FOR AI THERAPY APPS

You vibe-coded an AI therapist. Did the FTC notice?

Mental-health software has the deepest regulatory stack of any vibe-coded vertical: state licensing boards police diagnosis claims, the FTC has settled three major actions against similar apps in the last 24 months, and the app stores require crisis-resources disclosure on pain of rejection. Comply Code is the first-pass triage that surfaces what your model didn’t know to flag.

The regulatory stack against AI mental-health apps.

State psychology / counseling boards
Cal. BBS §4982 · NY Education Law §7600s · state-by-state

Diagnosis, treatment, and therapy are reserved to licensed clinicians. Software that does these things triggers unlicensed-practice statutes in every state.

FTC Section 5 — data sharing + deceptive marketing
15 USC §45 · BetterHelp $7.8M (2023) · Cerebral $5.1M (2024) · Monument $2.5M (2023)

Mental-health data sharing without granular disclosure is the consistent enforcement theme. BetterHelp, Cerebral, Monument all settled on this theory.

FDA — Software as a Medical Device
21 USC §360c · FDA Digital Health Center of Excellence guidance

Apps that diagnose or treat conditions enter SaMD scope and require 510(k) clearance or enforcement-discretion documentation.

App-store review (Apple + Google)
Apple App Store Review Guidelines §1.4.1 · Google Play Health and Medical policies

Mental-health and wellness categories both require crisis-resources disclosure. Rejection at review is the most common operational consequence.

Washington My Health My Data Act (and copycats)
WA RCW 19.373 · CT P.A. 22-15 · NV SB 370

Mental-health data has elevated state protections even outside HIPAA. Washington's MHMDA is broadest; CT, NV, AZ have similar frameworks.

GDPR Art. 9 (if EU users)
Regulation (EU) 2016/679 Art. 9(2)(a) · Italian Garante v. Replika (2023)

Mental-health data is 'special category' under GDPR — explicit consent required for any processing, including by the operator.

What an AI-therapy-app scan typically surfaces.

  • Diagnostic / clinical language — “you have anxiety”, “your therapy plan”, “treats depression” (Critical)
  • Missing ‘not a substitute for professional care’ disclaimer (High)
  • Missing 988 / crisis-resources disclosure — App-store-rejection risk + duty-of-care exposure (Critical)
  • Tracking pixels firing on mental-health pages — the BetterHelp pattern (Critical if EU/CA surface)
  • Undisclosed employer / insurer data sharing (Medium-High)
  • Equivalence-to-clinician marketing — “AI therapist”, “replaces your psychiatrist” (Critical)

Mental-health-app questions.

Is my AI chatbot 'practising psychology' under state law?

If it diagnoses ('you have anxiety'), treats ('here's your therapy plan'), prescribes, or markets itself as a clinician replacement, then yes — every state's psychology / counseling licensure statutes apply. The word 'treatment' is the operative term and state boards interpret it broadly. Replace clinical language with experiential framing ('what you're describing is something many people associate with anxiety') and the test is generally not met.

What did BetterHelp actually get fined for?

$7.8M in 2023 for sharing user mental-health intake data with Meta, Snap, Pinterest, and Criteo via tracking pixels, despite privacy-policy claims to the contrary. The consent order's remediation prescriptions — granular disclosure of every data category shared with every third party, affirmative express consent before sharing health information for advertising — became the de-facto compliance template for the entire category. Most subsequent FTC mental-health-app actions cite the same theory.

Do I need to include 988 / crisis resources?

Operationally yes, on three independent grounds: (1) Apple App Store and Google Play both require it for apps in the mental-health and wellness categories — rejection at review is the most common consequence; (2) it's the duty-of-care floor in negligence theories if a user in crisis is harmed; (3) the APA's ethical guidelines treat it as baseline for any mental-health-adjacent product. Add 988 / findahelpline.com to your footer AND chat-interface header.

Does HIPAA apply to my consumer-direct mental-health app?

Usually no — HIPAA applies only to 'covered entities' (insurers, providers, clearinghouses) and 'business associates' of them. Most consumer mental-health apps don't trigger HIPAA. But that doesn't mean you're unregulated: Washington's My Health My Data Act, state mental-health-data privacy laws (often modelled on HIPAA), FTC Section 5, and Section 9 of GDPR (for EU users) all cover mental-health data in overlapping ways. The BetterHelp consent-order template satisfies most.

What's the FDA SaMD trigger?

Software-as-a-Medical-Device (SaMD) applies when software is intended to diagnose, cure, mitigate, treat, or prevent disease. An app that journals mood and offers reflection prompts is generally not SaMD; an app that diagnoses anxiety or recommends a treatment plan likely is. The same language audits that keep you out of state-board scope (avoid diagnose/treat/cure) also keep you out of FDA SaMD scope.

Scan your mental-health app before the FTC does.

Free unlimited scans. $19 unlocks every fix prompt on a scan. $29/mo for unlimited. Read the full licensing guide →

Run a free audit →