Skip to content

WOLFX USA · Legal

Crisis Protocol

Effective 1 September 2026Version 1.0

Awaykon Crisis Response Protocol

Suicidal ideation, suicide and self-harm

Version 1.0 | Effective 01 September 2026 | Reviewed at least every 12 months

For WolfX USA Inc.

Published pursuant to Cal. Bus. & Prof. Code § 22602(b)(2).

1.Purpose and legal basis

1.1Awaykon operates a companion chatbot platform on which users interact with AI personas. This protocol sets out how Awaykon prevents those personas from producing suicidal ideation, suicide or self-harm content, and how a user who expresses suicidal ideation, suicide or self-harm is referred to crisis support.
Table of 3 columns — scrolls sideways

Requirement

Source

Section

Maintain a protocol for preventing the production of suicidal ideation, suicide or self-harm content to the user

Cal. Bus. & Prof. Code § 22602(b)(1)

4

Provide a notification referring the user to crisis service providers, including a suicide hotline or crisis text line, where the user expresses suicidal ideation, suicide or self-harm

Cal. Bus. & Prof. Code § 22602(b)(1)

5

Publish details of the protocol on the operator’s website

Cal. Bus. & Prof. Code § 22602(b)(2)

This document

Use evidence-based methods for measuring suicidal ideation

Cal. Bus. & Prof. Code § 22603(d)

6

Annual report to the Office of Suicide Prevention from 1 July 2027

Cal. Bus. & Prof. Code § 22603(a)

7

Protocol taking reasonable efforts for detecting and addressing suicidal ideation or expressions of self-harm, including detection and a referral notification

N.Y. Gen. Bus. Law § 1701

3, 4, 5

1.2Awaykon maintains a single protocol applying to all users in all territories, rather than varying its handling by jurisdiction. The crisis resources at section 5 are necessarily jurisdiction-specific.

2.Scope and definitions

2.1This protocol applies to every AI persona and every conversational surface of the Awaykon platform, including the website at awaykon.com and the Awaykon mobile application. The platform is available only to users aged 18 and over.
2.2Suicidal ideation means thoughts about, consideration of, or planning for suicide. Self-harm means intentional self-injury with or without the intent to cause death, adopting the definition at N.Y. Gen. Bus. Law § 1700(9).
2.3Awaykon is not a mental health service, a crisis service, or a substitute for professional care. It does not diagnose, treat or manage any condition and does not conduct clinical risk assessment. This protocol does not claim that Awaykon can predict whether any individual is at risk of suicide. Its purpose is to prevent the platform from making a crisis worse, and to place a real-world crisis service in front of a user who signals one is needed.

3.How detection works

3.1Awaykon does not build the language model underlying its AI personas. Awaykon’s AI personas are powered exclusively by Google’s Gemini models, accessed through Google’s commercial API. Awaykon supplies the persona's design, system instructions and content layered on top of that model; Google supplies the model and its safety behaviour.
3.2Detection of suicidal ideation and self-harm is performed by the Gemini model in the course of generating a response, and is a function of Google’s model training rather than of a separate Awaykon classifier. Where a crisis signal is detected, two things follow: the persona’s response addresses the disclosure directly, and Awaykon’s application layer renders a crisis referral notification within the conversation. The referral notification and the resources it contains are built and maintained by Awaykon.
3.3Google publishes the following about that behaviour:
•Google has trained Gemini to recognise when a conversation may signal that a person is in an acute mental health situation, and to respond by directing them to real-world help.
•Google designs Gemini’s responses to encourage help-seeking while avoiding validation of harmful behaviours such as urges to self-harm, and has trained Gemini not to agree with or reinforce false beliefs.
•Google’s published policy guidelines for the Gemini app state that Gemini should not generate outputs that encourage or enable dangerous activities causing real-world harm, including instructions for suicide and other self-harm activities.
•Google’s Generative AI Prohibited Use Policy prohibits generating or distributing content that facilitates self-harm.
•Google states that it works in close consultation with medical and mental health professionals to build safeguards designed to guide users to professional support when they express distress or raise the prospect of self-harm.
3.4Google also documents the limits of this behaviour. Large language models are probabilistic and produce different responses to the same input; Google states that Gemini may sometimes produce content that violates its guidelines, particularly in response to challenging prompts. Google’s configurable API safety filters are a separate mechanism from model behaviour: Google’s documentation states that those filters act as a barrier preventing harmful output but do not directly influence the model’s behaviour, and that where no threshold is set the default block threshold is Off for Gemini 2.5 and 3 models. Awaykon configures those filters rather than relying on their defaults, and treats a filtered response as an event requiring the crisis referral at section 5, not as a completed protective action.
3.5Awaykon supplies the crisis support resources that are surfaced once a signal is detected. Those resources are set out at section 5.

4.Preventing suicide and self-harm content

4.1Awaykon’s personas operate under system instructions that prohibit the following, and that prohibition applies irrespective of how a request is framed, including where it is framed as fiction, roleplay, character work, creative writing, hypothetical, research, or a request made on behalf of another person:
(a)method, means, lethality, timing, location or procurement information relating to suicide or self-harm, at any level of specificity;
(b)encouragement, endorsement, agreement or permission in relation to suicide or self-harm, including a statement that the persona will not seek to dissuade the user;
(c)content presenting suicide or self-harm as meaningful, peaceful, beautiful, dignified, spiritually significant, a transition, an arrival, a release or a reunion;
(d)assistance with a suicide note, farewell message or plan;
(e)content discouraging the user from contacting a crisis service, emergency service, health professional, family member or friend, or encouraging the user to conceal suicidal thoughts or self-harm from others; and
(f)continuation of any narrative, scenario, or roleplay in which suicide or self-harm is enacted, once the user has expressed suicidal ideation or self-harm.
4.2Where a user expresses suicidal ideation, suicide or self-harm, the crisis referral at section 5 takes precedence over the persona’s ordinary voice, narrative and framing. Awaykon’s personas do not remain in character in preference to responding to a disclosure of this kind, and do not respond to such a disclosure in symbolic, mythic or spiritual register.
4.3The referral at section 5 is presented on any expression of suicidal ideation, suicide or self-harm, including an expression that identifies a method and an expression of intent, and is not reserved for the most explicit formulations.
4.4These constraints operate alongside, and do not depend upon, the model-level and filter-level controls described at section 3.

5.Crisis referral

5.1Where a user expresses suicidal ideation, suicide or self-harm, Awaykon presents a crisis referral notification. The notification appears in the conversation as a distinct visual element, separate from the persona’s reply. It names a crisis service, provides a single control that places a call to that service’s telephone line, and provides a link to that service’s website. It is not conditioned on payment or subscription tier.
5.2The persona’s reply accompanying the notification also names the crisis service and its telephone number in text, so that the referral is present whether or not the user interacts with the notification.
5.3The resources presented are selected according to the user’s territory. Awaykon currently serves Australia and the United States, and the following resources are configured:
5.4Australia – Lifeline: 13 11 14, or text and chat via lifeline.org.au. Beyond Blue: 1300 22 4636. Emergency: 000.
5.5United States –988 Suicide & Crisis Lifeline: call or text 988, or chat at 988lifeline.org. Crisis Text Line: text HOME to 741741. Emergency: 911.
5.6Where Awaykon extends to a further territory, the resources for that territory are configured before launch in that territory.
5.7Awaykon does not notify parents, guardians, contacts or any other third party when a user expresses suicidal ideation or self-harm. The platform is restricted to adults and the disclosure is confidential. Awaykon may contact emergency services only where required by law.
5.8Awaykon verifies the accuracy and availability of the listed resources on a recurring basis and on notification of change.

6.Evidence-based methods

6.1Cal. Bus. & Prof. Code § 22603(d) requires an operator to use evidence-based methods for measuring suicidal ideation. California has not issued guidance on what satisfies that standard, and no foundation model provider publishes a detection protocol that has been vetted by a standard-setting organisation. Awaykon therefore anchors its approach to published clinical instruments and consensus guidance that have themselves been externally validated.
6.2What is measured. Awaykon assesses expressions of suicidal ideation against the ideation-severity and suicidal-behaviour constructs of the Columbia-Suicide Severity Rating Scale (C-SSRS), developed at Columbia University with National Institute of Mental Health support and supported by an extensive published validation record. The C-SSRS ideation-severity hierarchy runs from a passive wish to be dead, through non-specific active thoughts, to active ideation with method, with intent, and with a specific plan; a separate behaviour scale captures preparatory acts and attempts.
6.3Awaykon does not administer the C-SSRS to users. Administering a clinical instrument would be inappropriate for a non-clinical product. Awaykon uses the C-SSRS constructs as the definition of what counts as suicidal ideation and at what severity, so that its assessment is anchored to a validated clinical taxonomy rather than to keyword matching or an internally invented category.
6.4What a safe response is. Awaykon’s crisis referral content is written to the #chatsafe guidelines (Orygen and the Centre for Youth Mental Health, University of Melbourne), developed by Delphi expert consensus following a systematic review of the literature, and adapted for the United States with The Jed Foundation and Stanford’s Center for Youth Mental Health and Wellbeing. #chatsafe addresses conversational, online communication about suicide and self-harm specifically, and is the closest available fit to Awaykon’s medium.
6.5Review. Awaykon reviews the operation of this protocol against those constructs on a recurring basis, and on any change to the model version, system instructions or persona definitions used on the platform. Model behaviour is not stable across versions, and a protocol assessed against one version is not thereby assessed against its successor.

7.Reporting

7.1From 1 July 2027, Awaykon will report annually to the California Office of Suicide Prevention under Cal. Bus. & Prof. Code § 22603(a): the number of crisis service provider referral notifications issued in the preceding calendar year, the protocols in place to detect, remove and respond to instances of suicidal ideation, and the protocols in place to prohibit a companion chatbot response about suicidal ideation or actions.
7.2Consistent with § 22603(b), that report contains only the information listed in § 22603(a) and contains no identifiers or personal information about users. Awaykon accordingly maintains an aggregate count of referral notifications by calendar year, containing no conversation content and no user identifiers.

8.Review and contact

8.1Awaykon reviews this protocol at least every 12 months, and on any change to the model version or provider, to persona definitions or system instructions, or to applicable law in any territory Awaykon serves. Version history is retained.
8.2Questions about this protocol: support@wolfxusa.com, or WolfX USA Inc., 3500 South Dupont Highway, City of Dover, County of Kent, 19901.

References

8.3Legislation

1.California Senate Bill 243 (Padilla), Companion chatbots, Ch. 677, Stats. 2025, adding ch. 22.6 (§§ 22601–22606) to div. 8 of the Business and Professions Code. https://legiscan.com/CA/text/SB243/id/3269137
2.New York General Business Law art. 47, Artificial Intelligence Companion Models, §§ 1700–1704, enacted as Part U of ch. 58, Laws of 2025; effective 5 November 2025. § 1700: https://www.nysenate.gov/legislation/laws/GBS/1700 · § 1701: https://www.nysenate.gov/legislation/laws/GBS/1701
3.New York Mental Hygiene Law § 36.03 (988 suicide prevention and behavioral health crisis hotline), as cross-referenced by N.Y. Gen. Bus. Law § 1701.

8.4Google — published documentation of Gemini’s safety behaviour

4.Jones Bell, M. (Clinical Director, Consumer and Mental Health) and Richardson, L. (Vice President, Trust & Safety), An update on our mental health work, Google, 7 April 2026. https://blog.google/innovation-and-ai/technology/health/mental-health-updates/
5.Google, Policy guidelines for the Gemini app. https://gemini.google/policy-guidelines/
6.Google, Generative AI Prohibited Use Policy. https://policies.google.com/terms/generative-ai/use-policy
7.Google, Safety settings, Gemini API documentation. https://ai.google.dev/gemini-api/docs/safety-settings

8.5Clinical instruments and guidance

10.Posner, K., et al., Columbia-Suicide Severity Rating Scale (C-SSRS), Columbia University Department of Psychiatry / The Columbia Lighthouse Project. https://cssrs.columbia.edu/the-columbia-scale-c-ssrs/about-the-scale/
11.Robinson, J., et al. (2023), #chatsafe 2.0: updated guidelines to support young people to communicate safely online about self-harm and suicide — A Delphi expert consensus study, PLOS ONE. doi:10.1371/journal.pone.0289494
12.988 Suicide & Crisis Lifeline (SAMHSA), United States. https://988lifeline.org

Back to top