By Rovaryn Digital · 10 min read
Why RTW Coordinators Are Suddenly Reading AI Bills
Picture a typical Tuesday morning: a restricted-duty job description lands in your inbox at 8 a.m., the attending physician faxes an updated work-status form by 9, and your case-management tool surfaces three ranked light-duty options before your first cup of coffee is cold. The tool does exactly what you asked it to do — filter available positions against the physician's physical restrictions and sort them by fit.
Now picture a plaintiff's attorney — or a state labor agency investigator — asking one question: Who made the employment decision?
If your answer is "the software suggested it and we went with the top result," you may have a problem that has nothing to do with whether the duty match was medically appropriate. A new generation of state laws is drawing a compliance boundary around any automated or algorithmic tool that meaningfully shapes employment decisions — and return-to-work duty matching sits close enough to that boundary that every RTW coordinator and HR manager should understand where the line falls and how to document which side of it you are on.
This article explains what the emerging ai disclosure employment decisions framework looks like, how it applies to RTW workflows, and what a defensible, human-in-the-loop process requires in writing.
What "Automated Decision System" Means in the Laws That Matter
Three jurisdictions have moved furthest on employer-facing AI disclosure rules, and each uses slightly different language for the same core concept.
Colorado SB 26-189 (signed May 14, 2026; effective January 1, 2027). Colorado's Artificial Intelligence Act is the broadest. It covers "high-risk" automated decision systems — defined as systems that make, or substantially assist in making, "consequential decisions" about employment. Consequential employment decisions include hiring, firing, promotion, and — critically — the terms and conditions of employment. Whether a modified-duty assignment constitutes a term or condition of employment is a legal question, not a product question; verify the answer with qualified counsel. The law imposes obligations on deployers (employers who use the systems) as well as developers, including transparency notice and the ability for affected individuals to request human review. (SB 26-189 replaced the original 2024 Colorado AI Act, SB 24-205, and dropped its mandatory impact-assessment and risk-management-program requirements.) The law takes effect January 1, 2027, and enforcement is pending rulemaking and ongoing litigation; confirm the current operative date, scope, and any implementing regulations with the Colorado Attorney General's office or qualified counsel.
Illinois. Illinois has moved from narrow to broad. The Illinois AI Video Interview Act (effective 2020) governs the narrow use of AI to analyze video interviews. More significantly, HB 3773 — signed August 9, 2024 and effective January 1, 2026 — amended the Illinois Human Rights Act to prohibit employers from using AI in a way that produces discrimination against protected classes in recruitment, hiring, promotion, discipline, and other terms of employment. Confirm the current requirements and any Illinois Department of Human Rights guidance with counsel.
California. California has repeatedly attempted a broad automated-decision-system law without enacting one. AB 2930 was pulled by its sponsor in August 2024 before a floor vote — it was not vetoed — and a successor bill, SB 420, failed to advance out of committee in the 2025 session. As of this writing, no broad California ADS statute applies to private employers. Employers in California should treat this not as a settled "no" but as a regulatory direction of travel — toward disclosure, impact assessment, and human-review rights — and should structure workflows accordingly. Verify the current status of any California ADS requirement with counsel before publish.
The common thread across all three jurisdictions is a substantive requirement rather than a paperwork one: if software meaningfully shapes who gets what work assignment, the employer must be able to show that a human reviewed, understood, and made the final call — and must be able to explain the basis for that call to the affected worker on request.
Where Duty Matching Lives on the Spectrum
Not all software assistance is equal under these frameworks. A simple scheduling calendar that blocks hours has no AI-disclosure implications. A natural-language model that reads a physician's notes, scores a hundred open job descriptions against ten physical restrictions, and presents a ranked list is further along the spectrum — and the gap between "ranked recommendation" and "automated decision" is exactly where employers need clear internal documentation.
The relevant distinction is whether the system's output determines an outcome or informs a human who then determines an outcome. That distinction is not just semantic; it shows up in audit trails, grievance responses, and — in Colorado's framework — in the written notice provided to the worker.
Transitional Duty Manager's duty-matching feature operates as a ranked, explainable recommendation. The system surfaces candidate light-duty positions ordered by how well each position's task profile matches the physician's documented restrictions. A coordinator — a person with a name and a title — reviews the ranked list, considers factors the algorithm cannot weigh (the worker's seniority, a co-worker conflict, a supervisor's availability to provide accommodation support), and selects, modifies, or declines each option. The coordinator's choice is the employment decision. The software's output is the input to that decision, not the decision itself.
That distinction must be legible in your documentation — not just in how your IT team describes the product architecture, but in the records you produce if a carrier audits your program, a worker requests an explanation, or a state agency reviews your process. See how an audit-ready case file captures this chain of custody.
What a Compliant, Explainable Workflow Requires
A human-in-the-loop RTW process that holds up under AI-disclosure scrutiny has four documented components.
1. The algorithm's output is recorded as a recommendation, not a decision. The case file should show the ranked list the system produced — which positions were surfaced, in what order, and on what basis (restriction categories, task codes, match score). This is the explainability record. If a worker later asks "why was I assigned to the parts-sorting station instead of the receiving dock," you can point to the documented restriction profile and the task comparison, not just say "the system suggested it."
2. The coordinator's review is a documented affirmative act. A time-stamped entry showing that a named coordinator reviewed the recommendations, considered the physician's current restrictions (by date of the most recent work-status form), and selected the assignment is the dividing line between a human decision and a software output. A checkbox labeled "coordinator approved" is better than nothing. A date-stamped note field with the coordinator's name and a brief rationale is substantially stronger.
3. The affected worker receives an explanation on request. Under Colorado's framework — and the direction other states are moving — workers must be able to request human review and an explanation of any consequential employment decision in which an automated system played a role. Build your RTW program so that this explanation is retrievable: the restriction profile, the positions considered, the basis for the selection, and the name of the person who made the call. The ADA interactive process already requires a similar documentation posture — the AI-disclosure layer adds an explainability obligation on top of the accommodation record.
4. No auto-assignment, auto-scheduling, or auto-termination. These are not just product-design decisions — they are compliance constraints under both the emerging AI-disclosure framework and the foundational principle that duty matching is a human employment decision. Software should never move a worker from one assignment to another, end a transitional duty period, or trigger a benefits-suspension notice without a documented human review and approval at each step.
The Audit Trail Is the Compliance Record
AI-disclosure compliance is, at its operational core, a documentation problem. You do not need to stop using algorithmic tools; you need to be able to show what each tool did, when, and what a human did with the output.
The record that satisfies a carrier audit — physician restriction dates, job description approval, light-duty start and end dates, wage records for a reimbursement application — is largely the same record that satisfies an AI-disclosure inquiry. The additional layer is the explainability chain: what did the system recommend, and what did the coordinator decide.
If your current workflow produces that chain as a natural byproduct of the case file, you are in a defensible position. If the only record of how a duty assignment was chosen is "coordinator picked it," with no visible link to restriction data or position evaluation, you have a gap — not because the decision was wrong, but because you cannot explain it.
For a detailed walkthrough of what each document in a compliant RTW case file should contain and when it should be generated, the return-to-work case management guide covers the full lifecycle.
Practical Steps Before These Rules Mature
State AI-disclosure rules are early and fast-moving. Colorado's is the furthest along for employers; Illinois's is narrower than it may become; California's is in regulatory flux. The practical response is not to wait for each jurisdiction to finalize its framework — it is to build a workflow that would satisfy any of them, because the underlying logic (human reviews algorithmic output, makes a documented decision, can explain it) is consistent across all three.
Five things to confirm before your next open case:
- Map every automated or algorithmic step in your RTW workflow. For each step, document whether the output is a final action or a recommendation to a human.
- Assign a named decision-maker to each consequential step. "The system" is not a decision-maker. Every duty assignment, every modification to a restriction window, every close-out of a transitional period needs a name and a date.
- Verify whether your jurisdiction's ADS rules are in force. Colorado employers with operations or employees in that state should already be consulting counsel on SB 26-189 compliance. Illinois and California employers should confirm current requirements before assuming no obligation exists.
- Build the explainability record into your case file template now. Retrofitting documentation after a grievance or audit is harder than capturing it at assignment.
- Review your RTW program policy language. If your written policy says the coordinator "uses software to assign" transitional duties, update that language to reflect that the coordinator reviews software recommendations and makes the final selection.
The O*NET-based duty matching explainer walks through how restriction-to-task matching works at the task-statement level — useful background if you need to explain the basis of a duty recommendation to a worker or a reviewer.
The Disclosure Obligation Is Also a Program-Quality Signal
There is a practical upside to the documentation discipline that AI-disclosure rules are pushing toward. An RTW program that can explain every assignment — what restrictions applied, what positions were evaluated, who decided — is also a program that is harder for a carrier to challenge on reimbursement, harder for a plaintiff to characterize as pretextual, and more defensible in front of an administrative law judge.
The compliance obligation and the program-quality standard point in the same direction: document the human decision, not just the software output.
If you are building or auditing your RTW program documentation framework, the RTW Program Kit includes the written policy template, job description forms, coordinator-review checklists, and restriction-tracking worksheets that underpin a defensible, explainable process.
Nothing in this article is legal advice. AI-disclosure requirements vary by jurisdiction and are evolving rapidly. Confirm the current scope, effective date, and employer obligations for any state law with qualified employment counsel before making compliance decisions.
Want practical RTW compliance updates as AI-disclosure rules develop? Subscribe to the Transitional Duty Manager newsletter — brief, practitioner-focused, no hype.
Get the next RTW guide in your inbox
Practical guides and WA Stay-at-Work updates — no spam.
Automate the full RTW workflow
Transitional Duty Manager replaces manual RTW documentation with O*NET duty matching, WA SAW reimbursement packet export, and an immutable audit trail.
See how it works
