New Delhi: Platforms should stop relying on users to declare their own age, a Roblox executive said Wednesday, as representatives of OpenAI, Google and Roblox said the burden of protecting children online must move from parents to settings built into products.
“I think the age of self-declared age should be over. We live in the age of AI. We can age-verify,” said Nicky Jackson Colaco, Vice-President and Global Head of Public Policy at Roblox, at the Trust and Safety India Festival (TASI) 2026.
Speaking on the panel ‘Forces Reshaping Trust & Safety in 2026’, Colaco said parental controls remained necessary but could not be the main safeguard. “Every company is incumbent upon them to not only offer them, but to offer them in a robust way. But who we solve for should not be the parent,” she said. “We need to solve for the parents that aren’t able to use parental controls, not because they don’t love their kids.”
She said Roblox does not have a 13-plus divide, and children as young as five and six use the platform. Children can play without verifying their age but cannot communicate with anyone until they do. Roblox began requiring facial age checks for chat in December in Australia, New Zealand and the Netherlands before extending them worldwide in January.
“I don’t think people compete when it comes to safety,” Colaco said. “All it takes is one actor moving forward, a bold step for safety, to put downward pressure on the rest of the industry to do the same.”
Pragya Misra, Head of Strategy and Global Affairs, India, at OpenAI, said ChatGPT now predicts a user’s age from usage rather than what users declare. “It depends on the questions you ask, the time at which you are continuously accessing the device,” she said. “We are always veering towards the under-18 experience when we can’t really predict the age,” she said, adding that “many adults are now getting the by default under-18 experience”.
OpenAI rolled out age prediction in January. Adults placed in the under-18 experience can restore access by verifying through a selfie.
Misra said OpenAI took its teen settings to parent influencers to explain how to activate them, and asks users during onboarding whether they want to bring in a parent. “With AI being multimodal, hopefully we can reach a world where you can talk to the model and it can actually help you and guide you through that process,” she said.
On risks from AI itself, she said, “At every point the human should be in control and the AI should be serving us.” If development outstrips safety, she said, the company would hold back.
Snigdha Bharadwaj, Director of Search and Generative AI Trust Strategy at Google, said India requires a different approach. “There are shared devices, there’s multicultural understanding,” she said. “You have to move away from self-declared age to age verification to age assurance.”
She said safety policies must respond differently to children and adults. On a child asking about a dangerous online challenge, she said platforms need to respond “without having a persona” or telling them “I’m your friend”, and instead build in friction and disclaimers to steer the conversation elsewhere.
‘Decision to launch is with the client’
Akash Pugalia, Chief Digital Officer at TP, which runs red-teaming and adversarial testing for platforms, said model launches come with “tremendous pressure, but then there is always safety first”. His firm tests products against policies the companies write. “The decision to launch, not launch, get regulatory approval is with the client themselves,” he said.
Asked how devices such as camera-equipped smart glasses, which can record people without their knowledge, reach the market, Rohit Kumar, co-founder of Young Leaders for Active Citizenship, said the same glasses are “a lifesaver” for visually impaired users, and that manufacturers have changed such devices after release, including disabling the camera when the recording light is tampered with.
“Some harms companies put as they are going to market. Some, they perhaps downplay because there is an economic incentive,” Kumar said. “Should the regulatory set-up allow for sandboxing environments before there is mass exposure to a certain technology?”
He said India’s technology law rests on safe harbour, under which platforms are held liable only if they fail to take down third-party content when asked. “That is not a law that works for the current ecosystem of technology,” he said. Laws written for specific products, he said, leave gaps as new tools appear, and technology should be stress-tested against privacy and safety benchmarks before release.
Kumar said governments face “information asymmetry” and should set outcomes rather than prescribe solutions. He cited social media, where “the incentives are misaligned,” and the UK’s Online Safety Act, which names child safety as a risk to curtail.
India amended its intermediary rules in February to require labelling of AI-generated content.
Bharadwaj said there is “a wide gap today on how tech companies are thinking about this new technology versus how government is adapting”. Regulation, she said, “shouldn’t be prescriptive, but needs to work hand in hand with how technology is evolving.”
Watermarks and languages
Bharadwaj said companies share hashes of child sexual abuse material (CSAM) across the industry. On provenance, she said metadata-based standards fall short because “if you take a screenshot…that metadata actually just goes away,” and pointed to Google’s SynthID watermark, which more partners have adopted.
She said Google is working with MLCommons on a safety benchmark for Hindi and collecting language data across 750 districts. “What good looks like in Hindi cannot be compared with what good looks like in English,” she said.
Pugalia said AI labs are commissioning data with native speakers, including for “languages which I’ve never heard of in my life”.
Misra pointed to IndQA, OpenAI’s benchmark built with experts across India, which tests whether answers are “rooted in historical and cultural context of India,” not only correct in Hindi, Tamil or Telugu.
TASI is being hosted by the Centre for Social Research (CSR) and the Trust & Safety Forum at the India International Centre.
(Edited by Amrtansh Arora)
Also Read: An all-new teen-friendly ChatGPT is here