“I would rather speak to and share with an AI companion; at least I am not being judged.”
A 15-year-old girl from Delhi made this observation in the Expert Engagement Group on AI and Child Safety’s report. It captures something governments around the world are beginning to grapple with: children are not just using AI chatbots to complete homework; they are turning to them for companionship, emotional support, and conversations they might hesitate to have with people.
They are turning to them for companionship, emotional support and conversations they might hesitate to have with people.
New York has decided to do something about it. On October 28, Governor Kathy Hochul signed legislation restricting AI chatbots from using certain features when interacting with minors. The law targets systems that simulate human relationships, encourage emotional dependency or promote harmful behaviour. It does not ban children from using all AI chatbots. Instead, it attempts to regulate how these systems interact with them.
Europe, meanwhile, has proposed a broader framework combining age restrictions with safety-by-design requirements for AI companions and conversational chatbots.
India now faces a similar question. When a chatbot can become a child’s confidant, tutor and, in some cases, an apparent therapist, should the government restrict access by age, regulate the behaviour of these systems, or do both?
The answer is not straightforward. While chatbots can offer children a convenient space to ask questions and express themselves, their ability to simulate empathy also raises concerns about emotional dependency, mental health and the boundaries between human relationships and machine interactions.
When AI becomes a confidant
The risks ar ere not entirely hypothetical.
A 2025 clinical report in the Journal of Psychosexual Health described a 14-year-old Indian boy who survived a suicide attempt after growing dependent on chatbots, including ChatGPT, Gemini and DeepSeek. He called them as his only source of emotional understanding and used them to seek information about a lethal dose. The report also cites family conflict and social isolation, indicating that chatbot use was one of several factors involved.
The concerns extend beyond such extreme cases. Gauransh, 15, told PTI in 2025 that after a year or two of confiding in chatbots, he noticed “growing impatience and aggression” in himself. He stopped using them after learning that his conversations could be used to train the system.
A 2026 survey of Indian adolescents found that around 65% had used a chatbot at least once for emotional support.
However, evidence on the long-term effects of chatbot interactions on children’s mental health remains limited. Documented cases highlight potential risks, but they do not establish how widespread these outcomes are or prove that chatbot use alone caused them.
Clinical psychologist Prathvi Nayak has cautioned that children may not fully understand how the conversational AI works.
“Under age 13, children generally lack the metacognitive maturity to fully grasp that a conversational interface is an algorithmic pattern-matcher rather than an entity with feelings, intent, or genuine empathy,” she said.
The distinction matters because children may interpret a chatbot’s reassuring responses as genuine understanding. UNICEF India has also warned that the risks are not confined to applications marketed as AI companions. “General-purpose chatbots are also capable of relational interactions, like offering personal advice and emotional support,” the organisation has said.
Convenience, not trust
Research by Aanya Lakhanpal and Dr Shalini Duggal, published in the International Journal of Indian Psychology on April 26, 2026, surveyed 101 adolescents aged 13 to 18 about using chatbots for therapy.
The researchers reported an overall acceptability score of 2.64 on the study’s Likert scale, indicating relatively low acceptance among participants. Yet 69.3% identified ease of use, accessibility and round-the-clock availability as major benefits. Professional backing was the most important factor in building trust for 46.5% of respondents.
The findings suggest that convenience and trust are not necessarily the same thing. Adolescents may appreciate the availability of chatbots but remain reluctant to rely on them for therapy without professional oversight.
The study’s sample was small, largely urban and focused on therapy. Its findings should therefore be treated as indicative rather than representative of Indian adolescents as a whole.
The classroom dilemma
The debate also extends to education, where AI can help students understand difficult concepts but may also let them bypass the learning process.
Dr VP Niranjanaradhya, a development educationist and senior fellow and programme head of the Universalisation of Education Programme at the Centre for Child and the Law, National Law School of India University, Bengaluru, has taken a particularly strong position.
In his paper, Ban on AI Implementation in Schools in Karnataka, he argues that AI encourages students to outsource cognitive effort. “AI’s default use is for cognitive offloading, which is antithetical to the aims of education,” he writes.
The paper raises concerns about children’s sustained engagement with learning, the difficulty of auditing proprietary AI systems and the risks of introducing the technology without adequate safeguards.
It calls for an immediate pause on AI use in education until its safe applications can be established.
“We need to immediately pause the use of AI in education, while we figure out what areas may be possible to explore, where harm is completely ruled out,” he writes.
The paper recommends stopping AI use by children under 16 and prohibiting student access to digital devices in primary schools.
However, the classroom debate needs a distinction. An AI tutor designed to explain mathematics is not necessarily comparable to an AI companion designed to sustain personal conversations. A blanket restriction could treat very different applications as if they carried identical risks.
That distinction is increasingly becoming central to international regulation.
From age limits to regulating chatbot behaviour
New York’s recent legislation offers one approach.
Signed into law on October 28, 2026, by Governor Kathy Hochul, the legislation prohibits AI chatbot operators from providing minors with systems that have specified unsafe features.
These include chatbots that suggest they are real or fictional human characters, claim to have human emotions or personal relationships with users, or use previous conversations about a minor’s mental or physical well-being in subsequent interactions. The law also targets sexually explicit interactions, flattery and sycophancy, encouragement of secrecy, and content promoting suicide, self-harm or disordered eating.
The legislation followed concerns about children developing dangerous relationships with AI companions. It takes effect on June 1, 2027, giving companies time to comply.
The approach is significant because it does not simply prevent everyone below a particular age from accessing chatbots. It restricts what AI companions can do when interacting with minors. Companies can face enforcement action by the New York Attorney General, as well as civil proceedings under the law.
Europe is considering a broader framework.
In September 2026, the European Commission proposed the EU KIDS Act, which includes AI companions and general conversational chatbots within its safety-by-design requirements.
Under the proposal, chatbots accessible to minors would have to avoid design features and behaviours that simulate interpersonal relationships likely to create emotional dependency. They would also need to provide safer default settings, including restrictions on using information derived from a minor’s previous interactions in subsequent conversations.
The proposal also envisages parental controls for children under 13, while its broader age framework provides for parent-supervised social media accounts for 13- and 14-year-olds and independent accounts from 15.
The distinction is important. The EU proposal does not establish a blanket under-15 ban on all AI chatbots. Instead, it combines access controls for younger children with requirements for safer AI design.
Unlike New York’s enacted legislation, the EU KIDS Act remains a proposal and will need to go through the legislative process.
Together, the two developments point to a shift in the regulatory debate. Governments are beginning to look beyond the age at which children can access AI and examine how these systems are designed to interact with them.
Does India need a blanket ban?
India already has laws governing children’s personal data and online safety. However, it does not have a dedicated minimum age for accessing AI chatbots.
The Digital Personal Data Protection Act, 2023, requires verifiable parental consent for processing children’s personal data and restricts tracking, behavioural monitoring, and targeted advertising, subject to the law’s provisions and exemptions.
The Digital Personal Data Protection Rules, 2025, were notified in November 2025. However, most substantive obligations on companies, including verifiable parental consent, are scheduled to take effect in May 2027.
These provisions address how companies process children’s data. They do not, by themselves, establish rules governing whether a chatbot can simulate a personal relationship, encourage emotional dependency or provide inappropriate mental health advice.
The Expert Engagement Group on AI and Child Safety has recommended mandatory child-rights impact assessments for high-interaction AI systems, AI literacy in schools, privacy-preserving age tokens and greater algorithmic transparency.
A blanket age restriction might seem the simplest solution. But it raises questions about enforcement, age verification, and whether teenagers would simply migrate to less-regulated services.
Nayak has warned that restrictions could have unintended consequences.
“By age 13 to 15, digital usage usually shifts toward seeking autonomy. Blanket bans for this cohort can backfire by encouraging covert use on unmonitored platforms, driving adolescents away from safe, moderated tools to less regulated alternatives,” she said.
This is where India’s policy response needs to be more nuanced. A general-purpose chatbot, an AI tutor and a dedicated AI companion serve different purposes. Their potential risks also differ.
What should safeguards look like?
The international developments suggest that governments are moving beyond relying solely on parental supervision and age declarations.
For India, a risk-based approach could include restrictions on high-risk companion features, age-appropriate default settings, independent safety testing and mandatory escalation to human support when conversations involve suicide or self-harm.
Chatbots could also be required to clearly disclose that they are AI systems, avoid presenting themselves as human beings or qualified mental health professionals, and prevent emotionally manipulative interactions with minors.
Educational applications would need a separate framework that distinguishes between AI-assisted learning and systems that replace a child’s cognitive effort.
Companies should also be accountable for how their systems behave in practice, not just for publishing safety policies. Independent audits, transparent reporting of child-safety incidents and accessible complaint mechanisms could help establish that accountability.
New York has demonstrated one way to regulate chatbot behaviour, while Europe is attempting to combine age restrictions with safety-by-design requirements.
For India, the question is no longer simply whether children should be allowed to use AI chatbots. It is whether the systems they use should be allowed to behave like their friends, therapists or trusted confidants in the first place.