Sedigi Maps AI Companion Risks for Children and Flags ANPD
Brazil’s National Secretariat for Digital Rights (Sedigi) at the Ministry of Justice published a technical note on September 22, 2026 mapping risks from AI companion apps that simulate friendship or emotional bonds, and forwarded the findings to the National Data Protection Authority (ANPD) and the National Consumer Secretariat (Senacon). Consultancy coordinated by professor George Augusto Valença Santos of UFRPE reviewed thirteen platforms available in Brazil—including Character.AI, Replika, Nomi, Candy.ai, Kindroid, Chai, Paradot, Talkie, Crushon.ai, Eva AI, Swipey, Anima and SpicyChat.AI—citing intense attachment, compulsive use, sexualised content exposure, weak age verification and behavioural profiling of users who may be under eighteen.
Filed under Policy and dated September 27, 2026, this AI4Brazil briefing treats the Sedigi note as Brazilian child-protection AI news distinct from yesterday’s frontier-model access gap story. Secretary Victor Oliveira Fernandes said ANPD and Senacon should assess regulatory and consumer-protection angles under the Digital ECA (Law 15.211/2025) and Decree 12.880/2026, stressing that age self-declaration alone is easily bypassed and expressly barred when services offer unsuitable content for minors.
Why it matters: Brazilian teens already meet companion chatbots in app stores. Transparent agency mapping can limit harm—but only if age gates, data minimisation and human complaint paths become enforceable product requirements.
What it means in practice
Brazilian product, school and counsel leads should inventory whether staff or students use companion apps on managed devices; demand named age-assurance methods beyond self-declaration; assign an owner for ANPD/Senacon response tracking; run time-boxed reviews of emotional-profiling features; and prefer designs that keep humans on escalation when distress signals appear. Connect the note to OpenAI’s TSE compliance plan and Brazil’s delayed AI framework vote as parallel governance tracks.
Caveats come first. A preliminary mapping is not a sanction; Sedigi stressed due process for suppliers; and UNICEF download statistics describe a global market, not adjudicated Brazilian breaches. AI4Brazil therefore presents the companion note as directional policy context until formal inquiries conclude.
What to watch next: whether ANPD opens proceedings; how Senacon frames commercial practices; and whether app stores tighten age gates for Brazil. Readers can continue on the AI4Brazil homepage, or browse the Newsroom for additional briefings.
Bottom line: treat this update as orientation, not instruction. Brazilian digital-rights policy is extending child-protection rules to emotional AI products and remains early. Organizations that benefit most will treat age assurance as engineering, keep humans on welfare escalations, and refuse to confuse a technical note with finished enforcement.