AI is driving useful conversations about how we hire. Where can technology make the process more effective? How can it help us recognize candidate potential? And most importantly, what should we expect from the systems supporting those decisions?
It’s a lot to sort through if you’re buying hiring technology. You’re weighing product capabilities and promises while trying to understand how a tool will fit your hiring process and what responsibilities come with using it. Finding clear, practical answers can take more work than it should.
The evolving rules around AI add another layer to those decisions. They also offer a useful starting point for thinking about transparency, fairness, and the role of human judgment in hiring.
In this post, I’ll first cover the science behind the information we use to evaluate candidates and then look at the rules taking shape and how employers and technology providers share responsibility. The goal is to make these considerations easier to understand and help us move forward with greater confidence.
Better hiring AI starts with science-based inputs
Before evaluating what an AI tool can do, it’s worth looking at the information it uses. By science-based inputs, I mean evidence gathered through measures designed to assess clearly defined, job-relevant characteristics, with research supporting how the results should be interpreted. The Society for Industrial and Organizational Psychology (SIOP) recommends evaluating AI assessments for job relevance, consistent measurement, and evidence connecting scores to job performance or other relevant outcomes.
For buyers, that means asking what candidates are being given the opportunity to demonstrate. Consider a work sample built around an actual task, a structured interview with defined scoring criteria, or a validated assessment of a relevant ability. In each case, ask what the measure tells us about the role and what evidence supports using it that way. We should also ask whether that evidence still holds when AI interprets or combines the inputs.
This gives us two distinct areas to evaluate with a provider: (1) bias audits and (2) validation. A bias audit examines differences in outcomes across groups, while validation asks whether the interpretation and intended use of scores are supported by evidence. Both deserve attention when deciding whether a tool belongs in the hiring process.
A few examples help make the rules more understandable
For a buyer, a useful starting point is to ask what each requirement is trying to accomplish. Consider a few examples:
- Making outcomes visible. New York City’s Local Law 144 requires a recent independent bias audit, publicly available audit information, and notices for covered uses of automated employment decision tools. The audit measures differences in selection or scoring rates across demographic groups; it does not certify that a tool is free of bias.
- Explaining the technology’s role. Colorado’s enacted SB26-189 provides for notices when covered automated decision-making technology is used and plain-language descriptions of its role following adverse outcomes. For buyers, this raises a practical question: can we explain to a candidate how the technology contributed to a decision?
- Addressing discrimination. Illinois HB 3773, effective January 2026, prohibits AI use that results in unlawful employment discrimination and bars using zip codes as proxies for protected classes. That makes the information a system uses, and how it uses it, an important part of evaluation.
- Giving candidates visibility and choice. Illinois also requires notice, an explanation, and consent before employers use AI to analyze video interviews. It’s a useful prompt to think about what candidates should understand before they participate.
These examples are a starting point rather than a complete checklist. When evaluating a tool, we recommend working through the applicable requirements with counsel and connecting them to the actual hiring workflow.
Employers and providers each have a part to play
The distinction between a developer and a deployer can help clarify the conversation. In a typical purchasing relationship, the provider develops the technology and the employer deploys it within a hiring process.
Colorado makes this division explicit: beginning January 2027, developers must provide documentation covering intended uses, limitations, and other technical details, while deployers have separate duties concerning notices and consumer rights. That’s a helpful way to approach the working relationship, too.
“Are you compliant?” is an understandable question to ask a provider. It becomes more useful when we follow it with specifics: What has been evaluated? Does that evaluation reflect how we plan to use the tool? What documentation will we receive, and what happens when the system changes?
Those conversations should help employers make an informed decision without expecting every HR team to become an AI research group. Providers can make the process considerably easier by explaining their evidence in terms the people using the technology can understand.
Turn the requirements into useful hiring practices
There’s an opportunity here to connect technology evaluation with the hiring process we want to build. A few practices are worth making part of that work:
- Start with the job. Be clear about what the tool is intended to measure and why that matters for the role. Ask for evidence supporting that connection, including the limits of what a score or recommendation can tell us.
- Keep humans at the center. Decide who reviews the output and how they can question or override it. A reviewer needs enough context and authority to exercise judgment. These inputs are part of a broader human-led hiring process. AI should not make a complete hiring decision.
- Make the candidate experience understandable. Explain where AI is involved and provide a clear route for questions or accommodation requests. Review that communication from the applicant’s perspective.
- Keep learning after implementation. Agree on how outcomes will be monitored and when the setup should be revisited. Treat a change in the model or hiring workflow as a reason to check that the supporting evidence still fits.
Standards can also help organize this work. ISO/IEC 42001, for example, sets requirements for an AI management system rather than detailed requirements for a specific AI application. When considering a certification, look to understand the scope and separately ask how the particular hiring tool has been evaluated.
Questions to ask when evaluating AI for Hiring
We don’t need to become AI experts to ask useful questions. This short checklist can help organize the conversation with a technology provider:
- Job relevance: What does the tool measure, and why does that matter for the roles we’re hiring for?
- Scientific evidence: What research supports its use? Has the AI-generated score or recommendation been validated, as well as the underlying assessment?
- Data inputs and training: What information does the AI use, and how was that information selected and evaluated? How does our candidate information stay removed from training workflows?
- Fairness: What adverse impact testing or independent audits have been completed? Which products and candidate populations were included?
- Human judgment: Who reviews the results, and can they understand, question, or override a recommendation?
- Candidate experience: How will candidates know AI is involved, and how can they ask questions or request accommodation?
- Responsibilities: What documentation and support will the provider supply, and what will our team need to handle?
- Ongoing oversight: How is performance monitored, and what happens when the model or our use of it changes?
These questions make “explainable AI” a practical expectation: can we understand what shaped a recommendation, what evidence supports it, and how it should inform a hiring decision?
Criteria's Approach to AI in Hiring
At Criteria, we believe explainable AI starts with helping employers understand what a tool measures, how it reaches a recommendation, and where human judgment belongs. Our platform starts with job-relevant assessment science, and we apply those same standards to our AI products, evaluating their reliability, validity, and fairness before release (our scientific principles).
We take care with the data our AI uses and test whether the resulting insights support their intended purpose, with research and monitoring continuing after launch (our development approach). Our tools are designed to inform human judgment, with employers retaining control over when and how AI contributes to their hiring process (human oversight.
We also explain where AI is used and make independent audit summaries and our technical manuals available so employers can examine the evidence for themselves (transparency and evaluation). Behind that work, our ISO/IEC 42001-certified AI management system establishes documented processes for governance and risk assessment, including consideration of fairness and data privacy (our AI governance). The goal is to give hiring teams useful technology with a clear scientific foundation, supported by information they can understand and put into practice.
