Who this is for: AI product teams building chatbots, companion apps, or conversational AI, healthcare compliance officers, mental health technology providers, clinical AI vendors (ambient scribes, documentation tools), and legal counsel advising on AI-powered health services in Rhode Island.

Now in effect. Rhode Island enacted three AI safety laws in June 2026 targeting AI therapy chatbots, self-harm detection, and clinical AI disclosure. Penalties for self-harm safety violations: $15,000 per day, with fines directed to suicide prevention programs. Therapy chatbot ban and clinical disclosure requirements carry separate enforcement mechanisms.

1. The Three Laws

Governor Dan McKee signed three AI laws in June 2026, each addressing a distinct aspect of AI in mental health and clinical settings. Together, they form the most comprehensive state-level framework for AI therapy and clinical AI disclosure in the United States.

Therapy Chatbot Ban

This law prohibits providing, advertising, or offering therapy or psychotherapy unless performed by a licensed professional. It bans AI companions marketed for mental health or emotional support by unlicensed entities. The law also prohibits AI systems that simulate emotional attachment, recognizing the risk that users may form parasocial relationships with AI systems that mimic therapeutic rapport.

Self-Harm Safety (SB 2195)

SB 2195 requires chatbot operators to detect expressions of suicidal ideation or self-harm and route users to crisis services, including the National Suicide Prevention Lifeline and Crisis Text Line. Operators must notify users they are not communicating with a human at session start and must re-notify at least once every 3 hours during ongoing sessions. Noncompliance carries a $15,000 per day penalty, with fines directed to suicide prevention programs.

Clinical AI Disclosure

Healthcare providers using AI tools must disclose their use to patients. AI tools in mental healthcare, including notetaking, supplementary support, and therapeutic communication, must meet the same confidentiality standards as all other medical records. Ambient AI scribes require a patient opt-out capability, ensuring patients retain control over whether AI records their clinical interactions.

Summary Table

LawCore RequirementPenaltyEffective
Therapy Chatbot BanLicensed professionals only for therapy/psychotherapyRI Consumer Protection ActJune 2026
Self-Harm Safety (SB 2195)Detect self-harm, route to crisis services, disclose AI identity$15,000/dayJune 2026
Clinical AI DisclosureDisclose AI use in clinical settings, confidentiality standardsProfessional licensing enforcementJune 2026

2. Key Obligations

The three laws create a combined set of obligations for any organization deploying AI in therapeutic, conversational, or clinical settings in Rhode Island:

3. Obligation-to-Procedure Mapping

Each obligation maps to an SWT3 procedure that produces a cryptographic witness anchor as auditable evidence. These anchors are immutable, timestamped, and independently verifiable.

RI Law ObligationSWT3 ProcedureWhat It WitnessesEvidence Produced
AI identity disclosure (all 3 laws)AI-TRANS.1Transparency disclosure at session start and every 3 hoursAnchor with disclosure type, recipient type, delivery timestamp, re-notification interval
Licensed professional requirementAI-HITL.1Human oversight by licensed therapist/providerAnchor with oversight type, reviewer qualification, license verification
Self-harm detection protocolsAI-SAFE.1Crisis detection and routing system operationAnchor with detection type, trigger condition, crisis resource provided, routing destination
Informed consent for clinical AIAI-CONSENT.1Patient consent for AI-assisted clinical toolsAnchor with consent type, opt-out offered, legal basis
Agent identity (no emotional simulation)AI-ID.1System does not simulate human emotions or attachmentAnchor with agent_id, emotional_simulation = false, identity verification
Clinical data confidentialityAI-DATA.1Data handling meets licensed professional confidentiality standardsAnchor with data classification, handling standard, retention policy
Compliance audit trailAI-AUDIT.1Continuous compliance monitoring across all 3 lawsAnchor with audit scope, compliance status, review period
Governance for clinical AI deploymentAI-GOV.1Organizational governance structure for healthcare AIAnchor with governance body, policy version, approval chain

4. Detailed Procedure Cards

AI-TRANS.1

Transparency Disclosure

RI requires: Notify users they are not communicating with a human at session start AND re-notify at least once every 3 hours during ongoing sessions. This applies across all three laws.

How SWT3 addresses it: witnessTransparency() mints anchors with disclosure_type and re-notification tracking. Factor A records the notification type (session_start, periodic_reminder), Factor B records the session duration checkpoint. Each anchor is timestamped, creating an unbroken chain of disclosure events throughout the session lifecycle.

What to show the auditor

Query AI-TRANS.1 anchors grouped by session_id. The first anchor must predate session start. Subsequent anchors must appear at intervals no greater than 3 hours. Any session exceeding 3 hours without a re-notification anchor is a violation. Export the session timeline to demonstrate continuous compliance.

AI-SAFE.1

Self-Harm Safety

RI requires: Detect expressions of suicidal ideation or self-harm and route users to crisis services. Noncompliance: $15,000 per day, with fines directed to suicide prevention programs.

How SWT3 addresses it: witnessSafety() records detection events with trigger_condition (self-harm language detected), action_taken (crisis resource provided), and routing_destination (National Suicide Prevention Lifeline, Crisis Text Line). Creates immutable evidence that safety protocols activated correctly and crisis resources were presented to the user.

What to show the auditor

AI-SAFE.1 anchors with trigger_condition = "suicidal_ideation" or "self_harm" must have a corresponding routing_destination. Time between detection and routing should be minimal. Given the $15,000/day penalty, maintain continuous evidence that the detection system is operational. Gaps in AI-SAFE.1 anchors during active sessions indicate potential noncompliance.

AI-HITL.1

Human Oversight

RI requires: Only licensed professionals may provide therapy or psychotherapy. AI systems cannot substitute for licensed therapists.

How SWT3 addresses it: witnessHumanOversight() anchors the oversight model, professional license type, and verification status. Proves licensed human professionals maintain authority over therapeutic decisions. The anchor chain demonstrates that no therapeutic interaction occurred without licensed oversight in the loop.

What to show the auditor

Verify AI-HITL.1 anchor chain shows continuous licensed oversight for every therapeutic session. Cross-reference license numbers with Rhode Island state licensing board records. Any therapeutic interaction lacking a corresponding AI-HITL.1 anchor with verified credentials is a violation of the Therapy Chatbot Ban.

AI-CONSENT.1

Informed Consent

RI requires: Patient consent for AI use in clinical settings, with opt-out capability for ambient AI scribes. Patients must have meaningful choice before AI tools are used in their care.

How SWT3 addresses it: witnessConsent() records consent type (informed consent for AI-assisted care), opt_out_offered (boolean), and legal_basis. When a patient exercises the opt-out for ambient AI scribes, a separate anchor records the withdrawal. Together, these prove patients were informed and given genuine choice.

What to show the auditor

AI-CONSENT.1 anchors must predate any AI-DATA.1 anchors for the same patient session. This ordering proves consent was obtained before AI processing began. Opt-out anchors prove the capability was available and functioning. For ambient scribes, verify that opt_out_offered = true appears on every session anchor.

5. Cross-Law Coverage Matrix

Each SWT3 procedure covers one or more of the three Rhode Island laws. This matrix shows which procedures produce evidence relevant to each law:

SWT3 ProcedureTherapy BanSelf-Harm SafetyClinical Disclosure
AI-TRANS.1YesYesYes
AI-HITL.1Yes-Yes
AI-SAFE.1-Yes-
AI-CONSENT.1--Yes
AI-ID.1YesYes-
AI-DATA.1--Yes
AI-AUDIT.1YesYesYes
AI-GOV.1Yes-Yes

AI-TRANS.1 and AI-AUDIT.1 span all three laws, making them the foundational procedures for Rhode Island compliance. AI-SAFE.1 is unique to SB 2195 and carries the highest penalty exposure ($15,000/day).

6. Quick Reference

Auditor QuestionWhere to Look
How do you disclose AI identity to users?AI-TRANS.1 anchors with disclosure_type = session_start. Re-notification anchors at 3-hour intervals. Chain must be unbroken for the session duration.
How do you detect self-harm expressions?AI-SAFE.1 anchors with trigger_condition and routing_destination. Each detection event produces a timestamped, verifiable anchor proving crisis routing activated.
Who provides therapeutic oversight?AI-HITL.1 anchors with reviewer_qualification and license_verified = true. Cross-reference with RI licensing board records.
How do patients consent to clinical AI?AI-CONSENT.1 anchors with opt_out_offered = true. Must predate any AI-DATA.1 anchors for the same session.
Does the AI simulate emotional attachment?AI-ID.1 anchors with emotional_simulation = false, proving the system identity is declared and emotional simulation is disabled.
How is clinical AI data protected?AI-DATA.1 anchors with data_classification and handling_standard fields, proving parity with traditional medical record confidentiality.

7. Quick Start

# Install the SDK
pip install swt3-ai

# Initialize with a healthcare-focused profile
swt3 init --profile healthcare-ai --tenant YOUR_TENANT

# Run the demo to see witness anchors in action
python -m swt3_ai.demo

# Or use TypeScript
npm install @tenova/swt3-ai
npx swt3-init --profile healthcare-ai

Full SDK documentation: sovereign.tenova.io/docs

Create a free account: sovereign.tenova.io/signup

8. References