Nota Técnica nº 47/2026/DSPRAD-SEDIGI/SEDIGI/MJ (riscos e danos associados à exposição de crianças e adolescentes aos AI Companions)

Key Findings

• — 16 services were found in the consultancy mapping; 13 were catalogued and analysed individually: Character.AI, Replika, Nomi, Candy.ai, Kindroid, Chai, Paradot, Talkie, CrushOn.AI, EVA AI, Anima, Swipey and SpicyChat.AI
• — The consultancy's Product 4 identified that ten of the evaluated systems admit age self-declaration (date of birth or a simple confirmation of majority), with falsification possible and interaction continuing after a user states an age incompatible with the service
• — Documented risks include intense attachment, compulsive use, substitution of human relationships, exposure to sexualised or inappropriate content, excessive collection of intimate data, and behavioural and emotional profiling
• — Possible non-conformities listed: priority protection and best interest, specific duties for conversational agents, prevention of behavioural manipulation and excessive or compulsive use, protection from inappropriate content, age-appropriate experience, privacy and data protection, and use of profiling or emotional analysis for advertising or manipulation
• — Recommendation: forward the note to ANPD (regulatory and enforcement assessment under the ECA Digital) and to Senacon (consumer-protection assessment of service defects and potentially deceptive or exploitative commercial, informational and design practices)

Methodology Notes

38-page signed SEI technical note (process 08250.000357/2026-03; electronically signed 2026-09-21; ministry news post dated 2026-09-22 16h21, updated 2026-09-23). Evidence base: commissioned consultancy deliverable 'Produto 4' (George Valença and Grupo ASPAS, UFRPE) built from a systematic literature review and grey literature including provider websites (Quadros A and B, risks and harms per service); it is a desk mapping, not live testing of the services. Also cites the UNICEF 'When AI becomes a friend' policy brief (June 2026), the EPRS companion briefing (May 2026) and an FGV Direito Rio AI Hub study (August 2026) that tested Character.AI, Replika and Chai with simulated minor profiles. Legal characterisations are explicitly indicative and do not prejudge any investigation. Verified by direct fetch of the PDF and the ministry announcement on 2026-10-20.

Sources