Who must comply with NYC Local Law 144?

Any employer or employment agency using an automated employment decision tool (AEDT) to screen candidates for a job located in New York City, regardless of where the employer is headquartered. They must run an independent bias audit within the prior year, publish the summary, and notify candidates at least 10 business days in advance.

When do EU AI Act obligations apply to hiring systems?

AI systems used for recruitment and employee selection are high-risk under Annex III 4(a). High-risk obligations for hiring systems apply from August 2, 2026, covering risk management, data governance, human oversight, transparency, logging, and post-market monitoring.

An Automated Employment Decision Tool is any computational process derived from machine learning, statistics, or AI that produces a score, classification, or recommendation used to substantially assist or replace discretionary hiring or promotion decisions.

What are the penalties for NYC Local Law 144 non-compliance?

Fines start at $500 for a first violation and reach $1,500 for each subsequent violation. Each day of non-compliant use and each missed candidate notice counts as a separate violation.

Can our AI vendor perform our bias audit?

No. NYC Local Law 144 requires an independent auditor with no role in building, selling, or using the tool and no financial interest in it. Vendors can supply data but cannot serve as the independent auditor.

The Human in the Loop Is Mostly Asleep

What automation bias means for AI hiring compliance

The EU AI Act treats human oversight as a safeguard. Put a person between the AI system and the decision, and the risk goes down. It is a reasonable-sounding requirement, and it is written into law for high-risk hiring systems.

Behavioral science says that safeguard is weaker than it looks.

The reason has a name: automation bias. And once you understand it, "human in the loop" stops being a guarantee and starts being a design problem you have to actually solve.

What automation bias is

Automation bias is the well-documented tendency to over-trust an automated recommendation and under-use our own judgment when one is present. It shows up in two directions.

The first is errors of commission: we follow the machine when the machine is wrong. The recommendation is on the screen, it looks authoritative, and we go along with it even when the evidence in front of us should have given us pause.

The second is errors of omission: we miss the things the machine never flagged. If the system doesn't surface a problem, we don't go looking for one. Our attention narrows to what the tool chose to show us.

This isn't a quirk of careless people. It has been studied for decades in aviation, where pilots defer to flight automation, and in medicine, where clinicians defer to computerized decision support. In both fields, highly trained professionals who are paying close attention still drift toward ratifying the system rather than scrutinizing it. The bias operates below the level where good intentions can catch it.

How it shows up in hiring

Now move it into a hiring room.

A recruiter opens an AI-ranked shortlist. The candidate ranked second gets a call. The candidate ranked fortieth does not. The recruiter feels like they are exercising judgment. They are reviewing, comparing, deciding.

But look at what actually happened. The ranking set the frame. The recruiter's attention went to the top of the list, because that is where the tool pointed it. The candidates near the bottom were never really evaluated, they were filtered out by a number nobody interrogated. The felt experience is judgment. The functional reality is agreement.

The human is in the loop. They are just not resisting the model. They are rubber-stamping it.

The compliance trap

This is the gap most compliance work misses. The law assumes the person in the loop will catch the system's errors. Automation bias says that person can be fully engaged, fully well-intentioned, and still functionally supervising nothing.

Which produces the worst outcome of all, the one that hides in plain sight: a process that is technically compliant and functionally unsupervised. You have the human. You have the documentation. And you have almost no real oversight, because the oversight mechanism was undermined by a bias nobody designed around.

You can pass an audit this way. You cannot actually govern a hiring decision this way.

The fix is design, not disclaimers

Adding another notice or another sign-off box does not help, because the problem is not that people lack information. The problem is where the information sits in the decision, and how easy it is to defer. That is a design question, and it has design answers.

Have the human form a judgment before they see the score.
Anchoring is the whole mechanism here. If the reviewer commits to a view first and sees the model second, the model becomes a check on them rather than the other way around.

Make an override cost a sentence, in both directions.
Require a written reason to depart from the model, and require one to follow it on a close call too. A justification you have to type is a justification you have to actually have.

Show uncertainty, not just a clean rank.
A single confident number invites deference. A range, a confidence level, or the factors driving the score all give the reviewer something to push against.

Watch your override rate.
If nobody on your team ever disagrees with the model, that is not evidence the model is excellent. It is evidence the humans have stopped functioning as reviewers. A zero-percent disagreement rate is a red flag, not a win.

The real question

Compliance on paper is easy. You can satisfy the letter of the EU AI Act, or NYC Local Law 144, with a human whose role is purely ceremonial. The regulators asked for oversight. Nothing stops you from delivering the appearance of it.

But the point of the law is not the checkbox. The point is that a person, a real one, catches what the machine gets wrong before it costs a candidate a job they should have had.

So the question worth asking about your own hiring process is not whether you have a human in the loop.

It is whether that human is still allowed to think.

PeopleNotResumes helps teams meet NYC Local Law 144 and the EU AI Act, and treats the work as a chance to strengthen how hiring decisions actually get made. If you use AI anywhere in hiring, we can help you build oversight that functions, not just oversight that passes. Book a Compliance Score.