At the TASI 2026 Festival in New Delhi on Wednesday, October 14, executives from OpenAI, Google and Roblox, policy researchers and safety consultants debated AI safety, child protection and regulation. Panelists said AI is advancing faster than safeguards and discussed deepfakes, privacy risks from camera-equipped smart glasses, age verification and gaps in India’s regulatory framework. The two-day conference was jointly organised by the Centre for Social Research and TASI to discuss AI governance and online safety.

Participants debated whether companies can manage risks through testing and voluntary cooperation or whether governments should set clearer expectations before AI models and tools reach the public. Topics included red-teaming (testing systems with deliberate attacks), sharing safety signals across the industry through SynthID and hash matching, safer defaults for children, age prediction, and evaluations that account for India’s multicultural context. OpenAI’s Pragya Misra said the company had slowed reinforcement learning and training runs because it considers safety a priority, while also seeking to reach AGI first and have it benefit humanity.

Rohit Kumar, founding partner of The Quantum Hub, argued that competition can weaken companies’ incentives to invest in trust and safety, despite the value of user trust. He called for early, flexible regulation that sets clear outcomes and closer cooperation among governments, technology companies and researchers. Google’s Snigdha Bhardwaj said the company has zero tolerance for child sexual abuse material and may slow a model launch if it produces such content. She said Google continuously uses red-teaming to test harmful outputs and urged platforms to share risk signals through an interoperable safety layer, while addressing privacy and intellectual-property concerns. The panel also raised controlled testing of smart glasses before wider release.

Roblox’s Nicky Jackson Colaco said child safety should be the default, rather than depend on parents configuring controls. She described age-based limits on communication and content and said stronger safeguards at one company could encourage others to follow. Akash Pugalia of TP warned that children may rely on Gemini or ChatGPT for answers and raised concerns about critical thinking, grooming and convincing audio and video deepfakes. Misra said OpenAI’s research team found learning was a key use of its models among young people in India last year, but users said direct answers could undermine learning. She said the company released ChatGPT Study Mode early, using Socratic methods to guide learners without giving them the answer.

Kumar said there is no clear global model for AI safety regulation and that the information gap between technology companies and policymakers is widening. He called for non-prescriptive policies and cited the UK online safety law’s child-safety norm. He said India’s technology laws rely heavily on safe harbour, holding platforms accountable for third-party content when they fail to remove content they have been asked to take down, and argued that this approach does not fit the current technology ecosystem.

At the TASI 2026 Festival in New Delhi on Wednesday, October 14, executives from OpenAI, Google and Roblox, policy researchers and safety consultants debated AI safety, child protection and regulation. Panelists said AI is advancing faster than safeguards and discussed deepfakes, privacy risks from camera-equipped smart glasses, age verification and gaps in India’s regulatory framework. The two-day conference was jointly organised by the Centre for Social Research and TASI to discuss AI governance and online safety.

Participants debated whether companies can manage risks through testing and voluntary cooperation or whether governments should set clearer expectations before AI models and tools reach the public. Topics included red-teaming (testing systems with deliberate attacks), sharing safety signals across the industry through SynthID and hash matching, safer defaults for children, age prediction, and evaluations that account for India’s multicultural context. OpenAI’s Pragya Misra said the company had slowed reinforcement learning and training runs because it considers safety a priority, while also seeking to reach AGI first and have it benefit humanity.

Rohit Kumar, founding partner of The Quantum Hub, argued that competition can weaken companies’ incentives to invest in trust and safety, despite the value of user trust. He called for early, flexible regulation that sets clear outcomes and closer cooperation among governments, technology companies and researchers. Google’s Snigdha Bhardwaj said the company has zero tolerance for child sexual abuse material and may slow a model launch if it produces such content. She said Google continuously uses red-teaming to test harmful outputs and urged platforms to share risk signals through an interoperable safety layer, while addressing privacy and intellectual-property concerns. The panel also raised controlled testing of smart glasses before wider release.

Roblox’s Nicky Jackson Colaco said child safety should be the default, rather than depend on parents configuring controls. She described age-based limits on communication and content and said stronger safeguards at one company could encourage others to follow. Akash Pugalia of TP warned that children may rely on Gemini or ChatGPT for answers and raised concerns about critical thinking, grooming and convincing audio and video deepfakes. Misra said OpenAI’s research team found learning was a key use of its models among young people in India last year, but users said direct answers could undermine learning. She said the company released ChatGPT Study Mode early, using Socratic methods to guide learners without giving them the answer.

Kumar said there is no clear global model for AI safety regulation and that the information gap between technology companies and policymakers is widening. He called for non-prescriptive policies and cited the UK online safety law’s child-safety norm. He said India’s technology laws rely heavily on safe harbour, holding platforms accountable for third-party content when they fail to remove content they have been asked to take down, and argued that this approach does not fit the current technology ecosystem.