Dentists own the risk of AI hiring tools

Ninety-one percent of dentists who were recruiting or had recently recruited a hygienist called it very or extremely challenging (American Dental Association Health Policy Institute, April 2026, citing its Q1 2026 dental economy report). If you have posted a hygienist role and heard nothing for six weeks, an AI tool that reads every résumé overnight and hands you a ranked shortlist sounds like relief. The pitch usually comes with a reassurance: the vendor built it, the vendor runs it, so if it screens someone out unfairly, that is the vendor's problem.

That belief has a kernel of truth, which is why it survives. Vendors are being sued. But being sued and being responsible are not the same thing, and the contract and the statute book both point back at the practice that made the hire.

The part of the myth that is true

In Mobley v. Workday, a federal judge in July 2024 let a discrimination suit proceed against the software vendor on the theory that it acted as an agent of the employers using its screening tools, and in May 2025 the case reached nationwide collective certification covering applicants over 40 who were rejected by the system (Jones Walker, September 2025). Workday was also ordered to produce its customer list in discovery (Smith Gambrell Russell client alert, February 2026). Read that last detail twice: the employers who used the tool are the next names on the page.

A second case follows the same pattern. Kistler v. Eightfold AI, filed in January 2026 in the Northern District of California, alleges that the vendor builds hidden candidate dossiers and scores applicants from zero to five without telling them, and so should be treated as a consumer reporting agency under the Fair Credit Reporting Act (TechRepublic, March 2026; Smith Gambrell Russell, February 2026). Eightfold disputes the claims. Both cases are allegations, not findings, and neither involved a dental practice.

So what for you: vendor lawsuits show that courts will look at the tool, but they also show that every customer of the tool is a potential defendant. Read them as a warning about your exposure, not as proof that someone else carries it.

What the contract says

A Jones Walker analysis (September 2025) reported that 88% of AI vendors impose liability caps on themselves, often limited to monthly subscription fees, and that only 17% give warranties for regulatory compliance. The firm attributes both figures to an unnamed study and gives no sample size, so treat them as directional rather than precise. It adds that vendor contracts often require the customer to defend and hold the vendor harmless, and may restrict the customer's right to audit.

Run an illustrative case with assumptions you should replace. A screening tool costs $300 a month and the contract caps the vendor's liability at 12 months of fees. The cap is $3,600. That number is set by what you pay the vendor, not by what a discrimination claim costs to defend, and an indemnity clause can push the remaining bill the other way, toward you.

So what for you: before you sign, find two clauses, limitation of liability and indemnification, and ask who pays if a rejected applicant files a claim against the practice.

What the law now says, starting with Connecticut

Connecticut Public Act 26-15, known as the CART Act, changes the position from October 1, 2026. In a direct employer discrimination claim involving an automated employment decision tool, the statute says use of the tool is not a defense, although the state's human rights commission and courts may weigh anti-bias testing and similar proactive efforts (Ropes & Gray, June 2026). From October 1, 2027, employers must tell applicants when they are interacting with such a tool and give written notice before the tool factors into a decision, naming its purpose, trade name and the categories and sources of personal data it analyzes. Only the state attorney general can enforce, and a 60-day cure period is available at the attorney general's discretion for violations through December 31, 2027 (Ropes & Gray, June 2026).

Two gaps in what we could verify. The sources we read state no employee-count threshold, and secondary summaries disagree on Connecticut's dates and signing date, so we used the law firm's account. If you employ people in Connecticut, ask counsel whether your practice is covered.

Outside Connecticut, the baseline is older and does not depend on any AI statute. Title VII of the Civil Rights Act covers employers with 15 or more employees (EEOC, Federal Laws Prohibiting Job Discrimination, checked October 2026), and the EEOC removing its AI guidance in January 2025 did not change the underlying law, as we set out in our Q&A on AI hiring for solo clinicians. A state summary from HR for Health (2026, secondary) lists Illinois, effective January 1, 2026, which treats using ZIP codes as a proxy for protected classes as discriminatory and requires consent before AI analyzes recorded video interviews, and New York City, where a bias audit has been required for automated hiring tools since January 1, 2023.

So what for you: if you hire in Connecticut, Illinois or New York City, put the notice and consent duties on your calendar now. Everywhere else, assume federal anti-discrimination law already applies to the tool's output.

Where a dental practice is exposed

Dental hiring has two features that raise the stakes. The applicant pool is thin: only 60% of dentists report an adequate number of hygienists on staff (ADA Health Policy Institute, April 2026), so each rejection removes a scarce candidate. And most practices have no HR department, so nobody writes down why a person was turned down. If a claim arrives, that missing paper is the evidence you do not have. For the wider supply picture, see our data drop on the dental hygienist shortage.

The risk concentrates in one feature: automatic rejection. Ranking is a convenience, since you still read the top of the list. A score that silently removes the bottom half is a decision made without a human, and the usual proxies are the ones tools pick up from historical data: gaps in employment (which can track caregiving or disability), graduation year (which tracks age) and home ZIP code.

Here is the arithmetic, using assumptions of our own. A hygienist posting draws 60 applications. Reading each takes two minutes, so 120 minutes of your time or your office manager's. A tool that auto-rejects the bottom 40 saves about 80 minutes and leaves you with 40 decisions that nobody can explain. Eighty minutes against 40 undocumented rejections is a poor trade for a practice with no HR function.

So what for you: the time saved by automatic rejection is small and the unexplained decisions it creates are not. Keep the saving from drafting and scheduling, and give up the saving from auto-rejection.

What the evidence points toward

Frankly, if you are a principal weighing an AI hiring tool this quarter, the math says use it for the low-risk work and keep the judgment calls human. Three steps.

First, use AI for what carries little exposure: drafting the job posting, answering candidate questions about hours and pay, and scheduling interviews. Second, switch off automatic rejection and any candidate scoring that draws on data outside the application, until the vendor has answered four questions in writing: does the tool reject or score on its own, what data does it use beyond what the applicant submitted, what bias testing can you see, and what do the liability and indemnity clauses say. Third, make a person responsible for every no, and log one line of reason for each. A shared spreadsheet is enough. It takes seconds per candidate and it is the record that Mobley-style discovery would ask for.

If you also use AI tools that touch patient data, remember that a business associate agreement covers a different legal question from employment law (see why "HIPAA compliant" does not mean safe). If you want a ranked view of which admin and hiring workflows are worth automating first, the AI Opportunity and Growth Assessment builds one from your own numbers, or you can book a free 20-minute call.

So what for you: the one thing to do this week is open your vendor contract, or the vendor's terms page, and read the limitation of liability clause before you turn on any screening feature.

Related: The EEOC pulled its AI hiring guidance. The law did not move  ·  The dental hygienist shortage in numbers  ·  Why HIPAA compliant does not mean safe

Sources: ADA Health Policy Institute, HPI Perspectives on the dental hygienist shortage (April 2026); Jones Walker, AI Law and Policy Navigator, "AI vendor liability squeeze" (September 15, 2025), whose 88% and 17% figures it attributes to an unnamed study; Smith Gambrell Russell client alert on AI hiring litigation (February 2026); TechRepublic on Eightfold AI (March 2026); Ropes & Gray alert on Connecticut Public Act 26-15 (June 2026); EEOC, Federal Laws Prohibiting Job Discrimination (checked October 2026); HR for Health state AI hiring law summary (2026, secondary). Secondary summaries disagree on Connecticut's signing date and on some effective dates; we used Ropes & Gray. The $300 a month, 12-month cap and 60-application figures are our own illustrative assumptions. Whether Connecticut's act reaches a given small practice is unconfirmed. Court filings described are allegations, not findings. This article is for informational purposes only and does not constitute legal advice.