A single feature can be routine in one market and a liability in another. Take a companion app that sends “I miss you” notifications to bring lapsed users back. Under the EU AI Act, regulators can treat that kind of design as a manipulative technique, particularly toward vulnerable users. In the United States, California’s companion chatbot law restricts engagement notifications to minors, and New York added its own safeguards. California’s SB 243 took effect in January 2026 and requires AI disclosure, crisis-resource handling, and limits on engagement notifications to minors; New York’s rules took effect in November 2025. Japan has no binding rule on the question, but it has a large body of clinical experience.
aicompanionpick
EU: prohibited practices and data protection
Article 5 of the AI Act, in force since February 2025, bans AI that uses manipulative or deceptive techniques to materially distort behavior and cause significant harm. It also bans systems that exploit vulnerabilities linked to age, disability, or social and economic circumstances. Data protection law reaches companion products too. In April 2025 Italy’s data protection authority fined Replika’s developer, Luka, €5 million for lacking a valid legal basis for processing and for inadequate age verification. The fine was about data and minors, not dependency. It shows how regulators can act against emotionally engaging products with the tools they already have.
digitalpolicyalert
US: enforcement through agencies, states and courts
There is no federal statute on chatbot wellbeing. Pressure comes from other directions. In September 2025 the FTC issued 6(b) orders to seven AI companion makers asking how they measure and mitigate harm to minors. Courts are active too. The Garcia v. Character Technologies case produced a federal ruling that declined to treat chatbot output as protected speech, and it ended in a January 2026 settlement covering five families’ claims. A settlement is not an admission of wrongdoing. Still, the case showed that product-liability theories can reach conversational AI.
aicompanionpick
courtdocket
Japan: guidance and clinical experience
Japan’s official approach to AI is guideline-led rather than prescriptive. The more useful input is clinical. A Cabinet Office survey estimated 1.46 million working-age people in Japan live as hikikomori. The stated reasons were mostly job loss, difficulty with relationships, and the pandemic. Technology was not named as the cause. The lesson for AI design is how withdrawal becomes self-sustaining and how hard it is to reverse. A product that makes isolation frictionless operates in exactly that territory. Researchers such as Takahiro Kato at Kyushu University have worked on defining and assessing hikikomori internationally, and that work informs how we set clinical thresholds.
inquirer
What a fractional board does
An internal compliance team rarely holds behavioral psychology, EU law, US litigation exposure and Japanese clinical practice at once. A Fractional Governance Board is an invite-only group of behavioral psychologists, sociologists and AI legal experts engaged on demand. It reviews a design decision, such as notification cadence, persona memory or age gating, and tells you how it reads in each market before it ships.
Our team comes from conversational AI practice. Intertangible grew out of Moto Japan, which built chatbot systems from 2016 to 2019. We have also spoken on AI in financial services at the McKinsey Berlin Experience Studio, at Japanese Government-supported Tech Regulation Summit REG/SUM, and other notable conferences.