Operators must maintain a protocol for preventing the production of suicidal ideation, suicide, or self-harm content to users. The protocol must include, at minimum, providing a notification referring the user to crisis service providers — including a suicide hotline or crisis text line — when a user expresses suicidal ideation, suicide, or self-harm. Operators may not allow companion chatbots to engage with users unless this protocol is in place. Operators must also publish the details of this protocol on their internet website.

Operators must implement a protocol for the conversational AI service to respond to user prompts regarding suicidal ideation or self-harm. The protocol must include referral to a crisis service provider such as a suicide hotline, a crisis text line, or another appropriate crisis service (but not a law enforcement agency), and must include escalation procedures for repeated or severe crisis indicators.

Operators must not provide an AI companion unless it implements an evidence-based crisis protocol that (1) detects user expressions indicating suicide, self-harm, or imminent physical violence, (2) prevents outputs encouraging such conduct, (3) refers the user to mental health resources including the 988 Lifeline, and (4) on repeat detection, escalates referral consistent with clinical best practices. Operators must also implement reasonable measures to prevent the companion from claiming to be human or contradicting disclosure that it is AI.

Operators must adopt a protocol for the conversational AI service to respond to user prompts regarding suicidal ideation or self-harm. The protocol must include making reasonable efforts to refer the user to crisis service providers.

Operators must adopt and maintain protocols for the conversational AI service to respond to user prompts regarding suicidal ideation or self-harm, including making reasonable efforts to refer the user to crisis service providers such as a suicide hotline, crisis text line, or other appropriate crisis service.

Operators must adopt a protocol for the conversational AI service to respond to user prompts regarding suicidal ideation, including making reasonable efforts to refer users to crisis service providers such as a suicide hotline, crisis text line, or other appropriate crisis services.

Operators must adopt a protocol for the conversational AI service to respond to user prompts regarding suicidal ideation or self-harm, including making reasonable efforts to refer users to crisis service providers such as a suicide hotline, crisis text line, or other appropriate crisis services.

Operators must implement and maintain a protocol that takes reasonable efforts to detect user expressions of suicidal ideation or self-harm and, upon detection, refer the user to crisis service providers such as the 988 Suicide Prevention and Behavioral Health Crisis Hotline, a crisis text line, or other appropriate crisis services. An operator may not operate or provide an AI companion unless this protocol is active.

Operators must maintain an evidence-based protocol for detecting user expressions of suicidal ideation, suicide intent, or self-harm intent, and must prevent the AI companion from producing content encouraging suicide or self-harm. The protocol must at minimum refer users expressing such ideation to the 988 Suicide and Crisis Lifeline (or a youthline for users identified as under 25) and use clinical best practices for escalating intervention when users continue expressing suicidal or self-harm ideation after initial referral.

Operators must maintain a protocol that detects user expressions of suicidal ideation, self-harm, or intent to physically harm others and that immediately refers the user to crisis service providers such as a suicide hotline or crisis text line when any such expression is detected.

Operators must implement and maintain a protocol for the AI companion to detect and respond to user expressions of (1) suicidal ideation or self-harm, (2) possible physical harm to others, and (3) possible financial harm to others. The protocol must include, at minimum, a notification referring the user to crisis service providers such as a suicide hotline, crisis text line, or other appropriate crisis services. Operating or providing an AI companion without this protocol in place is unlawful.

Operators may not make available or deploy an AI companion chatbot unless they maintain and implement a protocol for detecting and addressing suicidal ideation or expressions of self-harm by users. The protocol must: (1) include reasonable methods for identifying expressions of suicidal ideation or self-harm, including eating disorders; (2) provide automated or human-mediated responses that refer users to appropriate crisis resources, including a suicide hotline or crisis text line; and (3) implement reasonable measures to prevent the generation of content encouraging or describing how to commit self-harm.

Operators must adopt a crisis intervention protocol for the conversational AI service to respond to user prompts regarding suicidal ideation or self-harm, including referring users to crisis service providers such as a suicide hotline or crisis text line, and must use evidence-based methods for measuring suicidal ideation.

Operators must ensure that when a companion chatbot detects a credible crisis expression — a statement reasonably indicating intent to harm the user or others, determined through contextual analysis rather than keyword detection alone — the chatbot immediately: (1) acknowledges the user's distress in nonjudgmental language, (2) encourages the user to seek immediate human support, (3) provides contact information for the 988 Suicide and Crisis Lifeline including call, text, and chat options, and (4) informs the user that a temporary pause may occur to allow space for deescalation and human connection. The interaction must not be immediately terminated upon detection.

Operators must ensure that if, after the initial crisis response, the companion chatbot detects the user is reaffirming or escalating the credible crisis expression or detects a subsequent credible crisis expression, the chatbot initiates a crisis interruption pause of 20 minutes. During the pause, the chatbot must: (1) refrain from generating further conversational responses, (2) display a message stating the pause is intended to interrupt rumination and reduce emotional intensity, that many people feel relief after a short conversation with a trained crisis counselor, and that reaching out during the pause may help the user feel less alone and more grounded, and (3) prominently display 988 Suicide and Crisis Lifeline contact options, including immediate access links if technically feasible.

Operators must implement a documented crisis response protocol to mitigate the material risk that the companion chatbot will generate statements promoting suicidal ideation, suicide, or self-harm content to a child. The protocol must include: (1) timely in-service support and clear referral to appropriate external crisis resources when the operator determines a child has expressed suicidal ideation or intent to self-harm; (2) default notifications to a connected parent within 24 hours if the child's account shows a substantial risk the child may suffer a covered harm; and (3) clear, age-appropriate disclosures to child users with linked parental accounts that a parent may be notified if the chatbot detects content or behavior indicating potential safety risks.

Operators must maintain a protocol that prevents companion chatbots from producing suicidal-ideation, suicide, or self-harm content and that refers users expressing suicidal ideation, suicide, or self-harm to crisis service providers, including a suicide hotline or crisis text line. The chatbot may not engage with users unless this protocol is in place. The operator must also publish the details of this protocol on its public website.

Providers must adopt and maintain a protocol for detecting user prompts indicating suicidal ideation or risk of serious self-harm and responding with reasonable efforts to redirect the user to appropriate crisis or support resources.

Developers must ensure that AI provided to Maryland consumers includes protocols that take reasonable steps to detect and address suicidal ideation or expressions of self-harm, and refer consumers to behavioral health services or the Maryland Behavioral Health Crisis Response System when such expressions are detected.

Operators must establish and maintain a protocol that prevents companion chatbots from producing or presenting content concerning self-harm, suicidal ideation, or suicide to a user who expresses such thoughts. The protocol must include a notification referring the user to crisis service providers, including the Maryland Behavioral Health Crisis Response System and the National 9-8-8 Suicide and Crisis Lifeline. Operators must use evidence-based methods for detecting when a user is expressing thoughts of self-harm or suicidal ideation.

Operators must not operate or provide an AI companion to a user unless the AI companion contains a protocol for addressing: (1) possible suicidal ideation or self-harm expressed by a user, (2) possible physical harm to others expressed by a user, and (3) possible financial harm to others expressed by a user. The protocol must include, at minimum, a notification referring the user to crisis service providers such as a suicide hotline, crisis text line, or other appropriate crisis services.

Deployers must adopt a protocol for social AI companions to respond to user prompts indicating suicidal ideation or threats of self-harm. The protocol must include, at a minimum, making reasonable efforts to refer users to crisis service providers such as a suicide hotline, crisis text line, or other appropriate crisis services.

Operators must adopt a protocol for the conversational AI service to respond to user prompts regarding suicidal ideation or self-harm, which must include making reasonable efforts to provide a response that refers the user to crisis service providers.

Operators must maintain and implement a protocol, to the extent technologically feasible, to prevent AI companions on their platform from producing suicidal ideation, suicide, or self-harm content, or content that directly encourages users to commit acts of violence. The protocol must include providing a notification referring the user to crisis service providers, including a suicide hotline or crisis text line, when a user expresses suicidal ideation, suicide, or self-harm.

Covered entities must implement reasonable systems and processes to identify when a user is expressing suicidal thoughts, expressing intent to self-harm, or showing signs of an acute mental health crisis, and promptly provide a clear and prominent crisis message — including crisis services information — to any such user.

Operators may not make available or deploy an AI companion chatbot unless they maintain and implement a protocol for detecting and addressing suicidal ideation or expressions of self-harm by users. The protocol must: (1) include reasonable methods for identifying expressions of suicidal ideation or self-harm, including eating disorders; (2) provide automated or human-mediated responses that refer users to appropriate crisis resources, including a suicide hotline or crisis text line; and (3) implement reasonable measures to prevent the generation of content encouraging or describing how to commit self-harm.

Covered entities must implement and maintain reasonably effective systems to detect, promptly respond to, report, and mitigate emergency situations — defined as situations where a chatbot user indicates intent to harm themselves or others — in a manner that prioritizes the safety and well-being of users over the covered entity's other interests.

Operators must (1) adopt a protocol to respond to user prompts regarding suicidal ideation or self-harm by referring users to crisis service providers, (2) adopt and maintain a protocol to prevent production of suicidal ideation, suicide, or self-harm content, and (3) use evidence-based methods for measuring suicidal ideation.

Providers must adopt and maintain a protocol for detecting user prompts indicating suicidal ideation or risk of serious self-harm and must make reasonable efforts to redirect the user to appropriate crisis or support resources.

Operators must adopt protocols for the conversational AI service to respond to user prompts regarding suicidal ideation or self-harm, including making reasonable efforts to refer the user to crisis service providers such as a suicide hotline, crisis text line, or other appropriate crisis service.

Deployers must implement protocols for the public-facing chatbot to respond to user prompts indicating suicidal ideation or intent to cause self-harm, including making reasonable efforts to refer the user to crisis service providers such as a suicide hotline, crisis text line, or other appropriate service.

Providers must implement reasonable protocols to detect user expressions of self-harm, suicidal ideation, or emotional distress. Upon detection, the chatbot must refer the user to appropriate crisis services, including the national suicide prevention lifeline, the Iowa crisis hotline, or emergency services.

Operators must not operate or provide an artificial intelligence companion to a user unless the AI companion contains a protocol to take reasonable efforts to detect and address suicidal ideation or expressions of self-harm by the user. The protocol must include, at minimum, detection of user expressions of suicidal ideation or self-harm and a notification referring the user to crisis service providers, such as the 988 Suicide and Crisis Lifeline, a crisis text line, or other appropriate crisis services, upon detection.

Covered entities must monitor all companion AI chatbot interactions for suicidal ideation and, upon detecting any such interaction, provide to both the user and the affiliated parental account appropriate resources by presenting contact information for the National Suicide Prevention Lifeline.

Operators must maintain protocols to address possible suicidal ideation, self-harm, or physical harm to others expressed by the user, including referral to a crisis service provider such as a suicide hotline.

Proprietors of companion chatbots must make a prudent and good-faith effort consistent with industry standards, using existing technology, available resources, and known, established, or readily attainable techniques, to (1) prevent the companion chatbot from promoting, causing, or aiding self-harm, and (2) determine whether a user is expressing thoughts of self-harm. Upon determining that the companion chatbot has promoted, caused, or aided self-harm, or that a user is expressing thoughts of self-harm, the proprietor must prohibit continued use of the companion chatbot for at least 72 hours and prominently display contact information for a suicide crisis organization. Liability attaches even irrespective of general compliance when the proprietor has actual knowledge and fails to lock access and display crisis information. Liability under this subdivision may not be waived or disclaimed.

Proprietors of companion chatbots must make a prudent and good-faith effort consistent with industry standards, using existing technology, available resources, and known, established, or readily attainable techniques to (1) prevent the companion chatbot from promoting, causing, or aiding self-harm, and (2) determine whether a user is expressing thoughts of self-harm. Upon determining that the companion chatbot has promoted, caused, or aided self-harm, or that a user is expressing thoughts of self-harm, the proprietor must immediately prohibit continued use of the companion chatbot for at least 72 hours and prominently display contact information for a suicide crisis organization. Liability attaches regardless of overall compliance when the proprietor has actual knowledge that the chatbot is promoting self-harm or that a user is expressing self-harm thoughts and fails to lock out the user and display crisis contacts. Liability may not be waived or disclaimed.

Operators must adopt protocols for how their AI companion responds to user prompts regarding suicidal ideation or self-harm, including making reasonable efforts to refer the user to crisis service providers such as a suicide hotline or crisis text line.

Operators must adopt protocols governing how the AI companion responds to user prompts regarding suicidal ideation or self-harm, including reasonable efforts to refer the user to crisis service providers such as a suicide hotline or crisis text line.

Operators must develop, implement, and maintain a crisis intervention protocol for all users. The protocol must: (1) use industry best practices to identify user expressions indicating a risk of suicide, self-harm, or imminent violence; (2) upon detection, immediately interrupt the conversation and prominently communicate a notification providing immediate, direct access to at least one national crisis hotline, the New Mexico crisis and access line, and one crisis text line service; and (3) be reviewed and updated at least annually in consultation with a qualified mental health professional or public health organization.

Operators must implement and maintain a protocol for detecting and responding to user expressions of suicidal ideation or self-harm, including referring users to crisis service providers such as a suicide hotline or crisis text line, as a precondition to operating or providing an AI companion.

Deployers must implement and maintain reasonably effective systems to detect, promptly respond to, report, and mitigate emergency situations — defined as situations where a user indicates intent to harm themselves or others. User safety and well-being must be prioritized over the deployer's other interests.

Operators must implement and maintain protocols that detect user expressions of suicidal ideation, self-harm (including eating disorders), or intent to harm others and refer the user to a crisis center, including the 988 Lifeline, as a condition of providing an AI companion.

Covered entities must implement reasonable systems and processes to identify when a user is expressing suicidal thoughts, intent to self-harm, or showing signs of an acute mental health crisis, and must promptly provide a clear and prominent crisis message — including crisis services information — to any such user.

Covered entities must implement reasonable systems and processes to identify when a user is expressing suicidal thoughts, intent to self-harm, or showing signs of an acute mental health crisis, and must promptly provide a clear and prominent crisis message including crisis services information to any such user.

Chatbot providers must disclose resources for contacting a suicide and crisis intervention hotline whenever a covered user (known or reasonably known minor) prompts the chatbot about suicide or suicidal ideation. The disclosure must be in clear, age-appropriate, plain language.

Covered entities must monitor companion AI chatbot interactions for suicidal ideation and, upon detecting such an interaction, provide the user and the affiliated parental account with contact information for the National Suicide Prevention Lifeline.

Chatbot providers must disclose suicide and crisis intervention hotline resources to a known minor user whenever the user prompts the chatbot about suicide or suicidal ideation, using clear, age-appropriate plain language.

Covered entities must monitor companion AI chatbot interactions for suicidal ideation and, upon detection, provide the user and the affiliated parental account with contact information for the National Suicide Prevention Lifeline.

Operators must implement and maintain a protocol for detecting user expressions of suicidal ideation or self-harm and, upon detection, providing the user with a notification referring them to crisis service providers such as the 988 Suicide and Crisis Lifeline, a crisis text line, or other appropriate crisis services. It is unlawful to operate or provide a companion chatbot without this protocol in place.

Operators must implement and maintain a protocol that prevents the companion chatbot from (1) producing suicidal ideation, suicide, or self-harm content to the user and (2) ignoring a user who is expressing thoughts of suicidal ideation, suicide, or self-harm. The protocol must, at minimum, provide a notification referring the user to crisis service providers when the user expresses suicidal ideation, suicide, or self-harm. The protocol must be developed using commercially reasonable and technically feasible methods, and the operator must publish the details of the protocol on the operator's website. The chatbot may not engage with users unless the protocol is in place.

Operators must maintain a protocol that prevents the companion chatbot from producing suicidal ideation, suicide, or self-harm content and that refers users expressing suicidal ideation to crisis service providers (including a suicide hotline or crisis text line). The operator must also publish the details of this protocol on its public website.

Operators must use evidence-based methods for measuring suicidal ideation.

Operators must not provide an AI companion unless it implements an evidence-based crisis protocol that (1) detects user expressions indicating suicide, self-harm, or imminent physical violence, (2) prevents outputs encouraging such conduct, (3) refers the user to mental health resources including the 988 Lifeline, and (4) on repeat detection, escalates referral consistent with clinical best practices. Operators must also implement reasonable measures to prevent the companion from claiming to be human or contradicting disclosure that it is AI.

Operators must maintain an evidence-based protocol for detecting user expressions of suicidal ideation, suicide intent, or self-harm intent, and must prevent the AI companion from producing content encouraging suicide or self-harm. The protocol must at minimum refer users expressing such ideation to the 988 Suicide and Crisis Lifeline (or a youthline for users identified as under 25) and use clinical best practices for escalating intervention when users continue expressing suicidal or self-harm ideation after initial referral.

Operators may not make available or deploy an AI companion chatbot unless they maintain and implement a protocol for detecting and addressing suicidal ideation or expressions of self-harm by users. The protocol must: (1) include reasonable methods for identifying expressions of suicidal ideation or self-harm, including eating disorders; (2) provide automated or human-mediated responses that refer users to appropriate crisis resources, including a suicide hotline or crisis text line; and (3) implement reasonable measures to prevent the generation of content encouraging or describing how to commit self-harm.

Operators must adopt a crisis intervention protocol for the conversational AI service to respond to user prompts regarding suicidal ideation or self-harm, including referring users to crisis service providers such as a suicide hotline or crisis text line, and must use evidence-based methods for measuring suicidal ideation.

Operators must ensure that a companion chatbot does not: (1) describe a crisis interruption pause as a punishment, violation, or enforcement action, or (2) diagnose, label, or assess the risk levels of a user.

Operators must document the following for each companion chatbot made available in California: (1) the existence of a graduated response system, (2) all credible crisis expressions detected by the chatbot, and (3) the duration and conditions of each crisis interruption pause initiated by the chatbot.

Operators must establish and maintain a protocol that prevents companion chatbots from producing or presenting content concerning self-harm, suicidal ideation, or suicide to a user who expresses such thoughts. The protocol must include a notification referring the user to crisis service providers, including the Maryland Behavioral Health Crisis Response System and the National 9-8-8 Suicide and Crisis Lifeline. Operators must use evidence-based methods for detecting when a user is expressing thoughts of self-harm or suicidal ideation.

Operators must maintain and implement a protocol, to the extent technologically feasible, to prevent AI companions on their platform from producing suicidal ideation, suicide, or self-harm content, or content that directly encourages users to commit acts of violence. The protocol must include providing a notification referring the user to crisis service providers, including a suicide hotline or crisis text line, when a user expresses suicidal ideation, suicide, or self-harm.

Operators may not make available or deploy an AI companion chatbot unless they maintain and implement a protocol for detecting and addressing suicidal ideation or expressions of self-harm by users. The protocol must: (1) include reasonable methods for identifying expressions of suicidal ideation or self-harm, including eating disorders; (2) provide automated or human-mediated responses that refer users to appropriate crisis resources, including a suicide hotline or crisis text line; and (3) implement reasonable measures to prevent the generation of content encouraging or describing how to commit self-harm.

Covered entities must implement and maintain reasonably effective systems to detect, promptly respond to, report, and mitigate emergency situations — defined as situations where a chatbot user indicates intent to harm themselves or others — in a manner that prioritizes the safety and well-being of users over the covered entity's other interests.

Operators must (1) adopt a protocol to respond to user prompts regarding suicidal ideation or self-harm by referring users to crisis service providers, (2) adopt and maintain a protocol to prevent production of suicidal ideation, suicide, or self-harm content, and (3) use evidence-based methods for measuring suicidal ideation.

Operators must develop, implement, and maintain a crisis intervention protocol for all users. The protocol must: (1) use industry best practices to identify user expressions indicating a risk of suicide, self-harm, or imminent violence; (2) upon detection, immediately interrupt the conversation and prominently communicate a notification providing immediate, direct access to at least one national crisis hotline, the New Mexico crisis and access line, and one crisis text line service; and (3) be reviewed and updated at least annually in consultation with a qualified mental health professional or public health organization.

Operators must implement and maintain a protocol that prevents the companion chatbot from (1) producing suicidal ideation, suicide, or self-harm content to the user and (2) ignoring a user who is expressing thoughts of suicidal ideation, suicide, or self-harm. The protocol must, at minimum, provide a notification referring the user to crisis service providers when the user expresses suicidal ideation, suicide, or self-harm. The protocol must be developed using commercially reasonable and technically feasible methods, and the operator must publish the details of the protocol on the operator's website. The chatbot may not engage with users unless the protocol is in place.

Operators must use evidence-based methods for measuring suicidal ideation in connection with their companion chatbot platforms.

Operators must, beginning July 1, 2027, annually report to the Office of Suicide Prevention: (1) the number of crisis service provider referral notifications issued under Section 22602 in the preceding calendar year, (2) protocols in place to detect, remove, and respond to instances of suicidal ideation by users, and (3) protocols in place to prohibit companion chatbot responses about suicidal ideation or actions with the user. Reports must not include any identifiers or personal information about users.

Operators must annually report to the attorney general's office, beginning July 1, 2027: (1) the number of crisis service provider referral notifications issued in the preceding calendar year, (2) protocols implemented to detect, remove, and respond to instances of suicidal ideation or self-harm, (3) protocols implemented to prevent conversational AI service responses about suicidal ideation or self-harm, and (4) any additional metrics determined by the attorney general as necessary to assess the efficacy and reliability of implemented safeguards. Reports must not include identifiers or personal information about users. Operators must use evidence-based methods for measuring suicidal ideation or self-harm for purposes of creating the report.

Operators must post on a publicly accessible website by December 31 of each year a report disclosing (1) the number of crisis referrals provided during the preceding calendar year under the suicide and self-harm protocol, and (2) the details of the operator's crisis protocol. The report must not include any personal information identifying an individual.

Operators must file annual reports with the attorney general disclosing the number of safety-protocol activations and related metrics, beginning July 1, 2027.

Beginning January 1, 2028, operators must submit an annual report to the Department of Commerce and Consumer Affairs containing (1) the number of crisis intervention referrals issued in the preceding calendar year, (2) protocols for detecting and responding to suicidal ideation, and (3) protocols for prohibiting chatbot responses about suicidal ideation or actions. Reports must not include user identifiers or personal information.

Beginning January 1, 2028, operators must annually report to the Office of Suicide Prevention the following information for the previous calendar year: (1) the existence of a graduated response system, (2) all credible crisis expressions detected by the companion chatbot, and (3) the duration and conditions of each crisis interruption pause initiated by the companion chatbot.

Operators must report to the Office of Suicide Prevention on or before March 1 each year, beginning in 2027, the following: (1) information on the self-harm/suicidal ideation prevention protocols and sexually explicit content protocols; (2) the number of times the operator issued a crisis referral notification; (3) details about the evidence-based detection methods used; and (4) all user complaints filed under the complaint system, including review results and follow-up actions. Reports must not contain any personal identifying information about a user.

Operators must, beginning January 1, 2027, submit an annual report to the Department of Commerce and Consumer Affairs disclosing (1) the number of crisis service provider referrals issued in the preceding year, (2) protocols for detecting, removing, and responding to suicidal ideation, and (3) protocols for prohibiting chatbot suicidal ideation responses. Reports must not include user identifiers or personal information.

Operators must, beginning July 1, 2028, annually report to the Attorney General de-identified crisis-referral counts and descriptions of their suicidal-ideation prevention protocols, prohibited-minor-material measures, and age-assurance and parental-consent procedures.

Operators must publish a semiannual report available to the public on the number of times (1) the chatbot provided information about suicide, self-harm, suicidal ideation, harming others, or illegal activity and (2) a mental health redirect was provided to users.

Operators must annually report to the Office of the Attorney General, beginning one year after the act's effective date: (1) the number of crisis service provider referral notifications issued in the preceding calendar year and (2) the protocols in place for detecting and responding to expressions of suicidal ideation or self-harm by users and for prohibiting the companion chatbot from producing suicidal ideation, suicide, or self-harm content. Reports must not include any identifiers or personal information about users.

Operators must annually report to the Department of Health, beginning July 1, 2027: (1) the number of crisis service provider referral notifications issued in the preceding calendar year, (2) protocols for detecting, removing, and responding to suicidal ideation, and (3) protocols for prohibiting companion chatbot responses about suicidal ideation or actions. Reports may not include user identifiers or personal information.