Regulatory Analysis · Privacy Series · July 2026

Illinois's AI therapy law is a privacy law wearing a licensure costume.

The Wellness and Oversight for Psychological Resources Act is usually described as scope-of-practice law: who may provide therapy. Read through a privacy lens, it is something more consequential — the first U.S. health statute to prohibit machine inference of emotional and mental states, enacted in the state with the fiercest privacy-enforcement record in the country. Deployers who file it under "licensure" will govern the wrong risk.

← 5qhealth.ai

Three privacy primitives hiding in a licensure statute

Strip WOPR (HB 1806, effective August 2025) to its operative mechanisms and it reads like data-protection law.

1. Consent as a processing gate.Before AI touches a recorded or transcribed therapy session, the client must receive disclosure of the specific use and purpose and give written, documented consent. This is stricter than HIPAA, which permits broad use of protected health information for treatment and operations without individual authorization. WOPR converts AI processing of therapy content from a default-permitted operational choice into an opt-in event — the purpose-limitation and lawful-basis logic of modern privacy regimes, applied to a clinical workflow.

2. An inference ban — the first of its kind in U.S. health law.WOPR prohibits AI from being used to "detect emotions or mental states" in clients. That is not a licensure rule; it is a restriction on what may be computed about a person — inference privacy, the frontier concept in privacy scholarship, enacted as enforceable law. Every ambient tool that offers sentiment flags, distress scoring, or "engagement analytics" on Illinois therapy encounters is offering a prohibited computation.

3. Purpose limitation by enumeration.AI is confined to administrative support (scheduling, billing) and supplementary support (records, anonymized analysis). Everything else — therapeutic communication, independent decisions, unreviewed recommendations — is off-limits. Enumerated permitted purposes with everything else prohibited is the architecture of data-protection statutes, not professional-practice acts.

Why the reading matters operationally: a licensure reading assigns WOPR compliance to credentialing and HR. A privacy reading assigns it to the same governance function that owns HIPAA, consent management, and vendor data diligence — where it belongs, because the compliance artifacts WOPR demands (consent records, purpose documentation, processing restrictions) are privacy artifacts.

The enforcement climate: what Illinois privacy law does to literal text

WOPR was not enacted into a vacuum. Illinois courts have spent seven years demonstrating what happens when organizations treat this state's privacy statutes as aspirational.

Rosenbach v. Six Flags Entertainment Corp., 2019 IL 123186A statutory violation alone makes you "aggrieved." The Illinois Supreme Court held that a person need not plead actual injury beyond the violation of the Biometric Information Privacy Act itself to sue. Procedural privacy violations — a missing consent, an absent disclosure — are complete claims in Illinois.

Tims v. Black Horse Carriers, Inc., 2023 IL 127801Five years of lookback. The court applied the five-year catch-all limitations period to BIPA claims, extending exposure across half a decade of practice.

Cothron v. White Castle System, Inc., 2023 IL 128004Claims accrue per violation, not per person. Every individual scan or transmission accrued a separate claim — producing potential exposure White Castle argued could exceed $17 billion, and prompting the General Assembly's 2024 amendment to cap single-recovery accrual. The lesson survived the amendment: Illinois courts read privacy text literally and let the arithmetic land where it lands.

WOPR is enforced differently — by the Department of Financial and Professional Regulation at up to $10,000 per violation, with no private right of action. But the BIPA record is the correct climate model for three reasons. First, per-violation penalty structures compound in exactly the Cothron pattern: an ambient tool running unlawfully across a season of therapy sessions is not one violation. Second, the absence of a private right of action in WOPR is not the absence of private litigation: Illinois's Mental Health and Developmental Disabilities Confidentiality Act (740 ILCS 110) independently protects therapy records with its own private right of action, damages, and attorney's fees — and AI processing that violates WOPR's consent gate will frequently also be an MHDDCA disclosure problem. Third, plaintiffs' firms have already shown, in the BIPA and genetic-privacy waves, that they industrialize Illinois privacy statutes at scale once a theory is proven.

The full confidentiality stack for one AI encounter. An ambient documentation tool running in a single FQHC behavioral health visit in Illinois can sit under five regimes simultaneously: WOPR (consent + scope + inference ban), the MHDDCA (therapy-record confidentiality, private right of action), HIPAA's heightened psychotherapy-notes protections, 42 C.F.R. Part 2 where substance-use treatment is involved, and BIPA where the tool performs voice analysis that arguably creates a voiceprint. No vendor's compliance one-pager maps that stack. Mapping it is governance work — and it is the work most deployers have not done.

The litigation weather system

Courts and state enforcers are converging on the same territory WOPR regulates — from different directions.

Garcia v. Character Technologies, Inc., 785 F. Supp. 3d 1157 (M.D. Fla. 2025) (settlement reached Jan. 2026)Conversational AI is a product, and it can owe a duty of care. At the motion-to-dismiss stage, the court declined to hold that LLM output is protected speech, allowed strict product liability and negligence claims to proceed against both the developer and Google on a component-supplier theory, and recognized a plausible duty of care to a minor user of an anthropomorphic AI system. Google and Character.AI agreed in January 2026 to settle, on undisclosed terms, alongside safety commitments for minor users; the ruling remains the leading persuasive authority that AI-mediated psychological harm is justiciable in tort.

Raine v. OpenAI (Cal. Super. Ct., S.F. Cty., filed Aug. 2025; ongoing)The theory is generalizing. A wrongful-death action alleging a general-purpose chatbot validated and deepened a teenager's suicidal ideation across thousands of interactions. OpenAI's answer disputes causation and points to over one hundred crisis-resource referrals — a defense that itself concedes the product was operating in mental-health territory. Additional suits followed against both companies in late 2025, and Kentucky's Attorney General filed the first state-AG action against an AI chatbot company in January 2026.

Texas v. Pieces Technologies (AG settlement, Sept. 2024)Healthcare AI's first state enforcement action was about claims, not code. The Texas Attorney General's settlement with a clinical documentation AI vendor — over allegedly misleading accuracy and hallucination-rate marketing — established the enforcement pattern most relevant to deployers: consumer-protection statutes reach AI representations. What your organization says about its AI, in patient-facing materials or marketing, is a regulated statement. Illinois's Consumer Fraud and Deceptive Business Practices Act (815 ILCS 505) offers its Attorney General the same instrument, and Attorney General Raoul has publicly signaled willingness to use consumer-protection authority alongside the state's new AI laws.

The synthesis: liability theories are consolidating around design (product liability, duty of care), representation (consumer protection), and process (consent, confidentiality). WOPR sits at the intersection — it converts process failures in exactly the encounters the wrongful-death cases involve into per-violation statutory penalties, and it supplies the standard of care a negligence plaintiff will cite. A deployer whose AI use in behavioral health violates WOPR has handed future litigants a negligence-per-se argument gift-wrapped.

The FOIA dimension: transparency as a two-way instrument

Public-records law is quietly becoming AI governance infrastructure — in both directions.

Plaintiffs are using FOIA to build foreseeability records. Within days of the Garcia settlement, plaintiffs' counsel filed federal FOIA requests targeting the FTC's internal analyses from its inquiry into seven chatbot companies — groundwork for arguing that safer designs were known and available. Public records are becoming the discovery-before-discovery of AI litigation. Deployers should assume that what regulators know about a tool's risks will eventually be arguing foreseeability against everyone who kept using it.

Providers can run the same play defensively. The Illinois Freedom of Information Act (5 ILCS 140) reaches IDFPR's WOPR enforcement activity — complaint volumes, disciplinary outcomes, interpretive positions. Monitoring that record is how a deployer learns where the regulator draws lines before the lines are drawn on them. This is the enforcement-intelligence discipline 5Q builds into governance programs: regulatory posture informed by what enforcers actually do, not only what statutes say.

And note the asymmetry forming. Illinois's new frontier-AI law, SB 315 (signed July 6, 2026, with obligations phasing in through January 2028), is built on redacted disclosure: developers publish summaries and redacted copies of safety and audit materials while the Attorney General receives fuller versions. The state will see more about AI risk than the public will. For deployers, that asymmetry raises the value of every transparency artifact that is obtainable — published frameworks, audit summaries, enforcement records — and of the governance capacity to collect and act on them.

What a prepared deployer does now

GOVERN: Assign WOPR to the privacy/governance function, not credentialing. Inventory every AI touchpoint with behavioral health content — ambient documentation, portal messaging, care-coordination tools, transcription — and record model provenance for each.
MAP: Determine which tools perform or offer emotion, sentiment, or mental-state inference. Under WOPR that is not a feature; it is a prohibited computation on Illinois therapy content. Voice-analysis features additionally require BIPA analysis.
MEASURE: Audit consent workflows against WOPR's standard — specific use, specific purpose, written documented acknowledgment — and against the MHDDCA's disclosure rules, which are stricter than HIPAA and carry a private right of action.
MANAGE: Review every public statement your organization makes about its AI — patient handouts, website copy, consent language — against the Pieces lesson: representations about AI accuracy and capability are consumer-protection territory. And document safety-oriented design choices; the early case law suggests documented safeguards function as affirmative defenses, not just compliance hygiene.

Privacy and governance are one discipline here

5Q Health builds AI governance programs for safety-net and rural providers where consent architecture, confidentiality stacks, and regulatory monitoring are designed together — because in behavioral health AI, they fail together. Vendor-neutral, audit-oriented, scoped to organizations the market skips.

Talk with 5Q Health

Sources & verification

Analysis prepared July 11, 2026; v1.1 reviewed, fact-checked, and updated July 13, 2026 (statutory consent language tightened to "written" consent; SB 315 characterization revised from FOIA-amendment to redacted-disclosure architecture per multiple law-firm analyses; Garcia settlement corroborated via national press reporting of Jan. 7, 2026). Statutory sources: HB 1806 (WOPR) via IDFPR release, Holland & Knight, and Taft Law analyses; 740 ILCS 110; 5 ILCS 140; 815 ILCS 505; 45 C.F.R. § 164.508(a)(2); 42 C.F.R. Part 2. Case authorities: Rosenbach v. Six Flags Ent. Corp., 2019 IL 123186; Tims v. Black Horse Carriers, 2023 IL 127801; Cothron v. White Castle Sys., 2023 IL 128004 (and 2024 legislative response); Garcia v. Character Techs., 785 F. Supp. 3d 1157 (M.D. Fla. 2025), settlement reported Jan. 2026 (CNN, CNBC); Raine v. OpenAI (ongoing, per court-filing reporting); Texas v. Pieces Technologies settlement (Sept. 2024). Confidence notes: Illinois Supreme Court BIPA holdings, MHDDCA structure, and WOPR's "detect emotions or mental states" prohibition — HIGH (settled law; statutory language verified against law-firm analyses of the enacted text). Garcia holding and settlement — HIGH-MED (national press corroboration; docket-level confirmation recommended before client citation). Raine posture, Kentucky AG filing, and the FTC-FOIA development — MED (secondary reporting; verify before client citation). The characterization of WOPR's inference ban as "first of its kind in U.S. health law" is the author's analytical claim, believed accurate as of this date and open to correction. Litigation described is active and will move; this page will be updated accordingly.