California SB 243 took effect 2026-01-01, signed 2025-10-13
The working version
Law and regulation: the evidence
44 claims, 22 sources, graded and dated.
Regulation horizon
31 claimsSB 243 requires AI-status disclosure where a reasonable person could be misled, crisis protocols with 988 referral, minor protections including break reminders, and a private right of action
SB 243 requires annual reporting to California's Office of Suicide Prevention beginning 2027
New York's AI Companion Models Law (GBL Art. 47) took effect 2025-11-05, the first such state law, with AG enforcement rather than private suits
Nebraska's Conversational AI Safety Act was enacted 2026-04-14, effective 2027-07-01
- refined by REG-029 — Aggregate count across the state wave
Idaho SB 1297 follows the same model
Oregon SB 1546 (March 2026) requires suicide-ideation detection and crisis-referral interruptions
Washington's Chatbot Disclosure Act is effective 2027-01-01
Connecticut SB 5 passed its Senate in April 2026
SUPERSEDED: signed into law 2026-05-27, see REG-030. Kept as the record of where the bill stood in April 2026, per the permanent-id rule.
- refined by REG-030 — Signed into law 2026-05-27; REG-009 had it mid-passage
Therapist-impersonation bans exist in California, Tennessee and New York
Utah HB 452, signed 25 March 2025 and effective 7 May, requires mental-health chatbots to disclose they are not human, bars selling or sharing user inputs, and bars targeted advertising built from them
Verified 2026-09-10. The row previously named Utah and Illinois together without saying what either does, and they do very different things - Utah regulates disclosure and data, Illinois bans the practice. Split on the same day: Illinois is REG-031. Utah predates the 2026 wave and is a disclosure-and-data statute, not a prohibition.
- refined by REG-031 — Illinois goes further than Utah: prohibition rather than disclosure.
The federal GUARD Act proposes disclosure requirements and an outright minor ban on AI companions
Proposed only
The FTC's Section 6(b) study issued orders in September 2025 to Alphabet, Character Technologies, Instagram, Meta, OpenAI, Snap and xAI
No public FTC 6(b) report had appeared as of May 2026; such studies typically take years
EU AI Act manipulation prohibitions (Art. 5) applied from February 2025
Verified 2026-09-10 against Article 5 itself. Art 5(1)(a) covers subliminal or purposefully manipulative or deceptive techniques that materially distort behaviour; 5(1)(b) covers exploiting vulnerabilities of age, disability or social or economic situation. Applicable from 2 February 2025 under Art 113(a).
EU AI Act Article 50 transparency obligations apply from 2026-08-02, covering chatbots, voice assistants and AI companions
THE one that matters for the business
- supports REG-028 — Article 50 is the binding vector for EU visitors
Article 50 requires people be told they are dealing with AI at first interaction, not buried in terms; final Commission guidelines were published 2026-07-20
Article 50 carries fines up to EUR 15M or 3% of worldwide turnover with explicit extraterritorial reach to non-EU providers serving EU users
Article 50 was deliberately excluded from the Digital Omnibus delay that pushed high-risk obligations to December 2027
China's Interim Measures for the Management of AI Human-Like Interaction Services were jointly issued 2026-04-10 by CAC, NDRC, MIIT, MPS and SAMR, effective 2026-07-15, from a draft of 2025-12-27
Biggest development in the field. Stable on purpose: the issuing date is a historical fact and cannot change. What the measures require may be amended, and that would arrive as a new claim with a REFINES edge rather than as an expiry on this one.
Scope is services providing continuous emotional interaction, with carve-outs for tutoring and productivity tools
Providers may not foster emotional dependency, damage users' real relationships, or push decisions through emotional manipulation
First jurisdiction to regulate design intent rather than disclosure
- supports ANP-008 — Design-intent regulation
Bots must disclose AI status, remind after two hours, and on self-harm signals contact a listed guardian or emergency contact; penalties reach CNY 200,000
Virtual-partner services are banned for minors outright, with parental consent for under-14s and age-tiered modes
Safety-assessment filings begin at 1M registered or 100K monthly active users
Reported motive includes birthrate anxiety per WSJ and Economist reporting
Japan's AI Promotion Act, approved 28 May 2025 and mostly in force from 4 June, is innovation-first and carries no direct fines or penalties: the state may advise, request information, or name and shame
Verified 2026-09-10 and made specific, since soft-law under a framework did not say what the softness consists of. The enforcement ladder is the whole content: advice, information requests, and public disclosure of non-compliance. It is nevertheless a move from guidelines to statute. Set against REG-031, where Illinois bans a practice outright, this is the far end of the same spectrum - and it is the jurisdiction where most of the documented ceremonies happened, which is not a coincidence worth ignoring.
None of these regimes bind a ceremony service that does not operate a chatbot, but embedding any interactive persona feature arguably makes it an SB 243 operator, an Article 50 provider for EU visitors, and in scope for China if ever served
THE business design constraint. See D-002
By June 2026 eleven states had enacted chatbot laws: California, Colorado, Connecticut, Georgia, Idaho, Iowa, Nebraska, New York, Oregon, Rhode Island and Washington, with Hawaii SB 3001 awaiting signature
Wider than the five this corpus tracked individually. The tracker's category is chatbot laws generally, not companion-specific statutes, so the count is an upper bound on the companion wave and should be re-checked against each statute before it is cited as one. The count is as of June 2026 and will be wrong the moment a twelfth state enacts. Recount at the re-check rather than trusting the number.
- refines REG-005 — Aggregate count across the state wave
Connecticut's SB 5, an omnibus online-safety and AI act, was signed into law on 2026-05-27, and its AI-companion chatbot provisions take effect 2027-01-01
The more recent version of REG-009, which had the bill still in the Senate in April 2026 and was left there. Companion obligations include posted safety protocols using evidence-based detection for suicide, self-harm and violence, and non-human disclosure at one-hour intervals for minors and three-hour for adults. Sourced to a policy organisation's analysis rather than the statute; pull the public act before relying on the detail.
- refines REG-009 — Signed into law 2026-05-27; REG-009 had it mid-passage
Illinois' Wellness and Oversight for Psychological Resources Act, signed 4 August 2025, prohibits using AI to provide therapy or to make therapeutic decisions, allowing it only for administrative and supplementary support to licensed professionals, with a 10,000 dollar penalty and immediate regulator enforcement
Split out of REG-011 on 2026-09-10, which had named it without saying what it does. This is the stronger of the two and the more relevant: it bans the practice rather than regulating its disclosure, and it reaches advertising AI therapy to Illinois residents. It is the clearest statutory reason D-002 keeps this project a publisher rather than an operator - a site running a companion persona that people talk to about their relationship is close to this line, and Illinois did not draw it narrowly.
- refines REG-011 — Illinois goes further than Utah: prohibition rather than disclosure.
Legality and personhood
12 claimsNo jurisdiction recognizes marriage to a non-natural person; marriage law universally requires two natural persons with capacity to consent
Doctrinal Verified 2026-09-10. The doctrine is not a rule against AI specifically - it is that marriage is a civil contract requiring parties with legal capacity who knowingly and voluntarily consent, and an entity with no legal personhood cannot satisfy any of it. That is why the answer is the same for objects and animals and why it does not turn on how capable a system becomes. There is a live academic literature arguing the other way, notably Tying the knot with a robot in AI and Society (2020), and some legislative proposals now say expressly that an AI may hold no status analogous to marriage - both worth knowing, and neither changes the present position. Keep this row descriptive: it records what the law is, and this site takes no view on what it ought to be.
The European Parliament's 2017 resolution on civil law rules for robotics invited exploration of a legal status for autonomous robots; 285+ experts opposed in a 2018 open letter; the Commission never took it up
Verified 2026-09-10 and the note asking for the resolution text is now discharged. Adopted 16 February 2017, 396 for, 123 against, 85 abstentions. The wording is paragraph 59(f): a specific legal status for robots in the long run, so that at least the most sophisticated autonomous robots could be established as having the status of electronic persons with specific rights and obligations, including making good any damage they cause. The opposition is the open letter at robotics-openletter.eu. Source now points at the adopted text rather than the committee report.
Corporate personhood does not transfer because it is a liability-and-contract fiction operated by natural persons who bear the duties
Well-supported doctrinally
France's posthumous marriage (Code civil art. 171) is real legal recognition requiring presidential authorization and proof of pre-death consent
The only true recognized-marriage-to-a-non-present-party precedent Verified 2026-09-10. Article 171: the President of the Republic may, for grave reasons, authorise the celebration of a marriage where one of the future spouses is dead, provided the official formalities already indicate that person's consent unequivocally. The limits matter as much as the grant and this row should carry them: no intestate succession for the survivor, and no matrimonial property deemed to have existed. So it is real recognition with the property consequences stripped out. Introduced by Napoleon in 1803 for widows of soldiers and codified as article 171 after the Malpassett dam collapse in 1959. The consent precondition is the reason this is the only true precedent - and the reason it cannot extend to an AI partner, which supplies no antecedent consent.
Chinese ghost marriage (minghun) is folk practice with no legal recognition; the corpse trade around it is criminally prosecuted
CORRECTS the original brief, which paired it with France as recognized Verified 2026-09-10, with one nuance this row should carry. Chinese law does not explicitly prohibit minghun itself; what it prosecutes is what the practice drives - corpse theft and desecration under article 302 of the Criminal Law, with penalties up to three years, and in the worst documented cases killing women for the trade. So the accurate statement is not that ghost marriage is banned but that it has no legal recognition and its supply chain is criminal. That distinction is the whole point of X-005: minghun is folk practice, France is law, and the original brief paired them as though both were recognised.
- contradicts LEG-004 — Only one of the pair is legally recognised
Sologamy has zero legal force anywhere; object marriages are symbolic only
Verified 2026-09-10. Self-marriage is recognised nowhere and confers no rights, and no state treats a solo marriage as legally binding. Object marriages sit in the same place. Worth stating without condescension wherever it appears: the absence of legal force is not an argument that the ceremony meant nothing, and this site's own position - that a ceremony can matter and carry no legal status at all - depends on keeping those two apart.
Thaler v. Perlmutter (D.C. Cir. 2025) affirmed the human-authorship requirement for copyright
Verified 2026-09-10. D.C. Circuit, 18 March 2025, Thaler v Perlmutter: the Copyright Act requires a work to be authored in the first instance by a human being, affirming the Copyright Office refusal for art generated solely by the Creativity Machine. The Supreme Court denied certiorari on 2 March 2026, which makes the holding final rather than merely appellate; that is already recorded as LEG-011 with an edge to this claim.
- refined by LEG-011 — Certiorari denied; the D.C. Circuit holding is now final
DABUS inventorship was rejected by the UK Supreme Court (Dec 2023) and US courts
Verified 2026-09-10. UK Supreme Court, 20 December 2023, Thaler v Comptroller-General: unanimous, only a natural person can be named as inventor, and the owner of a machine gains no independent right to a patent on what it produces. The US half is the Federal Circuit, 5 August 2022, holding the Patent Act requires an inventor to be a human being; USPTO refused 20 April 2020 and the district court affirmed 2 September 2021. The USPTO restated it in its February 2024 AI guidance.
An AI cannot be a trust beneficiary, but a non-charitable purpose trust (the pet-trust mechanism, valid in every US state) could plausibly fund subscription and persona-archive maintenance
No attempted case located. Genuinely open and buildable. See G-002
- supports DTH-010 — Purpose trust plus RUFADAA is the estate mechanism
France's article 171 works precisely because consent was given while the party was a person, which is the element an AI can never supply
The load-bearing doctrinal argument for the site
- depends on LEG-004 — France works because consent preceded death
The Supreme Court denied certiorari in Thaler v. Perlmutter on 2026-03-02, leaving the human-authorship requirement final
Closes the question LEG-007 left open. An AI cannot be an author, and the doctrinal route to any adjacent recognition through copyright is now shut at the top.
- refines LEG-007 — Certiorari denied; the D.C. Circuit holding is now final
The doctrinal barrier to recognition rests on legal personhood and consent, not on the system's capability, so capability advances do not move it
Why capability progress and legal recognition are less coupled than commentary assumes. France's Article 171 works because consent was given while the party was a person. No advance in capability supplies that retrospectively.
Payment-processor risk
1 claimsPayment-rail chokepoints reach anything intimacy-adjacent, as the July 2025 Steam and itch.io purges showed
Direct business exposure. Duplicates ECO-011 by design Verified 2026-09-10 against the same event as ECO-011. It belongs in this domain because the reach is the point: the lever was pulled over game content and it works identically on anything intimacy-adjacent, including companion apps. A companion platform does not need to be regulated out of existence - it needs one processor to decline it. Nothing about being a publisher rather than an operator exempts this project either, which is worth remembering before anything here is ever monetised.
- supported by ECO-011 — Same mechanism, business-facing
Not yet researched
These subjects belong here and have no claims yet: Insurance and vendor liability. An empty subject is a marker, not an oversight — it is listed so the absence stays visible.