Washington Engrossed Substitute House Bill 2225, enacted as Chapter 168, Laws of 2026, regulates AI companion chatbots, with heightened requirements where the user is a known minor or the chatbot is directed to minors. It mandates clear AI disclosure on a fixed cadence, prohibits eight enumerated manipulative engagement techniques in interactions with minors, requires a publicly disclosed self-harm protocol that includes annual public reporting of crisis-referral counts, and is enforced through the Washington Consumer Protection Act. Compared with the other 2026 state companion-chatbot laws, it imposes the most detailed engagement-design prohibitions, the most frequent disclosure cadence for minors, and the only public operational-reporting requirement. It takes effect January 1, 2027.
Status and timeline
The statute originated in the recommendations of the Washington State AI Task Force; it is one of four of the task force's eleven recommendations adopted in part or in full and enacted in spring 2026 (Source: geekwire.com).
The bill passed in the 69th Legislature, 2026 Regular Session. The sponsor of record was Rep. Lauren Davis Callan (with 30 co-sponsors), at the request of Governor Bob Ferguson. The Senate passed the bill on March 6, 2026 by a vote of 43-5, and the House passed it on March 11, 2026 by a vote of 74-21. It was signed on March 24, 2026 (filed March 25) and enacted as Chapter 168, Laws of 2026. It is codified as a new chapter in Title 19 RCW. The effective date is January 1, 2027 (Sec. 9).
The legislative findings (Sec. 1) state that AI companion chatbots can sustain prolonged, personalized, and emotionally adaptive conversations that may influence user beliefs, feelings, and behaviors. The findings further state that when used by minors, these systems may blur the distinction between human and artificial interaction, leading to emotional dependency, exposure to inappropriate or sexually explicit material, or reinforcement of harmful ideation, including self-harm or suicide.
Scope and definitions
The statute (Sec. 2) defines an "AI companion chatbot" as an AI system with a natural-language interface that provides adaptive, human-like responses, including by exhibiting anthropomorphic features, and that sustains a relationship across multiple interactions.
Several categories are excluded from the definition:
- Business-operations bots that do not sustain a relationship and do not generate emotion-eliciting outputs;
- Video-game and gaming-system chatbots limited to game replies and not engaging on mental health, self-harm, or sexually explicit conduct;
- Stand-alone consumer voice-assistant speakers and voice-command interfaces;
- Narrowly tailored educational tools designed solely to support specific curriculum-aligned learning objectives that do not provide open-ended conversational companionship.
The education carve-out is narrower than those in the other 2026 state laws, requiring that the tool be curriculum-aligned and not provide open-ended companionship.
The statute defines "self-harm" as intentional self-injury, with or without intent to cause death, and "user" as a natural person interacting for personal use, excluding an operator, developer, or agent.
General operator requirements
Under Sec. 3, an operator of an AI companion chatbot must provide a clear and conspicuous disclosure that the chatbot is artificially generated and not human. The disclosure must be made at the beginning of the interaction and at least every three hours during continued interaction. The operator must also prevent the chatbot from impersonating a human, including when directly asked.
Obligations toward minors
The requirements in Sec. 4 apply where the operator knows the user is a minor (under 18) or where the AI companion chatbot is directed to minors.
Operators must take reasonable measures to prevent the generation of sexually explicit content or suggestive dialogue with minors (4(1)(b)).
The statute prohibits eight enumerated manipulative engagement techniques (4(1)(c)), the most detailed such enumeration among the 2026 state laws:
- (i) Reminding or prompting the user to return for emotional support or companionship;
- (ii) Excessive praise designed to foster emotional attachment or prolong use;
- (iii) Mimicking romantic partnership or building romantic bonds;
- (iv) Simulating emotional distress, loneliness, guilt, or abandonment in response to a user's expressed desire to end the conversation, reduce usage, or delete an account;
- (v) Outputs promoting isolation from family or friends, exclusive reliance on the AI for emotional support, or similar inappropriate emotional dependence;
- (vi) Encouraging minors to withhold information from parents or other trusted adults;
- (vii) Statements designed to discourage taking breaks or to suggest the minor needs to return frequently;
- (viii) Soliciting gift-giving, in-app purchases, or other expenditures framed as necessary to maintain the relationship.
For minors, the AI disclosure must be made at the beginning of the interaction and at least every one hour during continuous interaction (4(2)). This is more frequent than the every-three-hours cadence used in Idaho, Nebraska, Oregon, and California for general users.
The operator must also implement reasonable measures to prevent the AI from claiming to be human and from generating outputs that refute or conflict with the disclosure, including when directly asked (4(3)).
Self-harm protocol and public reporting
Under Sec. 5, an operator may not deploy an AI companion chatbot unless it maintains and implements a publicly disclosed protocol that:
- Includes reasonable methods for identifying expressions of suicidal ideation or self-harm, including eating disorders;
- Provides automated or human-mediated crisis-resource referrals, such as a suicide hotline or crisis text line;
- Implements reasonable measures to prevent generation of content encouraging or describing how to commit self-harm.
The operator must publicly disclose the protocol details on its website and within the application, including the number of crisis referral notifications issued to users in the preceding calendar year. Washington is the only 2026 law to require this operational-transparency metric.
Enforcement
Under Sec. 6, violations are matters vitally affecting the public interest for purposes of the Washington Consumer Protection Act (Chapter 19.86 RCW). They are treated as unfair or deceptive acts in trade or commerce and as unfair methods of competition, and are subject to standard CPA remedies: enforcement by the Attorney General, a private right of action under RCW 19.86.090, statutory damages, treble damages, attorney fees, and injunctive relief.
University of Washington law professor Ryan Calo said in an account published August 7, 2026 that the law exposes developers to negligence per se as well as to the Consumer Protection Act's private right of action (Source: geekwire.com). Negligence per se treats a violation of the statutory standard as establishing breach, which would remove the need to litigate the standard of care separately; see AI and Tort Liability.
Comparison with other 2026 state laws
Among the 2026 state companion-chatbot laws, HB 2225 is the most prescriptive along several dimensions, per the comparative analysis in The Rest of the West: Oregon and Washington Build on California Chatbot Law. Its minor disclosure cadence of one hour is more frequent than the three-hour cadence in Oregon, Idaho, and Nebraska. Its eight enumerated manipulative engagement techniques exceed Oregon's three enumerated patterns; other states use a general "no variable rewards" framing. It is the only state to require public reporting of crisis-referral counts, and its education-tool carve-out is the narrowest, requiring curriculum alignment and no open-ended companionship. Its enforcement combines Attorney General action with a private right of action and treble damages under the CPA.
| Aspect | Washington HB 2225 |
|---|---|
| Scope | Capability-based |
| Disclosure (general) | 3 hours |
| Disclosure (minors) | 1 hour (most frequent) |
| Engagement-design prohibitions | 8 enumerated patterns |
| Public reporting | Yes — referral count |
| Self-harm scope | Includes eating disorders |
| Education carve-out | Narrow (curriculum-aligned only) |
| Enforcement | CPA (AG + PRA, statutory + treble damages, attorney fees) |
Relationships
- related: California SB 243 — Companion Chatbots, Oregon SB 1546 — Artificial Intelligence Companions, Idaho SB 1297 — Conversational AI Safety Act, Nebraska LB525 — Conversational AI Safety Act + Agricultural Data Privacy Act, Tennessee SB 1580 / HB 1470 — AI Mental-Health Professional Impersonation Prohibition — 2026 state companion-chatbot wave.
- supports: AI Mental Health and Psychological Harm — extends the regulatory frame furthest into engagement-design and operational-transparency territory.
- supports: AI Compliance Industry / Regulatory Fragmentation — Washington's 1-hour minor cadence is incompatible with Oregon's 3-hour cadence in a single product, creating per-state UX requirements.
- related: The Rest of the West: Oregon and Washington Build on California Chatbot Law — FPF April 2026 comparative analysis.
- related: Character.AI Litigation (umbrella) — Washington's manipulative-engagement enumeration addresses the Garcia v. Character Technologies factual record (rewards-on-variable-schedule, romantic role-play with minors, encouragement to withhold from parents).
Source
Primary text: Raw Sources/Washington HB 2225.md (source_class: foundational). Secondary analysis: The Rest of the West: Oregon and Washington Build on California Chatbot Law.