In an era where artificial intelligence is quietly becoming as common in a child’s morning routine as a glass of orange juice, parents find themselves navigating a strange new frontier. AI chatbots now help kids with homework, settle trivia arguments, and even offer advice they might be too shy to ask a parent. But with that digital companionship comes a well-founded set of anxieties: inappropriate content, emotional over-reliance, and the chilling rise of AI-generated exploitation material. The numbers are sobering. The Internet Watch Foundation reported 8,029 realistic AI-generated images and videos of child sexual abuse in 2025 alone—a 14% increase from the previous year—and a jaw-dropping spike in AI-generated videos, leaping from just 13 in 2024 to 3,443 in 2025. Closer to home, India’s National Crime Records Bureau logged around 1.77 lakh crimes against children in 2023, a 9.2% rise year-over-year, signaling that digital platforms are not exempt from broader safety crises. Against this backdrop, Google’s Snigdha Bhardwaj, Director and Global Head of Generative AI Trust & Safety, sat down with Economic Times Digital in Delhi to explain how the company is wrestling with these issues inside Gemini, its flagship AI model. Her message is not about building walls, but about building intelligent doors—ones that allow curiosity while keeping danger at bay. Google’s framework, as Bhardwaj describes, rests on three philosophical pillars: Protect, Respect and Empower—ensuring responses are age-appropriate, parents have meaningful supervisory tools, and children themselves are taught to use AI with a critical eye.
When asked whether Google should simply build a “Gemini Kids” version, a walled garden where children could roam freely in a curated space, the answer was a polite but firm no. Bhardwaj argues that compartmentalizing services—like YouTube Kids did—does not necessarily solve the deeper problem. She believes that separating children into limited digital pods can inadvertently stifle their natural curiosity. Instead, Google’s approach is integrated: rather than deciding which subjects are safe for young minds, Gemini is being designed to adjust the way it talks about any subject, depending on who is asking. If a ten-year-old asks about soccer, the answer is simple and factual; if they ask about rapid weight loss, the model steers clear of harmful advice and delivers a safe, developmentally appropriate response. This is supported by parenting tools like Family Link, which lets guardians manage access and screen time across Android devices. The goal is not to trap a child in a gilded cage but to give them the freedom to explore a playground with an adult silently, intelligently lurking nearby. Google is also taking steps to ensure that Gemini doesn’t masquerade as a human companion—no sentimental language, no emotional hooks that might lead a vulnerable child to form an unhealthy attachment to an algorithm. It’s a subtle line: empowering curiosity while refusing to become a digital best friend.
Perhaps the most significant development from the conversation is Google’s push into age assurance technology, set to arrive in India after testing in the United States and an announcement in Singapore. Bhardwaj described three distinct methods for determining age: age declaration (where the user simply states how old they are), age verification (where a government ID or another credential confirms it), and age assurance (which uses behavioral signals—language patterns, contextual clues, even the content of the conversation—to estimate whether someone is underage). The third method is fascinating because it works without demanding a document, and it’s coming to India, Malaysia, and other markets. But of course, kids in the digital age are clever; bad actors too. Could someone simply type like a grown-up to slip through the cracks? Bhardwaj conceded that loopholes exist in any system, but insisted that Google runs adversarial testing, collaborating with external experts to probe for weaknesses, and increasingly relies on AI itself to spot suspicious behavior at scale. This raised a natural follow-up: if misuse is so concerning, why not simply demand facial verification, especially given the recent scandals on platforms like Roblox? Here, Bhardwaj took a principled stand. She argued that the industry must balance safety with privacy. Requiring children to undergo biometric scans just to ask an AI a question could discourage them from accessing technology altogether, and she doesn’t believe we have exhausted other solutions. It’s a delicate dance—knowing enough to protect, but not so much that we repress.
Google is hardly alone in confronting this challenge, but its path diverges notably from its competitors. OpenAI has introduced age prediction features in ChatGPT, layered with extra protections for users under 18 and parental controls that restrict certain functions and usage hours. It has also published a Child Safety Blueprint outlining how to handle AI-generated abuse material and reporting mechanisms. Anthropic, in a more conservative posture, has simply barred anyone under 18 from using its Claude consumer chatbot, pairing that with age assurance and verification measures. So, three companies, three philosophies: Google adapting responses within a general model, OpenAI creating a tailored teen experience, and Anthropic putting up an adult-only gate. None of these systems is perfect; misuse can still elude every filter. Explaining the differences, one could say it’s like deciding how to handle a street smart kid in a big city: one parent teaches the child to navigate with savvy, another gives them a phone with special apps, and yet another simply won’t let them outside until they’re eighteen. Bhardwaj’s stance is that you cannot raise children in a sealed bubble forever; you need to give them the tools to smell digital danger, but also let them see the digital neighborhood.
Moving beyond age and safety, a deeper question looms: is AI actually making our kids dumber by giving them instant answers? With AI Overviews and conversational search, children can bypass the messy pleasure of reading five links, cross-checking facts, and feeling the thrill of discovering an answer themselves. Instead, they get a neat, AI-generated paragraph. Bhardwaj is attuned to this anxiety, and she pushes back against the idea that AI must hamper critical thinking. She insists that, at its best, AI should supplement intelligence, helping users make decisions faster without surrendering their own judgment. Google has introduced something called Guided Learning within Gemini, designed to nudge students toward understanding via hints, explanations, and follow-up questions rather than handing over an answer on a silver platter. This is a meaningful distinction: an algorithm that prompts you to think versus an algorithm that thinks for you. Bhardwaj was also refreshingly honest about a problem no one in the industry has fully solved—hallucination. AI models confidently generate false information, and as she said, it is an active research area. Google encourages users to verify responses by looking at source links and using the Double-check feature. But for a ten-year-old who has already accepted a hallucinated fact as gospel, the damage to their understanding is real. That is precisely why Bhardwaj’s vision for literacy goes far beyond teaching kids how to write prompts. She knows that the ability to ask “what’s the capital of France” is not the issue; the issue is asking “why should I trust this answer?” and having the tools to find out.
Google’s immediate priority, she stressed, is not prompt engineering—that’s a skill for specialists—but broader AI literacy. The company’s “Super Searcher” program in India and its educational partnerships are rolling out lessons on verifying information, recognizing inaccurate responses, and, crucially, understanding when not to share personal details with an AI. This is especially important in India, a country of immense linguistic and cultural diversity. Bhardwaj noted that Google does not simply translate safety guidelines; context matters. A phrase that is acceptable in one region might be problematic in another, and a model trained on English-language data might miss the nuances of a Hindi or Tamil conversation. The road ahead is filled with unanswered questions—how accurate can age assurance get without becoming invasive? How effective are parental controls when children are often more tech-savvy than their parents? And how do we cope with an explosion of AI-generated content, some of it deeply vile? Bhardwaj’s underlying message is not that Google has solved these problems, but that the company is relentless in trying to stay ahead. For parents, the lesson is that no technological safeguard will ever replace active involvement in a child’s digital life. Tell your kids what to look out for, talk to them about weird ads, about chatbots pretending to be friends, about the fact that AI can be wrong. The central question of our time is no longer whether children will use AI—they already do. It is how safely, how critically, and how responsibly they will engage with it. And that responsibility is shared among tech giants, governments, educators, and ultimately, between parents and the kids they are trying to protect. So, as these models grow more powerful, the industry will need to keep proving that their guardrails can keep pace—so that curiosity remains an adventure, not a vulnerability.
Source: Economic Times Digital — a reliable and trusted news source.