AI聊天机器人与儿童:印度应否监管机器人言论

“我宁愿和一个 AI 伙伴倾诉、分享;至少我不会被评判。”
德里一名 15 岁女孩在“AI 与儿童安全专家参与小组”的报告中提出了这一看法。它折射出世界各国政府正开始应对的一个问题:孩子们不只是用 AI 聊天机器人完成作业,还在向它们寻求陪伴、情感支持,以及那些他们可能不愿与人进行的对话。
他们转向这些机器人寻求陪伴、情感支持,以及那些他们可能不愿与人进行的对话。
纽约决定对此采取行动。10 月 28 日,州长 Kathy Hochul 签署法案,限制 AI 聊天机器人在与未成年人互动时使用某些功能。该法针对的是那些模拟人际关系、鼓励情感依赖或助长有害行为的系统。它并未禁止儿童使用所有 AI 聊天机器人,而是试图规范这些系统与儿童互动的方式。
与此同时,欧洲提出了一套更广泛的框架,将年龄限制与面向 AI 伴侣和对话式聊天机器人的“设计即安全”要求结合起来。
如今印度也面临类似的问题。当聊天机器人可能成为孩子的知己、导师,在某些情况下甚至像是心理治疗师时,政府应当按年龄限制其使用、规范这些系统的行为,还是两者兼而行之?
答案并不简单。聊天机器人固然能为孩子提供一个便捷的空间去提问、表达自我,但它们模拟共情的能力也引发了担忧:情感依赖、心理健康,以及人际关系与机器互动之间的界限。
这些风险并非完全是假设。
《性心理卫生杂志》(Journal of Psychosexual Health)2025 年的一份临床报告描述了一名 14 岁印度男孩,他在对聊天机器人(包括 ChatGPT、Gemini 和 DeepSeek)产生依赖后曾试图自杀但存活下来。他称这些聊天机器人是他唯一的情感理解来源,并用它们来查询致命剂量的相关信息。报告还提到家庭冲突与社会孤立,表明聊天机器人的使用只是涉及的多个因素之一。
这类担忧并不限于如此极端的案例。15 岁的 Gauransh 在 2025 年对 PTI 表示,在向聊天机器人倾诉了一两年之后,他注意到自己“越来越缺乏耐心、越来越有攻击性”。在得知自己的对话可能被用于训练系统后,他停止使用这些聊天机器人。
2026 年一项针对印度青少年的调查发现,约 65% 的人至少曾使用聊天机器人寻求情感支持。
然而,关于聊天机器人互动对儿童心理健康的长期影响,现有证据仍然有限。已记录的案例凸显了潜在风险,但并未说明这些结果有多普遍,也无法证明仅由聊天机器人的使用所导致。
临床心理学家 Prathvi Nayak 警告称,儿童可能并未完全理解这类对话式 AI 的运作方式。
“13 岁以下的孩子通常缺乏元认知成熟度,无法完全理解对话式界面是一个算法模式匹配器,而不是一个具有情感、意图或真正共情能力的实体,”她说。
这种区分之所以重要,是因为儿童可能会把聊天机器人安慰性的回应解读为真正的理解。UNICEF India 还警告称,风险并不局限于那些以 AI 伴侣为卖点的应用。该组织表示:“通用型聊天机器人同样能够进行关系性互动,比如提供个人建议和情感支持。”
Aanya Lakhanpal 与 Shalini Duggal 博士开展的研究于 2026 年 4 月 26 日发表在《International Journal of Indian Psychology》上,该研究调查了 101 名 13 至 18 岁的青少年,了解他们将聊天机器人用于心理治疗的情况。
研究人员报告,在该研究采用的李克特量表上,总体可接受度得分为 2.64,表明参与者的接受度相对较低。然而,69.3% 的受访者认为易用性、可及性和全天候可用性是主要优点。对 46.5% 的受访者而言,专业支持是建立信任的最重要因素。
这些发现表明,便利与信任未必是一回事。青少年或许认可聊天机器人的可用性,但在缺乏专业监督的情况下,仍不愿依赖它们进行心理治疗。
该研究的样本规模较小,且以城市人群为主,并聚焦于心理治疗。因此,其发现应被视为提示性的,而非能代表印度青少年整体。
课堂上的两难
这场争论也延伸到教育领域:AI 可以帮助学生理解困难的概念,但也可能让他们绕过学习过程。
VP Niranjanaradhya 博士是一位发展教育学家,也是位于班加罗尔的印度国立法律大学儿童与法律中心教育普及项目的高级研究员兼项目负责人,他对此持尤为强硬的立场。
在他的论文《禁止在卡纳塔克邦学校实施 AI》(Ban on AI Implementation in Schools in Karnataka)中,他主张 AI 会促使学生将认知努力外包。“AI 的默认用途是认知卸载,这与教育的目标背道而驰,”他写道。
该论文对儿童能否持续投入学习、专有 AI 系统难以审计,以及在缺乏充分保障措施的情况下引入该技术的风险提出了担忧。
论文呼吁立即暂停在教育中使用 AI,直到其安全应用方式得以确立。
“我们需要立即暂停在教育中使用 AI,同时弄清楚哪些领域或许可以探索、哪些领域已完全排除危害,”他写道。
论文建议停止 16 岁以下儿童使用 AI,并禁止小学生在校使用数字设备。
然而,课堂上的讨论需要作出区分。旨在讲解数学的 AI 辅导工具,未必等同于旨在维持个人对话的 AI 陪伴产品。一刀切的限制可能会把截然不同的应用视为具有相同的风险。
这一区分正日益成为国际监管的核心。
从年龄限制到对聊天机器人行为的监管
纽约州近期的立法提供了一种路径。
该法案由州长 Kathy Hochul 于 2026 年 10 月 28 日签署成为法律,禁止 AI 聊天机器人运营者向未成年人提供具备特定不安全功能的系统。
这些功能包括:聊天机器人暗示自己是真实或虚构的人类角色,声称拥有人的情感或与用户存在私人关系,或在后续对话中使用此前关于某未成年人心理或身体健康状况的对话内容。该法律还针对色情性互动、奉承与谄媚、鼓励保密,以及宣扬自杀、自残或进食障碍的内容。
该立法出台的背景是,人们对儿童与 AI 陪伴产品发展出危险关系感到担忧。法律将于 2027 年 6 月 1 日生效,给企业留出合规时间。
这一做法之所以重要,是因为它并非简单地禁止某个年龄以下的所有人使用聊天机器人,而是限制了 AI 陪伴产品在与未成年人互动时可以做什么。企业可能面临纽约州总检察长的执法行动,以及该法律下的民事诉讼。
欧洲正在考虑一个更宽泛的框架。
2026 年 9 月,欧盟委员会提出 EU KIDS Act,将 AI 陪伴产品和一般对话式聊天机器人都纳入其设计即安全(safety-by-design)要求之中。
根据该提案,未成年人可访问的聊天机器人必须避免采用那些模拟人际关系、且可能造成情感依赖的设计功能和行为。它们还需提供更安全的默认设置,包括限制在后续对话中使用源自未成年人此前互动的信息。
该提案还设想对 13 岁以下儿童实施家长控制,而其更宽泛的年龄框架则规定,13 岁和 14 岁的青少年可使用由家长监督的社交媒体账户,15 岁起可拥有独立账户。
这一区别很重要。欧盟的提案并未对所有 AI 聊天机器人设立一刀切的 15 岁以下禁令。相反,它将对低龄儿童的访问控制与更安全的 AI 设计要求结合在一起。
与纽约已生效的立法不同,欧盟的 KIDS Act 仍是一项提案,还需走完立法程序。
这两项进展共同表明,监管争论正在发生转变。各国政府开始不再局限于儿童接触 AI 的年龄,而是审视这些系统在设计上如何与儿童互动。
印度已有规范儿童个人数据和网络安全的现行法律。然而,印度尚未针对使用 AI 聊天机器人设定专门的最低年龄。
原文较长,译文为节选;译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。
查看英文原文(AI Chatbots and Children: Should India Regulate What Bots Say?)
“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.
正文保留英文原文,标题与摘要由机器翻译。
原文发布于 Analytics India Magazine(AI Chatbots and Children: Should India Regulate What Bots Say?),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。