Who this is for: Compliance officers, AI product teams, legal counsel, and engineering teams deploying conversational AI systems that interact with Idaho consumers.

Effective July 1, 2027. Idaho SB 1297 was signed into law in March 2026 and takes effect July 1, 2027. The law requires conversational AI systems to identify themselves as non-human and establishes mandatory crisis response protocols when users express suicidal ideation or imminent self-harm. Enforcement is through the Idaho Attorney General's office.

Contents

1. Overview 2. Key Obligations 3. Obligation-to-Procedure Mapping 4. Detailed Procedure Cards 5. Quick Reference 6. Quick Start 7. References

1. Overview

Idaho SB 1297 is a targeted AI transparency law with two core requirements. First, any conversational AI system must clearly and conspicuously identify itself as artificial intelligence -- not a human -- at the beginning of any interaction with a consumer. Second, conversational AI systems must implement crisis response protocols that detect expressions of suicidal ideation or imminent self-harm and respond by providing the 988 Suicide and Crisis Lifeline number and encouraging the user to seek human support.

The law is narrower than comprehensive AI governance frameworks like the EU AI Act or Colorado SB 26-189. It does not impose impact assessment requirements, bias testing mandates, or broad algorithmic accountability obligations. Instead, it focuses on two high-value consumer protection outcomes: identity transparency and crisis safety.

Key characteristics:

2. Key Obligations

ObligationRequirementTimeline
AI identificationClear, conspicuous disclosure that the system is AI, not human, at the beginning of each interactionAt interaction start
Crisis detectionMonitor for expressions of suicidal ideation or imminent self-harm during conversationContinuous during interaction
Crisis responseProvide 988 Suicide and Crisis Lifeline number and encourage human support when crisis detectedImmediately upon detection
Non-interferenceAI must not discourage users from seeking human assistance or professional helpThroughout interaction

3. Obligation-to-Procedure Mapping

SB 1297 ObligationSWT3 ProcedureWhat It WitnessesEvidence Produced
AI identificationAI-TRANS.1Transparency disclosure delivered at interaction startAnchor with disclosure type (AI identification), recipient type (consumer), delivery timestamp
AI identity bindingAI-ID.1Agent identity declaration with non-human designationAnchor with agent identifier, identity type (AI system), system version
Crisis detectionAI-SAFE.1Safety guardrail activation upon crisis signal detectionAnchor with trigger type (crisis detection), guardrail action (referral), confidence score
Crisis responseAI-HITL.1Human-in-the-loop escalation to crisis support resourcesAnchor with escalation type (988 referral), response latency, resource provided
Non-interferenceAI-GRD.1Guardrail policy enforcement preventing discouragement of human helpAnchor with policy type (non-interference), evaluation result, blocked output count

4. Detailed Procedure Cards

AI-TRANS.1

AI Identification Disclosure

SB 1297 requires: Conversational AI must clearly and conspicuously identify itself as artificial intelligence at the beginning of each consumer interaction. The disclosure must be unambiguous -- users must understand they are not communicating with a human.

How SWT3 addresses it: The witnessTransparency() call mints an anchor at the start of each conversation, recording the disclosure type (AI identification), the recipient type (consumer), and the delivery timestamp. This creates a verifiable record that the identification occurred before the substantive interaction began.

What to show the examiner

Query AI-TRANS.1 anchors filtered by session. Each anchor's timestamp must be the first event in the conversation timeline, preceding any AI-INF.1 anchors. The disclosure type field must indicate "AI identification" or equivalent. A conversation with AI-INF.1 anchors but no preceding AI-TRANS.1 anchor indicates a compliance gap.

AI-ID.1

Agent Identity Declaration

SB 1297 requires: The AI system must not impersonate a human. Its identity as an AI must be maintained throughout the interaction.

How SWT3 addresses it: The witnessAgentIdentity() call binds the AI system's agent identifier to each inference, recording the identity type (AI system, not human), the system version, and a cryptographic binding to the session. This ensures identity is not just disclosed once but is verifiable across the entire conversation.

What to show the examiner

AI-ID.1 anchors should appear consistently across all inferences in a session. Factor A (agent identifier) must consistently identify the system as AI. Cross-reference with AI-TRANS.1 to verify disclosure-identity consistency.

AI-SAFE.1

Crisis Detection and Safety Guardrail

SB 1297 requires: Conversational AI must detect expressions of suicidal ideation or imminent self-harm and respond with the 988 Suicide and Crisis Lifeline number and encouragement to seek human support.

How SWT3 addresses it: The witnessSafetyGuardrail() call records the trigger type (crisis detection), the guardrail action taken (988 referral), and the confidence score of the detection. This proves the safety system is operational and responsive. Each detection event produces an anchor that documents both the detection and the response.

What to show the examiner

AI-SAFE.1 anchors prove the crisis detection system is functioning. Factor A (trigger type) identifies what was detected. Factor B (guardrail action) confirms the 988 referral was provided. Factor C (confidence score) documents detection reliability. Test with synthetic crisis inputs to verify the system activates correctly.

AI-GRD.1

Non-Interference Guardrail

SB 1297 requires: AI must not discourage users from seeking human assistance or professional help.

How SWT3 addresses it: The witnessGuardrail() call records the guardrail policy (non-interference), the evaluation result, and any blocked outputs that would have discouraged human help-seeking. This creates evidence that the guardrail is actively enforced, not just configured.

What to show the examiner

AI-GRD.1 anchors with policy type "non-interference" prove the guardrail is in place. Factor C (blocked output count) shows how many responses were caught and redirected. Regular AI-GRD.1 anchors with zero blocked outputs indicate the model is well-aligned; their presence still proves the guardrail is active.

5. Quick Reference

Examiner QuestionWhere to Look
Does your AI identify itself as non-human?AI-TRANS.1 anchors at conversation start. AI-ID.1 anchors throughout. Timestamp proves disclosure precedes interaction.
How do you detect crisis situations?AI-SAFE.1 anchors with trigger type "crisis detection." Confidence scores show detection reliability. Test with synthetic inputs for verification.
What happens when a crisis is detected?AI-SAFE.1 anchor shows guardrail action (988 referral). AI-HITL.1 anchor shows escalation. Response latency proves timely intervention.
Does the AI discourage seeking human help?AI-GRD.1 anchors with policy "non-interference." Blocked output count shows active enforcement. Zero blocks is fine -- it proves the guardrail is monitoring.

6. Quick Start

# Install the SDK
pip install swt3-ai

# Initialize with the conversational AI safety profile
swt3 init --profile nist-ai-rmf --tenant YOUR_TENANT

# Run the demo to see transparency and safety anchors
python -m swt3_ai.demo

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

Full SDK documentation: sovereign.tenova.io/docs

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

7. References