep receipts. You’ll need to know whether outcomes are skewed against a protected class and what you’re going to do about it. (And remember that protected classes are often defined more broadly by state laws, like the NJLAD, than federal.)
- Put a real human in the loop. For real. (like a trained reviewer who is monitoring and can actually intervene, not another AI)
- Have accommodations in place. If a timed test or format disadvantages certain applicants, you need a way to adjust the process without drama. My take: always offer an alternative recruitment path, and don’t diss the ones who use it.
- Save audit-ready records. Inputs, outputs, scoring/ranking, override logs, vendor materials. And monitor that data. If you ever have to explain your process, you’ll want more than just vibes.
- Tighten vendor contracts. Again, farming out the work, even to a big tech-based vendor, does not relieve you of the potential liability. Do your own diligence, demand transparency, audit rights, and stay vigilant.
- Train your team. The best policy in the world fails if you and your recruiters don’t know when to override, accommodate, or escalate.
Don’t let your efficiency tool become a liability machine or a talent-repellent. If you’re using AI anywhere in hiring and want to sanity-check your workflow (and your risk), we’re happy to talk.
