Key takeaways
- AI is most useful today for drafting, summarizing, answering routine questions, and scheduling.
- Using AI to screen, rank, or evaluate people carries the highest legal and fairness risk.
- Anti-discrimination law applies to AI-assisted decisions, and several states and cities have added AI-specific rules.
- Keep a person accountable for every employment decision AI influences.
- Start governance with an inventory of every AI tool already in use, including features inside existing software.
AI in HR refers to the use of machine learning and generative AI tools in people processes, such as writing job descriptions, screening applicants, answering employee questions, recommending learning, and analyzing workforce data. Because these tools can influence decisions about people, they carry legal and ethical obligations.
A recruiting team turns on an AI feature in its applicant tracking system that ranks candidates by predicted fit. Time to shortlist drops dramatically, and everyone is pleased. Six months later, someone notices that candidates from certain colleges, and candidates with employment gaps, almost never reach the top of the list. Nobody knows exactly why, the vendor can't fully explain it, and some of the company's job postings were open to candidates in New York City.
The team didn't set out to create a problem. They turned on a feature. That's how most AI risk enters HR: quietly, through tools that were bought for something else. This guide covers where AI helps, where it creates risk, and how to govern it.
Where AI genuinely helps HR teams
The most reliable uses today save time on work that a person reviews before anything happens.
- Drafting and editing. Job descriptions, interview questions, policy drafts, offer letters, and internal announcements. A person edits and approves every word.
- Answering routine employee questions from your own handbook and benefits documents, with links to the source and an easy way to reach a person.
- Scheduling interviews and meetings across many calendars.
- Summarizing interview notes, survey comments, and long documents, with the original available for review.
- Learning recommendations based on role and skills.
- Analytics, such as asking workforce questions in plain language and getting charts back, with the underlying data visible.
Notice the pattern. In each case AI produces a draft or suggestion, and a person stays accountable for the result.
Where AI carries the most risk
Risk rises sharply when AI screens, ranks, scores, or evaluates people. That includes resume screening, candidate ranking, video interview analysis, personality or skills assessments scored by AI, productivity monitoring, and tools that predict attrition or performance for individual employees.
These tools can reproduce patterns in historical data, including past bias, and they can be hard to explain. A model trained on the profiles of past successful hires may learn to prefer people who resemble them, for reasons that have nothing to do with job performance.
Federal law still applies to AI decisions
There is no single federal AI employment law, but existing laws apply fully. Title VII prohibits employment practices that have a disparate impact on protected groups unless they are job-related and consistent with business necessity, and that includes selection tools that use AI. The Americans with Disabilities Act requires reasonable accommodations, which can mean offering an alternative to an AI assessment that disadvantages someone with a disability. The Fair Credit Reporting Act can apply to certain third-party reports about candidates. And employers, not vendors, generally remain responsible for the outcomes of tools they choose to use.
State and local AI rules
States and cities have moved faster than Congress. Some notable examples:
- New York City requires employers using automated employment decision tools for hiring or promotion to obtain an independent bias audit, publish a summary, and notify candidates.
- Illinois requires notice and consent before using AI to analyze video interviews, and amended its Human Rights Act, effective January 1, 2026, to prohibit discriminatory use of AI in employment decisions and require notice.
- California civil rights regulations on automated decision systems in employment took effect in October 2025, clarifying that existing anti-discrimination rules apply to these tools and setting record-keeping expectations.
- Colorado passed a broad law on high-risk AI systems, including those used in employment decisions, whose effective date has been pushed back. Check its current status.
Outside the US, the EU AI Act treats many employment uses of AI as high-risk, which matters for multinationals. This area changes quickly. Review it at least twice a year.
Questions to ask every AI vendor
- What exactly does the AI do, and what data does it use?
- Can we turn the AI features off, individually?
- Has the tool been independently audited for bias, and can we see the results?
- How do you support compliance with laws like New York City's bias audit requirement?
- Is our data used to train models shared with other customers?
- Can the tool explain why it produced a particular result?
- How do candidates or employees request an alternative process or accommodation?
- What records do you keep, and for how long?
Write the answers into your contract where possible, especially on data use and audit cooperation.
A governance framework you can start this quarter
Governance doesn't have to be heavy to be useful. A practical starting point:
- Inventory. List every AI tool and feature in use across HR, including AI built into software you already own. Most companies find more than they expected.
- Classify risk. Low risk for drafting and scheduling. Higher risk for anything that screens, ranks, or evaluates people.
- Set rules. Write a short policy: approved tools, what data can and can't be entered, human review requirements, and who approves new tools.
- Human accountability. For every higher-risk use, name the person accountable for decisions and make sure they can override the tool.
- Test and monitor. Check outcomes for higher-risk tools by group, such as pass-through rates by gender and race where data is available, and repeat regularly.
- Notice and alternatives. Tell candidates and employees when AI is used in decisions about them, as required and as good practice, and offer an alternative process where appropriate.
- Review. Revisit the inventory and policy at least twice a year.
Employee data and privacy
Generative AI tools create a new data risk: employees pasting sensitive information into public tools. Make your rules explicit. Never enter Social Security numbers, health or leave information, performance notes, compensation, or investigation details into tools that aren't approved and contracted for that purpose. Use enterprise versions of AI tools with contractual data protections. Several states also have privacy laws that give employees rights over their personal information, and California's privacy law covers employee data.
Talking with employees about AI
Employees are often more worried about AI than leaders realize. They wonder whether it is evaluating them, whether their jobs will change, and whether their data is safe. Silence makes the worry worse. Explain clearly which AI tools the company uses, what they are for, what data they use, and where people remain in charge of decisions. Invite questions, and give employees a way to raise concerns. Being open about AI builds the trust you need to adopt it well.
Measuring whether AI is actually helping
Many AI rollouts are judged by enthusiasm rather than results. Pick a few measures before you start, such as time saved on job description drafting, time to schedule interviews, the share of employee questions resolved without a ticket, or recruiter hours per hire. Measure them for a baseline period, then again after rollout. Also track quality: error rates in AI-drafted content, employee satisfaction with AI answers, and any fairness measures for higher-risk tools. If a tool doesn't move the measures you chose, it is fine to switch it off.
Preparing HR teams and managers
The HR teams getting the most from AI treat it as a skill, not just a tool. Train HR staff and managers on how to write good prompts, how to check AI output for errors and bias, and when not to use AI at all, for example in sensitive employee relations conversations. Share examples of good use internally. And be honest with employees about how the company is using AI, including how it may change roles over time, which our workforce planning guide discusses.
Our ATS guide and AI transparency label explain how we evaluate AI features in HR software.
AI in HR governance checklist
Work through this list with your team. The printable PDF version has space to tick each item off.
- Inventory every AI tool and feature used in HR, including built-in features
- Classify each use as lower risk or higher risk
- Write a short AI use policy for HR and managers
- List data that must never be entered into unapproved tools
- Name an accountable person for each higher-risk use
- Confirm people can override AI recommendations
- Ask vendors for bias audit results and explanations
- Confirm whether vendors train shared models on your data
- Check New York City, Illinois, California, and Colorado rules where you hire
- Give candidates and employees notice where required
- Offer alternative processes and accommodations
- Monitor outcomes by group for higher-risk tools
- Train HR staff and managers on responsible use
- Review the inventory and policy twice a year
Where to go from here
Wherever you are in the process, there is a next step that fits.
Compare software categories
Email templates for this topic
Ready-to-edit emails with subject lines, compliance notes, and a free PDF.
- Skills assessment request email templateAsk a candidate to complete a skills test or work sample, with a clear scope, deadline, and purpose.
- Candidate sourcing outreach email templateA short, specific first message to a passive candidate that gets replies instead of deletes.
- Application received email templateConfirm an application, set expectations on timing, and keep the candidate warm while you review.
- Interview invitation email templateInvite a candidate to interview with every detail they need: format, time, people, and how to prepare.
Frequently asked questions
Is it legal to use AI to screen job applicants?
Who is responsible if an AI hiring tool discriminates?
What is a bias audit?
Can employees use ChatGPT or similar tools at work?
What is the safest way to start using AI in HR?
Sources
We checked the facts in this guide against these primary sources. Laws and thresholds change, so confirm current rules before you act.
- EEOC: Prohibited employment policies and practices
- NYC Department of Consumer and Worker Protection: Automated employment decision tools
- Illinois General Assembly: Artificial Intelligence Video Interview Act
- California Civil Rights Council: Automated decision systems regulations
- NIST: AI Risk Management Framework
Written by the HRTally Research Team
Our guides are written and reviewed by people with hands-on US HR, payroll, and HR technology experience. Legal and tax facts are checked against primary sources, such as the Department of Labor, IRS, EEOC, and USCIS, and linked at the end of each guide. Vendors never review or pay for our editorial content.