I was one of 15 practitioners quoted in GoHire's 2026 Recruiting AI Adoption Playbook. My contribution was about bias and screening; what happens when you point an algorithm at a hiring history that was never diverse to begin with.
The line they pulled:
"If your AI was trained on your last ten years of hires, and those hires weren't diverse, congratulations - you've automated your bias."
That will garner nods of agreement, but a nod isn't a safeguard. So, here's a 9-point self-check on whether your AI screening is defensible, not just installed.
The Ethical AI Screening Scorecard
9 questions every TA leader should be able to answer before an algorithm touches a candidate.
Most "AI in hiring" advice is about adoption — which tool, how fast, how much it'll save you. Almost none of it is about whether you can defend what you've adopted when a candidate, a regulator, or your CEO asks how the decision got made.
Check every box you can answer truthfully — one point each. The boxes you can't check are your exposure, the gap between the tools you're running and the guardrails you can prove.
[ ] 1. Training data & inputs — You know what your screening tool was trained on, what inputs it scores candidates on, and whether the hiring history it learned from was actually diverse.
[ ] 2. Independent audit — Your tool has been bias-audited in the last 12 months by someone who doesn't work for the vendor.
[ ] 3. Explainability — You can tell a rejected candidate, in plain language, exactly how AI factored into their outcome.
[ ] 4. Human authority — A human makes the final advance-or-reject decision; the tool recommends, ranks, or flags.
[ ] 5. Jurisdiction — You know which AI hiring laws apply based on where your candidates sit and where the process runs— NYC Local Law 144, Colorado, Illinois, EU AI Act — not just where your office is.
[ ] 6. Vendor accountability — Your vendor has documented bias testing and a clear position on responsibility if the tool contributes to discrimination.
[ ] 7. Notice & recourse — Candidates are told AI is part of the process and have a real path to request human review, appeal, or accommodation.
[ ] 8. Ongoing monitoring — You track selection rates and disparate impact across groups after go-live, not just during the procurement phase.
[ ] 9. Governance & escalation —There's a named owner for your AI screening decisions and a defined escalation path for when the tool produces questionable or contested results, and when the laws or the model change.
Your score:
7–9 — Leading. AI screening is governed, monitored, and defensible at your org.
4–6 — Exposed. Your tools are running ahead of your guardrails. Fixable, but not yet defensible.
0–3 — At risk. Your screening runs on the vendor's word, not on evidence you can produce. Right now, you couldn't defend how a single decision was made.
How'd you score?
If you landed in Leading — send this to a peer who wouldn't.
If you landed in Exposed or At risk, it's a starting point. Pressure-testing screening tools against bias, explainability, and compliance is the work I do. Book a call and we'll walk your stack through all nine.
And if this saved you a meeting, you can buy me a coffee so these stay free.
— Keirsten Greggs, Founder & Principal Consultant of TRAP Recruiter, LLC, and host of the TRAP Chat podcast. A talent acquisition leader and strategist with 26 years across defense, tech, and nonprofit sectors, and creator of the TRAP Framework — Trust, Relationship Building, Accountability, Proactive Approach.