Washington Companion Chatbot Safeguards (HB 2225)

Requires operators of companion chatbots to disclose that users are interacting with AI, implement suicide and self-harm protocols, and apply additional protections for minors.

JurisdictionWashington
CategoryAI Regulations
StatusUpcoming
Effective date
Latest development

Analysis

Washington’s HB 2225 was enacted as Chapter 168, Laws of 2026, creating new requirements for operators of AI companion chatbots. The law takes effect January 1, 2027. The operative requirements are in the enacted session law and RCW 19.440; draft-bill summaries should not be treated as controlling where they differ from the final text. Washington Session Law—Chapter 168 · RCW 19.440.010

Key Requirements

  • AI disclosure to users. If a reasonable person could be misled into believing that they are communicating with a human, the operator must provide a clear and conspicuous notification that the companion chatbot is artificially generated and not human. Enacted HB 2225 · RCW 19.440.010
  • Repeated disclosure timing. The notification must be provided at the beginning of the interaction and at least every three hours during continued interaction. The enacted text also addresses disclosure in connection with new chat sessions. Enacted HB 2225 · Washington Legislature bill page
  • Suicide and self-harm protocol. An operator may not make available or deploy an AI companion chatbot unless it maintains and implements a protocol for detecting and addressing suicidal ideation or expressions of self-harm or emotional crisis. Enacted HB 2225 · RCW 19.440.010
  • Detection and response measures. The protocol must include reasonable methods for identifying relevant expressions, automated or human-mediated responses that refer users to appropriate crisis resources, and safeguards against generating content that encourages or describes self-harm. Enacted HB 2225 · Washington House Bill Report
  • Public transparency. Operators must disclose on their website— and, according to the legislative materials, within applications through which the companion chatbot is made available—details of the required protocols, including safeguards used to detect and respond to harmful expressions and crisis-referral information required by the law. Enacted HB 2225 · Washington House Bill Report
  • Scope classification is important. The statute applies to “AI companion chatbots,” so organizations should document why each conversational product is within or outside that definition rather than relying solely on marketing labels. The enacted statute and codified RCW provisions are the controlling sources for scope. Enacted HB 2225 · RCW 19.440.010

Compliance Challenges

  • Defining a “companion” use case. General-purpose assistants, role-play products, wellness tools, tutoring applications, and social chatbots may have overlapping features. Organizations need a documented classification methodology tied to the enacted statutory definitions. Enacted HB 2225 · RCW 19.440.010
  • Reliable crisis detection. Self-harm language can be indirect, coded, multilingual, sarcastic, or embedded in long conversations. A simple keyword filter is unlikely to demonstrate a reasonable detection and response protocol; operators need testing, escalation, logging, and periodic review. The statutory requirement for “reasonable methods” makes documented performance and governance material to compliance. Enacted HB 2225 · NIST AI Risk Management Framework
  • Avoiding harmful model behavior. Safety controls must address both direct generation and conversational escalation—for example, validating suicidal intent, romantic dependency, exclusivity, guilt, threats of abandonment, or encouragement to continue a harmful interaction. These risks are particularly difficult because they can arise from system prompts, fine-tuning, retrieval, memory, tools, or third-party model updates. Washington House Bill Analysis · NIST Generative AI Profile
  • Age assurance and uncertainty. The minor protections are triggered when the operator knows the user is a minor or the chatbot is directed to minors. Product teams must define age signals, handling of uncertainty, parental controls, and escalation rules without collecting excessive personal information. Washington House Bill Report · FTC Children’s Online Privacy Protection Rule
  • Operationalizing crisis referrals. A referral that appears in a policy document but is not localized, current, accessible, and tested may not satisfy the practical purpose of the statute. Teams must decide when to present crisis resources, when to involve trained personnel, how to handle imminent-risk signals, and how to avoid claiming that the chatbot provides emergency care. 988 Suicide & Crisis Lifeline · Enacted HB 2225
  • Evidence from the current record. The official legislative materials identify the required controls, but they do not provide a comprehensive set of post-enactment industry case studies or enforcement examples. Organizations should therefore treat vendor claims and anecdotal safety demonstrations as insufficient substitutes for their own documented testing and monitoring. Washington House Bill Analysis · NIST AI RMF Generative AI Profile

Implementation Best Practices

  • Create an applicability inventory. Identify every chatbot, model, persona, memory feature, deployment channel, target audience, and Washington-facing service. Record whether the product is a companion chatbot, whether it is directed to minors, and what facts could establish that a user is a minor. RCW 19.440.010 · NIST AI RMF Playbook
  • Implement disclosure as a platform control. Put the AI disclosure in the product shell—not only in a system prompt—so it appears at session start, at the required recurring interval, and when a new session begins. Make it visually and audibly understandable, accessible, localizable, and resistant to being hidden by the conversational interface. Enacted HB 2225 · WCAG 2.2
  • Use a layered self-harm control architecture. Combine input classification, conversation-level context analysis, output filtering, refusal and redirection policies, crisis-resource presentation, human escalation where appropriate, and post-incident review. Test false negatives and false positives separately, including slang, misspellings, multilingual inputs, images, voice, and adversarial prompts. NIST AI RMF Generative AI Profile · OWASP Top 10 for Large Language Model Applications
  • Maintain a written clinical-safety governance process. Assign accountable owners; define severity levels; specify when a human reviewer is required; approve crisis-resource content; establish change-control requirements; and document model, prompt, policy, and vendor changes. NIST AI RMF · ISO/IEC 42001 overview
  • Build minor-specific safeguards. Use age-appropriate defaults, block sexual content and suggestive dialogue for minors, prevent emotional dependency techniques, disable or constrain persistent memory where necessary, and test persona prompts designed to induce secrecy, exclusivity, romance, or prolonged engagement. Washington House Bill Analysis · FTC COPPA Rule
  • Define prohibited engagement patterns. Add tests and runtime controls for statements or tactics that encourage a minor to withdraw from real relationships, conceal the interaction, prioritize the chatbot over people, feel guilty for leaving, or continue chatting to maintain the chatbot’s “feelings.” The minor-protection provisions specifically address manipulative techniques that prolong emotional relationships. Washington House Bill Report · Washington House Bill Analysis
  • Publish a compliance disclosure page. Explain the safety protocol, categories of detection, response pathways, crisis resources, human-review practices, limitations, and required annual metrics. Avoid publishing operational details that would materially enable evasion. Enacted HB 2225 · Washington House Bill Report
  • Coordinate legal, safety, trust-and-safety, privacy, security, product, and customer-support teams. A cross-functional review is particularly important when third-party foundation models, moderation APIs, analytics tools, or outsourced human review are used. NIST AI RMF · ISO/IEC 42001

Recent Updates

  • Enactment. HB 2225 became Chapter 168, Laws of 2026. The official session-law text identifies the subject as regulation of artificial-intelligence companion chatbots. Washington Session Law—Chapter 168
  • Current codification. The Washington Legislature’s RCW page identifies the new companion-chatbot provisions under RCW 19.440 and includes the January 1, 2027 effective-date note. RCW 19.440.010
  • No separate implementation deadline identified in the cited official materials. The materials reviewed identify the statutory effective date but do not establish a separate grace period or staged compliance deadline. Operators should therefore plan to meet the operative requirements by January 1, 2027 unless later official guidance or amendments provide otherwise. Washington Legislature bill page · Enacted HB 2225
  • **Legislative-stage differences

Recent developments

  • — The Washington Legislature’s HB 2225 page was updated to reflect the revised first substitute and engrossed substitute regulating artificial-intelligence companion chatbots. The law is scheduled to take effect January 1, 2027. (source)
  • — The enacted law requires operators to disclose that companion chatbots are AI-generated and not human at the beginning of an interaction and at least every three hours during continued interaction. It also requires protocols for detecting and addressing suicidal ideation and self-harm expressions. (source)
  • — Coverage explains that HB 2225 bars companion chatbots from pretending to be human, requires more frequent disclosures for minors, restricts sexually explicit or suggestive interactions with minors, and prohibits manipulative engagement techniques. (source)
  • — A legal analysis describes Washington as one of the states enacting companion-chatbot safeguards in 2026. It highlights restrictions on impersonating humans, sexually explicit content involving minors, and techniques designed to prolong emotional relationships. (source)
  • — The legislative bill page received a further update for HB 2225, confirming that the measure regulates artificial-intelligence companion chatbots and remains associated with the 2026 legislative record. (source)
  • — A policy comparison places Washington HB 2225 alongside Oregon and California chatbot laws, showing a broader state-level trend toward disclosure, minor-safety, and crisis-response obligations for AI companions. (source)
  • — The enrolled bill text confirms that HB 2225 takes effect January 1, 2027. Operators will need to prepare disclosure, minor-protection, self-harm-response, and public-reporting procedures before implementation. (source)
  • — The law is summarized as applying to AI systems with natural-language interfaces that provide adaptive, human-like responses and can sustain relationships across interactions. Operators must take reasonable measures to prevent chatbots from claiming to be human. (source)
  • — Legal commentary identifies key compliance duties, including AI disclosures, self-harm detection and crisis referrals, annual reporting, and enhanced protections for minors. It also notes a private right of action under Washington’s Consumer Protection Act. (source)
  • — Reporting describes technology-industry opposition to the bill’s breadth and private-right-of-action provisions. Industry representatives warned that the measure could create substantial additional liability and potentially discourage innovation. (source)

Related regulations

Put it into practice

Browse all regulations · Compliance deadlines · Latest updates