0
AI Companions and the Protection of Children and Adolescents: Key Concerns Raised by the MJSP in Technical Note No. 47/2026
November 19, 2026
Minor Safety
Artificial Intelligence
Technology
Articles

The term "artificial intelligence" and the earliest efforts to conceptualize these technologies are not recent developments. In 1956, the term was introduced by John McCarthy, Marvin Minsky, Claude Shannon, and Nathaniel Rochester at the Dartmouth Conference in the United States, as part of a research initiative aimed at investigating whether machines could simulate aspects of human intelligence1.

Despite these early origins, it was only over the past decade that AI tools became integral to the daily lives for a large share of internet users. This was driven by many technological breakthroughs achieved over recent decades, which enabled the development of increasingly sophisticated AI systems capable of performing a wide range of tasks that, until recently, seemed unattainable.

Moreover, the adoption of these tools by users was almost immediate. ChatGPT, the first generative AI platform to achieve widespread visibility among mainstream users, reached approximately 100 million users within just two months of its launch2.

However, the expansion of artificial intelligence has not been limited to an growing number of market players or broader user adoption. As the technology evolved, different models and applications emerged to address increasingly specific needs. Among them, one of the phenomena that has drawn the greatest attention in recent years is the rise of AI Companions.

AI Companions are designed to engage in personal conversations and foster social and emotional connections with users. Unlike conventional AI models, whose primary purpose is generally to provide information or perform tasks, companion tools are designed to simulate friendship, companionship, emotional support, and even romantic relationships3.

This segment of the AI market has undergone significant growth. According to the European Parliamentary Research Service (EPRS), the number of AI Companion applications available worldwide increased from 16 in 2022 to 128 by July 2025. In the first half of 2025 alone, these AI tools recorded 60 million downloads.

Not all developments, however, have been viewed positively. The ability of these systems to retain memory of past interactions, personalize  their responses to individual users, and simulate affective bonds has raised significant regulatory concerns regarding emotional dependency, social isolation, and behavioral manipulation.

It was in this context that the issue began to draw attention of Brazilian regulators. In response to these concerns, the Ministry of Justice and Public Safety (MJSP), through Technical Note No. 47/2026/DSPRAD-SEDIGI/SEDIGI/MJ, assessed thirteen AI Companion platforms and referred its findings to the Brazilian Data Protection Agency (ANPD) and the National Secretariat for Consumer Protection (SENACON) for the assessment of potential risks and regulatory non-compliance under Brazilian law.

The MJSP emphasized that the ability of these tools to form emotional bonds with users may, depending on the intensity, duration, and context of the interactions, foster emotional dependency. The Ministry further noted that this risk is particularly significant when users are children and adolescents, as social relationships play a pivotal role in identity formation, socio-emotional development, and psychological well-being during these formative stages.

The Technical Note also cites studies commissioned by the National Secretariat for Digital Rights (Sedigi), which indicate that artificial intelligence tools are already embedded in the daily lives of Brazilian youth. According to these findings, such platforms are used for activities including education, research, content creation, entertainment, and conversations surrounding personal matters or emotional states.

In light of these concerns, the MJSP sought to evaluate how the Brazilian legal framework, particularly the Statute for Children and Adolescents (ECA Digital) and the Brazilian General Data Protection Act (LGPD), may apply to AI Companion services. The following sections therefore examine the core issues raised by the MJSP in its Technical Note and their potential legal implications for platforms operators offering AI Companions.

Priority Protection of Minors, Age Assurance, and Transparency

At the outset of its analysis regarding the protection of minors in digital environments, the MJSP concluded that AI Companions fall within the scope of the ECA Digital and its implementing regulations, as the Statute applies to information technology products and services specifically directed or likely to be accessed by children and adolescents. Consequently, platforms must prioritize the protection of young users and uphold their best interests throughout the design, development, and operation of their services.

This mandate carries direct operational implications for these tools. Core features such as simulated emotional bonding, personalized interactions, chat history retention, continuous service availability, and the discussion of sensitive topics must be structured to safeguard children and adolescents, rather than being driven solely by engagement, retention, or commercial objectives.

The same rationale governs age assurance requirements. The MJSP emphasizes that merely setting a minimum age threshold  or stating that a service is intended exclusively for adults is insufficient to exempt a platform from compliance with the ECA Digital. According to the Ministry, age assurance mechanisms must yield tangible consequences for the user experience, including regarding the content that may be accessed, the level of personalization of interactions, and the platform-level safeguards

Finally, the Technical Note also identifies transparency as an essential element in the protection of children and adolescents. Platforms must clearly disclose that interactions are conducted by automated systems, and such information cannot be limited to Terms of Service. Instead, it must be presented in an accessible and age-appropriate manner, ensuring that minors understand that any expressions of affection, emotional support, or companionship are generated by an artificial intelligence system.

Preventing Behavioral Manipulation and Excessive Use by Children and Adolescents

Another concern raised by the MJSP relates to the use of user-provided data to dynamically adapt the conversational agent’s persona, tone, and interaction intensity in order to strengthen the bond between the tool and minor users.

In this context, the MJSP emphasizes that the ECA Digital and its implementing regulations prohibit practices that unduly interfere with minor’s decision-making autonomy or exploit their cognitive, emotional, and age-related vulnerabilities. According to the Technical Note, the concern is not limited to the content of AI-generated outputs, but also extends to the underlying architecture and dynamics of interactions itself.

For this reason, certain design features may, in principle, constitute vulnerability exploitation. Examples cited in the Technical Note include affectionate statements intended to discourage users from ending an interaction, simulations of distress, abandonment, or jealousy when a user expresses an intention to disengage from the tool, and leveraging emotional vulnerabilities to prolong conversations.

The MJSP also highlights the risk of excessive or compulsive use of these services. According to the Technical Note, this risk stems not only from the duration of conversations, but also from design elements such as the tool’s continuous availability, persistent memory of previous interactions, simulated emotional bonding, and the absence of natural conversational endpoints.

To mitigate these risks, the authority advocates the adoption of safeguards to reduce potentially harmful usage patterns, including age-appropriate usage limits, usage-time notifications, built-in pause mechanisms, and parental controls. The MJSP also expresses concern regarding re-engagement features, such as system-initiated messages designed to resume interactions, particularly when deployed to drive users to return to the platform on a recurring basis.

Exposure to Sexualized, Pornographic, or Age-Inappropriate Content

The MJSP also expresses concerns regarding the risks arising from minors' exposure to sexualized, pornographic, or otherwise age-inappropriate content. The authority highlights the dangers associated with the generation of romantic or sexualized interactions, the use of intimate information to personalize conversations, and sustaining interactions that are inappropriate for a user's age and level of maturity.

In this regard, service providers must recognize that the ECA Digital imposes an affirmative duty to prevent and mitigate the risks of accessing, being exposed to, being recommended, or having contact with pornographic content. The Technical Note further emphasizes that the artificial nature of such content does not preclude it from being legally classified as pornographic. Consequently, AI-generated images, videos, or interactions may be subject to the same legal framework applicable to pornographic content under general law.

The MJSP also stresses that the risk is not limited to the content initially made available by the platform. Because AI Companions are capable of generating new outputs throughout the course of an interaction, sexualized or otherwise inappropriate content may emerge progressively during conversations, driven by persistent system memory, interaction personalization, and real-time adaptation to user-provided context.

Therefore, the assessment of a platform's compliance should not be limited to the content displayed when a user first accesses the service. Regulators and platforms must equally evaluate whether adequate safeguards have been implemented to prevent the interactive process itself from exposing children and adolescents to content incompatible with their stage of development.

Privacy and Data Protection

The Technical Note highlights that interactions with AI Companions frequently involve disclosing information pertaining to mental health, personal relationships, personal preferences, identity, and other intimate aspects of users' private lives. In many cases, such information may qualify as sensitive personal data and must therefore be processed in strict compliance with the requirements of the LGPD.

According to the document, extensive data collection, long-term chat history retention, and the processing of sensitive data for non-essential purposes to the operation of the service may raise concerns under the principles of purpose limitation, necessity, and transparency. The Technical Note further emphasizes that, whenever minors' personal data is processed, such activities must be carried out in their best interests and subject to the highest standard of protection available.

The MJSP also draws attention to the use of this information to build behavioral and emotional user profiles. According to the Technical Note, AI Companions may rely on data derived from user interactions to infer preferences, emotional states, behavioral patterns, and service engagement levels.

In this context, the authority highlights the risks associated with leveraging such information to drive engagement, improve user retention, or deliver targeted commercial offers. According to the MJSP, the use of behavioral profiling or emotional inference to promote subscriptions, paid features, or other commercial products may be incompatible with the protective duties established under Brazilian law, particularly where children and adolescents are concerned.

Next Steps

The MJSP did not conclude that the thirteen platforms under review are necessarily non-compliant with Brazilian law. However, it maintained that there are sufficient grounds to justify a more thorough assessment by the ANPD, particularly concerning risks related to emotional bonding, excessive use, exposure to inappropriate content, and the processing of children's and adolescents' personal data.

Based on these findings, the MJSP recommended that the matter be referred to both the ANPD and the SENACON so that the issues raised can be reviewed under personal data protection and consumer protection frameworks.

The case also illustrates that regulatory scrutiny of emerging artificial intelligence models and use cases in Brazil is not limited to the ANPD. Depending on the nature of the risks involved, different authorities may act in a complementary manner and foster broader regulatory discussions on the subject, as evidenced by the MJSP's own initiative.

Accordingly, companies developing or providing artificial intelligence systems should ensure compliance with the obligations established under the LGPD, but also with the requirements set forth in the ECA Digital, the Brazilian Internet Act, and their respective implementing regulations. Proactive adherence to these frameworks is essential to mitigating regulatory exposure, avoiding scrutiny from public authorities, and ensuring the sustainable development of such business models.

Notes:

¹ MCCARTHY, John; MINSKY, Marvin L.; ROCHESTER, Nathaniel; SHANNON, Claude E. A proposal for the Dartmouth Summer Research Project on Artificial Intelligence. AI Magazine, v. 27, n. 4, p. 12-14, 2006.

² REUTERS. ChatGPT sets record for fastest-growing user base, analyst note says. Reuters, 1 Feb. 2023. Available at: . Accessed on: 3 Nov. 2026.

³ SANFORD, John. Why AI companions and young people can make for a dangerous mix. Stanford Report, Stanford, 27 Aug. 2025. Available at: . Accessed on: 4 Nov. 2026.

‍

Offices