Can I make my AI vendor responsible for compliance in my contract?

You can ask for it, but it does not protect you the way you think. These laws put the duty on the employer or the agency that uses the tool, not the company that built it. A vendor indemnification clause might help you recover money later. It does not stop a candidate complaint or a state investigation from landing on your firm first. Get the vendor to hand over their bias-audit results and technical documentation. Then keep your own records too.

Do these laws apply if my AI only screens resumes and does not make the final decision?

In most cases, yes. NYC Local Law 144 covers any tool that substantially assists or replaces a human decision. If your AI ranks, filters, or scores candidates before a recruiter ever sees them, it is shaping the outcome. That counts. The fact that a human signs off at the end does not exempt you. Regulators look at whether the tool influenced who got advanced, and a screening filter clearly does.

How much does a bias audit actually cost and how often do I need one?

For a single tool, a third-party bias audit usually runs between 5,000 and 25,000 dollars depending on complexity and how much clean data you can provide. NYC requires an annual audit. Plan on refreshing it every 12 months and any time you change the tool or the way you use it. If you run several screening tools, budget for each one. The cheaper move is to standardize on fewer tools so you audit fewer systems.

AI Hiring Bias Audits: 4 State Rules Staffing Firms Face

Four state AI-hiring laws are now in force at the same time. Your software vendor will not carry the liability when a candidate files a complaint. Here is what you need to document this quarter.

Four states now regulate AI in hiring at the same time, and all four rules touch staffing firms directly. NYC, California, Illinois, and Texas each have active requirements in 2026. If you use an AI tool to screen, rank, or score candidates, you are on the hook in every state where those candidates apply.

Most staffing leaders I talk to assume the software vendor carries this risk. That is the expensive assumption. These laws put the duty on the party that uses the tool to make employment decisions. That is you. A vendor can hand you a bias audit and a nice slide deck. When a candidate files a complaint, the complaint names your firm.

Here is what each regime actually asks for, and what you need to have on file before someone asks.

The Four Regimes and What They Demand

These are not four versions of the same law. They overlap in spirit and differ in the details. You need to know which one applies to which candidate pool.

NYC Local Law 144

This is the one with teeth and the one people know. If you use an automated employment decision tool on a candidate for a job in New York City, you need three things. A bias audit completed by an independent auditor within the last year. A public summary of that audit posted on your website. And notice to candidates at least 10 business days before you use the tool on them.

The penalty is per violation, and each day counts as a separate violation. Run an unaudited tool for a month and the math gets ugly fast.

California

California's regulations under the Civil Rights Department took effect in 2025 and are fully active in 2026. They treat automated decision systems as covered under existing anti-discrimination law. The key move here is record retention. You have to keep records related to automated decision systems for four years. That includes the data the tool used and the outputs it produced.

California does not hand you a tidy checklist like NYC. It holds you to the same discrimination standards as a human decision, and it expects you to prove the tool did not create a disparate impact.

Illinois

Illinois has two things in play. The AI Video Interview Act has been active for years and covers any AI that analyzes video interviews. You need candidate consent, an explanation of how the AI works, and limits on who sees the video. On top of that, an amendment to the Illinois Human Rights Act now bars using AI in a way that produces discrimination, and it bars using zip code as a proxy for a protected class. That second part catches more tools than firms realize.

Texas

The Texas Responsible AI Governance Act is the newest of the four and takes effect January 1, 2026. It focuses on intent and prohibits developing or deploying AI with the intent to unlawfully discriminate. It also sets up a state oversight structure and a way to cure violations. Texas is less about paperwork and more about being able to show you did not build or use a tool to discriminate. Your documentation is your defense.

Why You Cannot Push This to Your Vendor

Read your vendor contract. It almost certainly says the vendor is not responsible for how you use the product. That is standard. It means the compliance duty sits with you.

There is a practical reason too. The laws regulate the decision, not the code. Your vendor does not know which candidates you screened, in which states, for which clients, with which settings. You do. The audit trail lives in your systems and your process. Only you can produce it.

So stop treating this as a software problem. It is an operations problem. It belongs to whoever owns your hiring workflow, not whoever manages your tech stack.

The Documentation You Need on File

You do not need a legal department to get compliant. You need a folder and some discipline. Here is the short list every staffing firm using AI screening should be able to pull up today.

A current bias audit for every AI tool that screens, ranks, or scores candidates. Dated within the last 12 months. Done by an independent party, not your vendor's marketing team.
An inventory of every AI tool in your hiring flow. Most firms cannot name all of them. Your ATS may have scoring baked in. Your sourcing tool may rank candidates. List each one and what it does.
A candidate notice process. Written notice that AI is used, sent before you use it, with a record of when it went out. NYC wants 10 business days. Build for the strictest rule and you cover the rest.
Records tied to each automated decision. Inputs, outputs, and the settings in place. California wants four years. Keep everything that long.
A proxy check. Confirm your tools are not using zip code, name, or school as a stand-in for a protected class. Illinois names this directly and the others punish the outcome.

The firms that get burned are not the ones using bad tools. They are the ones who cannot show what their tools did. When a regulator or a plaintiff's attorney asks, silence looks like guilt.

Your Move This Week

Pull one thing together this week: a written inventory of every AI or automated tool in your hiring process, and the states where you use each one. Include your ATS scoring, your sourcing tools, any video interview software, and any resume ranker. Next to each one, write whether you have a current bias audit for it. Yes or no.

That single page tells you your exposure. The tools with a "no" next to them are your open risk. Fix those first. If you use tools in NYC without a posted audit, that is the one to close this month, because the penalties there stack by the day.

Your AI Is Deployed. Now Prove It's Legal: A Staffing Firm's Guide to the 4 Active Bias-Audit Regimes in 2026