Ch 7 — Compliance & Risk

Navigating the regulatory landscape for AI in employment — federal, state, local, and international requirements
Regulatory content current as of November 2026. Verify before acting on legal guidance
High Level
account_balance
Federal
arrow_forward
map
State
arrow_forward
location_city
Local
arrow_forward
public
Global
arrow_forward
fact_check
Audit
arrow_forward
shield
Govern
-
Click play or press Space to begin the journey...
Step- / 8
gavel
The Regulatory Landscape
AI in employment is now regulated at every level — and the landscape is moving fast
The New Reality
AI in employment is now regulated at every level of government: federal (Title VII, ADA, ADEA), state (California, Illinois, Texas, Colorado, Connecticut, New Jersey), local (NYC), and international (EU AI Act). Federal agencies have pulled back: the EEOC removed its 2023 AI guidance and, since June 2026, no longer pursues disparate impact cases. That does not reduce employer liability. Title VII, ADA, and ADEA are unchanged, and private plaintiffs and state agencies can still sue over AI that produces disparate impact. As an HR ops leader, compliance with AI regulations is a present-day operational requirement.
Why It’s Accelerating
High-profile cases: Amazon’s biased hiring AI, HireVue’s facial analysis controversy, the iTutorGroup EEOC settlement, Mobley v. Workday (an age-bias collective action against an AI screening vendor, still in litigation), and the Eightfold AI class action (Jan 2026), which alleges secret applicant scoring dossiers that may violate FCRA, put AI bias squarely on lawmakers’ radar.

State momentum: California CRD regulations (Oct 2025), Illinois HB 3773 (Jan 2026), and Texas HB 149 (Jan 2026) are in force, and 2026 added more: California’s SB 947 and Connecticut’s Public Act 26-15 both take effect in 2027. Colorado’s first-in-the-nation AI Act (SB 24-205) was repealed before it ever took effect and replaced by a narrower notice law (see State Laws below). At the federal level the Trump administration is pushing the opposite direction — Executive Order 14365 (Dec 2025) created a DOJ task force to challenge state AI laws — so the near-term story is preemption fights, not harmonizing legislation.

Public awareness: Employees and candidates increasingly know that AI is making decisions about them — and they expect transparency.
Regulatory Layers
FEDERAL (applies everywhere) Title VII, ADA, ADEA apply to AI decisions // EEOC dropped disparate impact cases // (June 2026); statutes still in force, // private and state suits continue FCRA may apply to AI scoring tools STATE (varies by jurisdiction) California: CRD Regs (Oct 2025) — most detailed yet: bias testing, human oversight, 4-yr recordkeeping Also SB 947 (July 2027) and CCPA ADMT rules (Jan 2027) Illinois: HB 3773 (Jan 2026) + AI Video Interview Act Texas: TRAIGA / HB 149 (Jan 2026) Colorado: SB 26-189 (Jan 2027) Connecticut: PA 26-15 (2026–27) New Jersey: disparate impact rules Maryland: Facial recognition consent LOCAL (city-level) NYC Local Law 144: Bias audits required Other cities: none known as of Nov 2026 INTERNATIONAL (if you operate globally) EU AI Act: High-risk classification Canada: AIDA lapsed in 2025 UK: Pro-innovation AI framework WATCH (as of Nov 2026) No federal AI law has passed; Washington is pushing to preempt state AI rules instead
Ops reality: The removal of federal guidance does not create a compliance vacuum — it makes the landscape harder to navigate, not easier. You need to know which regulations apply to your organization, which AI tools are in scope, and who owns compliance. This chapter gives you the map.
account_balance
EEOC & Federal Requirements
Title VII applies to AI decisions — and the employer stays liable, whatever the vendor promised
The Legal Foundation
Title VII of the Civil Rights Act prohibits employment discrimination based on race, color, religion, sex, or national origin. The EEOC removed its 2023 guidance on AI in hiring, and in June 2026 it went further: its new enforcement plan says the agency will not pursue disparate impact cases, and a Justice Department opinion called the EEOC’s disparate impact guidelines unconstitutional. The underlying law has not changed. Neither action amends Title VII or binds the courts. If your AI screening tool disproportionately rejects candidates from a protected class, private plaintiffs and state agencies can still bring that claim.

Private litigation is now the main front: In Mobley v. Workday, a federal court let age-bias claims proceed as a collective action against an AI screening vendor (no liability finding yet; Workday is seeking to decertify). The Eightfold AI class action (January 2026) alleges the platform created secret applicant scoring dossiers without disclosure, potentially violating the Fair Credit Reporting Act and California consumer reporting laws. This case signals a new front for AI hiring liability beyond Title VII.
Disparate Impact vs. Disparate Treatment
Disparate treatment: Intentionally treating people differently based on protected class. Rare in AI, but possible if protected characteristics are used as input features.

Disparate impact: A facially neutral practice that disproportionately affects a protected group. This is the primary AI risk. An algorithm doesn’t need to “intend” to discriminate — if the outcome is discriminatory, you have a legal problem.
The Four-Fifths Rule
The Four-Fifths (80%) Rule: The selection rate for any group must be at least 80% of the rate of the group with the highest selection rate. Example: Group A selection rate: 60% Group B selection rate: 40% Ratio: 40/60 = 66.7% Below 80% = flag for adverse impact review // This applies to every stage where AI // is making or influencing selection: // resume screening, interview scheduling, // assessments, final recommendations
Critical: You are liable as the employer even if the vendor built the AI. The EEOC’s retreat does not change this: using an algorithmic tool does not change an employer’s obligations under Title VII. The statute, not the agency’s priorities, creates the liability. You cannot outsource compliance to a vendor.
map
State Laws: The Patchwork
A growing web of state-level AI regulations with different requirements
Key State Laws
California CRD Regulations (effective October 2025): The most detailed state AI employment regulations yet. Unlawful to use automated decision systems that discriminate on protected traits. Requires 4-year recordkeeping, and treats bias testing and meaningful human oversight as evidence in your defense if a tool is challenged, which in practice makes both expected. This is now the most detailed state standard.

California SB 947 (signed Sept 2026, effective July 2027): The “No Robo Bosses Act.” Employers may not discipline or fire a worker based solely on an automated decision system; a human must corroborate the output, and the worker gets written notice. Separately, California’s CCPA automated-decision rules bring pre-use notice, opt-out, and access rights to employment decisions from January 2027; their risk-assessment duties began phasing in during 2026.

Colorado SB 26-189 (effective January 2027): Replaced the original Colorado AI Act (SB 24-205), which was repealed in May 2026 before it took effect. The impact assessments and risk programs are gone. What remains: notice before an automated tool materially influences a decision, an explanation within 30 days of an adverse outcome, and a path to human review. Enforced by the Attorney General only.

Illinois HB 3773 (January 2026): Amends the Illinois Human Rights Act. Using AI that has a discriminatory effect in hiring, promotion, discipline, or discharge is a civil rights violation, and so is failing to notify people that AI is being used.

Texas TRAIGA / HB 149 (January 2026): Applies to private employers as well as government agencies. AI cannot be used with intent to discriminate, but the law rejects disparate impact as standalone liability, diverging from California and Illinois. There is no private right of action; enforcement runs through the Attorney General.

Connecticut PA 26-15 and New Jersey: Connecticut requires written notice before AI-assisted employment decisions from October 2027. New Jersey’s disparate impact rules (Dec 2025) expressly cover automated employment tools.

Illinois AI Video Interview Act (2020): Employers must notify candidates that AI is analyzing their video interview, explain how the AI works, and obtain consent.

Maryland HB 1202 (2020): Bans the use of facial recognition in job interviews unless the applicant signs a waiver.
The Compliance Patchwork Problem
Risky Approach
Trying to comply with each state’s law individually. Different disclosure language for Illinois candidates, different consent forms for Maryland, different data rights for California. Creates operational complexity, errors, and audit nightmares.
Smart Approach
Use the strictest standard as your baseline. If Illinois requires AI disclosure for video interviews, disclose for all candidates everywhere. If California grants data rights, extend them to all employees. One baseline, universally applied, plus a check of each jurisdiction’s specific duties.
Strategy: Build your AI governance to the highest common standard — currently California CRD. It’s simpler to administer, reduces legal risk, and future-proofs you as more states pass laws. No federal AI law has passed, and Washington is moving to preempt state AI rules (Executive Order 14365, Dec 2025), so expect a federal-vs-state fight, not harmonization. At the state level the trend is still toward more regulation, not less.
location_city
NYC Local Law 144: The Template
The first major local law on AI in hiring — and why other cities are watching
What It Requires
NYC Local Law 144, enforced since July 2023, regulates automated employment decision tools (AEDTs) — tools whose output “substantially assists or replaces” human decisions in hiring or promotion for jobs in New York City. The requirements are specific:

1. Annual bias audit: An independent third party must conduct a bias audit annually, analyzing selection rates by race/ethnicity and sex.

2. Published results: The audit summary must be publicly posted on the employer’s website.

3. Candidate notification: Candidates must be notified at least 10 business days before the AEDT is used, told what job qualifications the tool assesses, and told how to request an alternative process or accommodation (the law requires the instructions, not that the request be granted).

Enforcement is tightening: A December 2025 New York State Comptroller audit found the city had barely enforced the law, and the city’s consumer protection agency agreed to step up.
Enforcement & Impact
Enforcement status: Penalties: up to $500 first violation $500-$1,500 each subsequent // Each day of use counts separately. // Even so, the real risk // is the reputational damage of publishing // a bias audit that shows adverse impact What counts as an AEDT: YES: AI resume screeners, chatbot evaluators, ML-scored assessments, AI interview analysis NO: Spam filters on applications, scheduling bots, basic keyword search in ATS Why it matters nationally: // NYC was first. No other city is known to // have copied it, but states are passing // notice and review laws. Prepare now.
Ops action: If you use any AI in hiring — even through a vendor — inventory those tools now. Determine if they qualify as AEDTs. If you hire in NYC (or plan to), you need a bias audit process in place. Even if you don’t, state laws in Illinois, Colorado, and California now reach the same tools.
public
EU AI Act: High-Risk Classification
Employment AI is classified as “high-risk” — the strictest regulatory tier
What the EU AI Act Means for HR
The EU AI Act, which began phased enforcement in 2024, classifies AI systems by risk level. (The EU’s “Digital Omnibus,” adopted in mid-2026, moved the high-risk employment-AI obligations from August 2026 to December 2027.) Employment AI is explicitly classified as “high-risk,” meaning it faces the strictest requirements short of an outright ban. (Some uses are banned already: inferring employees’ emotions from their faces or voices at work has been prohibited since February 2025.) The high-risk category covers AI intended for: recruitment, resume screening, job advertising, interview evaluation, promotion decisions, termination decisions, task allocation, and performance monitoring.

If your organization employs people in the EU, or uses AI tools whose output is used in the EU, this applies to you.
High-Risk Requirements
// Most of these fall on the vendor (provider). // As the employer (deployer) you must use the // tool as instructed, assign human oversight, // keep its logs 6 months, and tell workers // before it is used on them. Conformity assessment: Formal evaluation that the AI system meets all EU AI Act requirements before deployment Data governance: Training data must be relevant, representative, free of errors, and complete Human oversight: Humans must be able to understand, monitor, and override AI decisions Transparency: Users must be informed when AI is being used and understand how it works Record-keeping: Automatic logging of all AI system operations for traceability and audit Penalties: High-risk duties: up to €15 million or 3% of global revenue // Banned practices: up to €35M or 7%
Global impact: Like GDPR before it, the EU AI Act is setting the global standard. Even if you only operate in the US, your AI vendors are likely building to EU requirements — and US regulations are borrowing heavily from this framework.
handshake
Vendor Liability: Who’s Responsible?
If your AI vendor’s tool discriminates, YOU are liable as the employer
The Liability Reality
Here is the most important legal principle in AI compliance: if your vendor’s AI tool discriminates, you — the employer — are liable. A vendor may end up liable as well (that is the question in Mobley v. Workday), but that does not take you off the hook. Courts and regulators have been consistent on this: employers cannot outsource their Title VII obligations. A vendor’s AI is your tool once you deploy it. If it creates adverse impact, it’s your problem in court.
Vendor Due Diligence
Before signing: Request bias audit results, model documentation, and data governance practices. If they can’t provide these, walk away.

During contract: Negotiate audit rights, indemnification clauses, and transparency requirements.

After deployment: Conduct your own bias monitoring. Don’t rely solely on vendor assurances.
What Your Contracts Need to Say
Must-Have Contract Provisions: 1. Bias audit rights Right to conduct independent bias audits at any time; vendor pays if bias is found 2. Model documentation Vendor must provide model card, training data description, and validation results 3. Indemnification Vendor indemnifies employer for claims arising from AI bias or discrimination 4. Audit access Employer can inspect the model, data, and decision logic upon request 5. Data deletion All candidate/employee data deleted upon contract termination 6. Change notification Vendor must notify before any model updates that could affect outcomes
Red flag: If a vendor resists any of these provisions, that tells you something about their confidence in their own product. Serious AI vendors welcome transparency requirements because they’ve already done the work.
checklist
Building a Compliance Framework
A practical framework for AI governance in HR operations
The Five-Step Framework
1. Inventory: Catalog every AI tool in use across HR — including AI features embedded in your ATS, HRIS, and other platforms that you may not think of as “AI tools.”

2. Classify: Assign a risk level to each tool based on what decisions it influences: high (hiring, firing, compensation), medium (scheduling, routing, categorization), low (drafting, summarization, search).

3. Assign ownership: Every AI tool needs a named owner responsible for compliance, monitoring, and incident response.

4. Schedule audits: High-risk tools need bias audits at least annually. Medium-risk tools need periodic review. Low-risk tools need annual inventory confirmation.

5. Document everything: Every decision about AI deployment, every audit result, every incident — documented and retained.
Compliance Checklist Template
AI COMPLIANCE CHECKLIST [ ] Complete AI tool inventory [ ] Classify each tool by risk level [ ] Assign compliance owner per tool [ ] Review vendor contracts for required clauses [ ] Obtain/schedule independent bias audits [ ] Verify candidate/employee notification process [ ] Document AI decision-making rationale [ ] Establish incident response protocol [ ] Set up ongoing monitoring cadence [ ] Map applicable regulations by jurisdiction [ ] Train HR team on AI compliance basics [ ] Brief legal counsel on AI tool usage [ ] Schedule annual governance review [ ] Create regulatory change monitoring process
Start here: If you do nothing else from this chapter, complete the AI tool inventory. You can’t govern what you can’t see. Most organizations are surprised by how many AI-powered tools they’re already using.
shield
The Risk Register
Tracking AI risks alongside your existing risk management processes
Why a Dedicated AI Risk Register
AI risks are different from traditional HR technology risks. They involve model drift (performance degrades over time), emergent bias (bias that appears only at scale), and regulatory velocity (laws changing faster than your compliance cycle). A dedicated AI risk register, integrated with your existing risk management framework, gives you visibility and control.

Each AI tool should have an entry tracking: what it does, what data it uses, what decisions it influences, who owns it, when it was last audited, and its current compliance status across all applicable jurisdictions.
Risk Register Template
AI RISK REGISTER Tool: [Name of AI tool/feature] Vendor: [Vendor name] Use case: [What it does in your process] Risk level: High | Medium | Low Data used: [PII, demographics, performance, etc.] Sensitivity: [Data classification level] Owner: [Named individual] Last audit: [Date of last bias/compliance audit] Next audit: [Scheduled date] Jurisdictions: [Which regulations apply] Status: Compliant | Review needed | Non-compliant Incidents: [Count and date of last incident] Notes: [Open issues, pending actions]
Integration tip: Don’t create a separate risk management process for AI. Add AI risk entries to your existing risk register and governance review cadence. AI risk is just a new category of operational risk — and you already know how to manage that.
Sources & Further Reading
• Title VII of the Civil Rights Act of 1964; Americans with Disabilities Act; Age Discrimination in Employment Act
• EEOC, “Select Issues: Assessing Adverse Impact in Software, Algorithms, and AI” (guidance removed 2025)
• NYC Local Law 144 (2023) — Automated Employment Decision Tools
• California Civil Rights Department, Regulations on Automated Decision Systems (effective Oct 2025)
• EEOC National Enforcement Plan, FY 2025–2029 (June 2026); DOJ Office of Legal Counsel opinion on disparate impact liability (June 2026)
• Colorado SB 26-189 (signed May 2026, effective Jan 2027; replaced SB 24-205)
• California SB 947 (effective July 2027); CCPA automated decisionmaking regulations (employer compliance Jan 2027)
• Illinois HB 3773 (effective Jan 2026); Texas HB 149 (effective Jan 2026)
• Connecticut Public Act 26-15; New Jersey N.J.A.C. 13:16 (Dec 2025)
• EU Artificial Intelligence Act, Regulation 2024/1689, as amended by the 2026 Digital Omnibus
• Mobley v. Workday (N.D. Cal.); Eightfold AI class action (Jan 2026)
• New York State Comptroller, audit of Local Law 144 enforcement (Dec 2025)

Regulatory content current as of November 2026. Verify before acting on legal guidance.