In an exclusive conversation with Economic Times Digital on the sidelines of the Trust & Safety India Festival in Delhi, Snigdha Bhardwaj, Director and Global Head of Generative AI, Trust & Safety at Google, discussed how the company is approaching child safety, age verification and AI literacy as younger users increasingly turn to chatbots.Artificial intelligence is becoming an integral part of how children learn, search for information and interact with technology. From completing homework to seeking personal advice, AI chatbots are increasingly being used by younger audiences. However, this also raises concerns about inappropriate content, emotional dependence and online exploitation.The numbers are concerning. According to the Internet Watch Foundation (IWF), 8,029 realistic AI-generated images and videos depicting child sexual abuse were identified in 2025, a 14% increase over the previous year. The organisation also reported a sharp increase in AI-generated videos, from just 13 in 2024 to 3,443 in 2025.In India, the National Crime Records Bureau recorded approximately 1.77 lakh crimes against children in 2023, a 9.2% increase over the previous year. While these figures are not specific to AI, they highlight the broader child safety challenges facing digital platforms.Against this backdrop, Economic Times Digital spoke with Snigdha Bhardwaj, Director and Global Head of Generative AI, Trust & Safety at Google, to understand how the company is building safeguards into Gemini.“The way Google looks at it, especially around Gemini safety, when I say Gemini, I mean all models of Gemini, across different surfaces and platforms, because what we do is build guardrails right at the model level,” Bhardwaj explained.Google’s approach is built around three pillars: Protect, Respect and Empower. These focus on age-appropriate responses, parental supervision and helping younger users develop the skills to use AI responsibly.Why Google doesn’t want a separate Gemini Kids experienceOne of the questions I raised was whether Gemini should have a dedicated child mode, similar to YouTube Kids, where parents could restrict the subjects their children can explore.Interestingly, Google does not believe creating separate compartments is necessarily the answer.“On our Gemini models, we want to make sure that we’re not creating those compartments,” Bhardwaj said.Instead, Google is relying on Family Link, which allows parents to manage Gemini access, alongside Android’s screen-time and focus controls.The company’s argument is that restricting children to specific subjects could limit their curiosity. Instead, Gemini should provide age-appropriate responses regardless of the topic.“What we want to do is whether you’re searching for soccer or how to lose weight in 10 days, you give them age-appropriate responses,” she explained.Google is also working to ensure that Gemini does not present itself as a human companion or encourage inappropriate emotional relationships with younger users.Google’s age assurance technology is coming to IndiaPerhaps the most interesting development from our conversation was Google’s plan to bring age assurance technology to India.Bhardwaj explained that Google uses three approaches to determine a user’s age: age declaration, age verification and age assurance.While the first two depend on users declaring or verifying their age, age assurance uses behavioural signals, including language and the context of interactions, to estimate whether a user may be underage.“The third area, which we rolled out right now in the US, announced it in Singapore, and now it’s coming to India, Malaysia and other countries, is age assurance,” she said.However, this raises another concern. Could bad actors deliberately manipulate their language to bypass these protections?“Bad actors have been trying to find loopholes in technology irrespective of what solution you offer them,” Bhardwaj acknowledged.She explained that Google conducts adversarial testing and works with external experts to identify vulnerabilities. AI is also helping the company detect suspicious behaviour and potential threats at scale.Why facial verification isn’t necessarily the answerI also asked whether facial verification could eventually become necessary for AI platforms, particularly considering the child safety concerns that have emerged on platforms such as Roblox.Bhardwaj believes the industry needs to balance safety with privacy.“Knowing age is an important idea, but should you then intervene into that privacy? I don’t think we’ve proven that none of the other solutions would work for us to resort to that,” she said.Google already uses identity verification in certain services, but Bhardwaj argued that requiring children to provide additional sensitive information could discourage them from accessing technology.The company is currently evaluating its age assurance systems in the United States before expanding them to other markets.How OpenAI and Anthropic are approaching child safetyGoogle is not the only AI company dealing with these challenges, although its approach differs from its competitors.OpenAI has introduced age prediction capabilities for ChatGPT, alongside additional protections for users identified as being under 18. It also offers parental controls that allow families to manage certain features and restrict usage hours. The company has additionally published a Child Safety Blueprint addressing AI-generated abuse material and reporting mechanisms.Anthropic has taken a different approach. Its Claude consumer service is restricted to users aged 18 and above, with age assurance and verification measures designed to identify underage users.This creates three distinct approaches across the industry. OpenAI is building a dedicated experience for teenagers, Anthropic is restricting its consumer service to adults, while Google is focusing on age-appropriate responses within Gemini rather than creating a separate children’s application.None of these approaches completely eliminates the possibility of misuse.Are AI-generated answers affecting children’s critical thinking?Beyond child safety, another concern I raised was whether AI-generated summaries are discouraging younger users from exploring information independently.With AI Overviews and conversational search, users can often get answers without visiting the original sources. This raises questions about whether children are receiving enough context to understand complex subjects.“AI shouldn’t hamper your critical thinking. It should actually supplement and help you make decisions faster,” Bhardwaj said.She highlighted Gemini’s Guided Learning experience, which encourages students to understand concepts through hints, explanations and follow-up questions rather than simply receiving answers.“It should empower and improve your curiosity as against disabling it,” she added.Bhardwaj also acknowledged that AI-generated information is not always accurate.“AIs can hallucinate and hallucination is an active research area. None of the people in the industry have yet solved it,” she said.Google encourages users to verify responses through source links and Gemini’s Double-check feature.AI literacy needs to go beyond promptingAnother important part of our discussion was whether children should be taught how to use AI more effectively, particularly when the quality of responses can depend on how questions are framed.Bhardwaj explained that Google’s immediate priority is broader AI literacy rather than simply teaching prompt engineering.“First of all, we want to make sure that everybody feels confident in embracing this change, and then we are learning as we are going,” she said.She highlighted Google’s Super Searcher programme in India and educational partnerships aimed at teaching students how to verify information, recognise inaccurate responses and avoid sharing sensitive personal details.Google is also consulting clinicians, child development specialists and education experts while developing its safety policies.For India, localisation remains another important consideration.“What we don’t do is we do not want to rely just on translation because we don’t feel that even that is sufficient. The context is important,” Bhardwaj said.The challenge aheadGoogle’s approach to child safety is built around model-level safeguards, parental supervision, age assurance and education. However, the broader industry still faces questions about the reliability of age estimation, the limitations of parental controls and the growing risks associated with AI-generated content.The challenge is particularly relevant in India, where children are increasingly gaining access to AI-powered devices and services, often without continuous parental supervision.As Gemini, ChatGPT and Claude become more capable, companies will have to demonstrate that their safety measures can keep pace with the technology itself.For parents, meanwhile, safeguards can help, but they cannot completely replace supervision and digital literacy.Ultimately, the question is no longer whether children will use AI, but how safely and responsibly they can use it.Add as a Reliable and Trusted News Source
————————————————
Source link
