Crisis Safety Protocol

Drafted July 9, 2026 · Last revised August 9, 2026

1. Purpose and legal basis

FutureSelf includes an AI guide (“Coach”) that provides adaptive, natural-language responses across the app. California Business and Professions Code Chapter 22.6 (SB 243) includes the following duties for operators that fall within its scope:

• Maintain a protocol designed to prevent the chatbot from producing content related to suicidal ideation, suicide, or self-harm
• Refer users who express suicidal ideation, suicide, or self-harm to crisis service providers
• Publish details of that protocol on the operator’s website

This draft describes FutureSelf's internal protocol. Future Self LLC maintains it while counsel reviews applicability and wording.

2. Product scope

Coach is task-scoped AI guidance inside FutureSelf. It is disclosed as artificial intelligence at first contact and is not a human and not a health professional.

Your AI guide is artificial intelligence. It reflects your own words and supports your goals. It is not a substitute for professional care.

Companion AI features may not be suitable for some minors. FutureSelf is intended for adults age 18 and older.

3. Protocol overview

Our protocol has four layers that work together:

1. Prevent harmful generation: system instructions and response filters that prohibit Coach from producing content that encourages suicide, self-harm, or dangerous acts.
2. Detect user signals on-device: a deterministic, offline language screen that runs when you write free text, before that text is handled as ordinary coaching input.
3. Refer to crisis services: when a signal is detected, the app surfaces national crisis resources (988 and Crisis Text Line) in a calm, dismissible card or message.
4. Do not label or trap: detection never blocks you from saving your words, never risk-scores you, never stores a crisis label on your profile, and never replaces emergency services.

Release verification pending. Current source includes a support-type-aware app host plus local hosts for screened full-screen authoring and meditation surfaces. This draft must not be approved as a complete coverage claim until the combined crisis and eating-distress presentation paths are retested on the release candidate.

4. Preventing production of self-harm content

To reduce the chance that Coach generates suicidal ideation, suicide, or self-harm content:

• Coach system instructions prohibit encouraging self-harm, suicide, or dangerous acts.
• Post-generation quality monitors and coaching validators strip or reject outputs that violate safety and compliance rules.
• When a user message trips the crisis screen on a conversational surface, Coach does not continue an ordinary coaching turn. Instead, the app delivers a fixed safety response that discloses Coach is AI and points to 988 and Crisis Text Line.
• Marketing and in-app copy are written to avoid implying licensed healthcare authority (California AB 489).

No automated system is perfect. If Coach ever responds in a way that feels unsafe, stop the conversation and use the crisis resources above, then email support@futureslf.com.

5. Detecting user expressions of risk

Detection runs on your device, synchronously, without requiring a network call. That design choice exists so a safety tripwire cannot silently fail when you are offline.

5.1 What we look for

• Hard signals: explicit language about suicide, wanting to die, ending one’s life, self-harm, overdose, and closely related phrases.
• Soft signals: indirect phrasing associated with despair or hopelessness (for example “what’s the point,” “better off without me,” “can’t go on”), tuned to over-catch rather than miss. A false positive costs a calm resource card; a false negative can cost a life.
• Separate eating-distress screen: a narrower detector for genuine eating distress on nutrition surfaces. On a hit, the app offers eating-support resources (including Crisis Text Line pathways) rather than meeting every food description with a crisis card.

5.2 Where detection runs

The shared screen is wired to task-scoped AI surfaces and registered free-text persistence boundaries, including the daily reflection composer, day tasks and plans, current pillar intakes, meditation statements and reflections, Self Identity and Action authoring, Connection moments, and current onboarding input fields. Onboarding does not author a global identity statement; any retained free text is screened before persistence. Retired morning and evening writers are not current surfaces. A standing entry under Profile → Crisis support remains available without needing to trip a detector.

6. Crisis referral notification

When the protocol fires, FutureSelf is intended to provide a notification that refers you to crisis service providers, including:

• 988 Suicide and Crisis Lifeline: call or text 988 (available 24/7 in the United States)
• Crisis Text Line: text HOME to 741741

On conversational surfaces, the referral may appear as a Coach message that clearly states Coach is AI and that a real person is the right next step. On intake surfaces, a resource card offers a primary action to contact 988 and a dismiss action so you remain in control. The app does not place a call or send a text without your action.

7. What this protocol does not do

• It does not label or score you.
• It does not provide emergency dispatch or active rescue.
• It does not replace professional care.
• It does not guarantee that every expression of distress will be detected.
• It does not claim Coach is a health professional.

8. Minors

FutureSelf is intended for adults 18 and older and is not directed to children. Companion chatbots may not be suitable for some minors. We do not knowingly design or market Coach for minors. If we learn that a minor is using the Service, we will take appropriate steps consistent with the approved Terms and Privacy Policy.

If FutureSelf ever knowingly serves minors, SB 243 requires additional safeguards (repeated AI / break reminders during long sessions, and measures against sexually explicit content directed at minors). Those product controls are not the current adult-only posture; expanding to minors would require shipping them first.

9. Measurement and annual reporting

SB 243 includes annual-reporting requirements beginning July 1, 2027 for covered operators. If counsel determines that FutureSelf is covered, the company must confirm the required measures, privacy treatment, submission process, and first due date before launch. This draft does not itself establish that duty.

10. Governance

• Product and engineering own day-to-day operation of the detectors, prompts, and referral UI.
• Product safeguards are documented in our internal AB 489 / SB 243 standard and tested by automated crisis-surface tests. Counsel determines legal applicability and compliance.
• Material changes to this protocol will be reflected on this page with an updated revision date.

11. Contact

• Safety / protocol questions: support@futureslf.com
• Privacy: privacy@futureslf.com
• Operator: Future Self LLC

Related documents: Terms of Use · Privacy Policy · Support