India's AI Crossroads: Crafting Our Own Safety Path

The global conversation around artificial intelligence safety is truly heating up, and you can feel the urgency in the air. Just recently, the United States made headlines with its White House accord on super intelligence, a voluntary pact signed by tech giants like Google, OpenAI, and Meta. It’s a significant step, sure, but here in India, our own industry leaders are making it abundantly clear: we can’t just cut and paste someone else’s homework. Instead, there's a powerful and growing consensus that India absolutely needs to forge its very own, custom-fit AI safety framework, one that truly understands our unique societal fabric and technological landscape.

It’s not just about being different for the sake of it; there are very real, pragmatic reasons. Amit Khanna, a key figure at Grant Thornton Bharat, put it perfectly, emphasizing that any commitment needs to be "suited to Indian market conditions." Think about it: our focus here might be more on AI applications that touch everyday lives, rather than just the foundational models themselves. We’re talking about ensuring AI systems are reliable across a myriad of vernacular languages and cultural contexts – a far cry from a one-size-fits-all solution. And let's not forget the "AI knowledge asymmetry" issue, where understanding and access to cutting-edge AI aren't evenly distributed. A tailored approach is simply non-negotiable.

Now, when it comes to how these rules should look, that's where things get really interesting, with a healthy debate brewing among the experts. Srikanth Velamakanni, who chairs Nasscom and leads Fractal Group, doesn’t mince words. He views purely voluntary approaches as "woefully insufficient." For him, it’s about binding rules – the kind that mandate transparency, clear incident reporting, and undeniable accountability. He envisions India's domestic commitment not just as an internal safeguard but as a vital building block for a broader global understanding of AI safety.

But others see a more nuanced path. Soumen Mandal from Counterpoint Research, for instance, suggests we can certainly draw inspiration from the US model, but the real push needs to be on accelerating our own efforts to build a truly trusted AI ecosystem here at home. He, much like Amit Khanna, leans towards a "hybrid approach." Imagine this: mandatory rules for the high-stakes areas, like defence, healthcare, banking, or government services, where failure simply isn't an option. Then, for the faster-moving technical domains, perhaps more agile, voluntary standards would make sense. Khanna even calls this a "techno-legal" approach, which sounds just about right given the complexities involved.

However, not everyone is convinced that mandatory regulations are the best way forward for everything. Ganesh Gopalan, co-founder of Gnani.ai, raises a really important point about ensuring a level playing field. He argues quite compellingly that global frontier models, those developed by the big players, should absolutely be subject to the very same rules as Indian-developed ones. He personally favors self-regulation, expressing a valid concern that overly strict, mandatory regulations could inadvertently stifle healthy competition and perhaps even lead to preferential treatment for certain established companies, which nobody wants to see happen in a rapidly evolving field like AI.

It’s clear the Indian government is also very much attuned to these discussions. Earlier this year, IT Secretary S Krishnan hinted that our existing legal frameworks might well need a significant overhaul, noting that "the time is getting right" for additional AI-specific regulation or even a new law. Echoing this sentiment, IT Minister Ashwini Vaishnaw has also acknowledged that the current information technology law was drafted long before AI truly took off, underscoring the potential need for a fresh legal framework to tackle this new frontier.

Beyond the philosophical debates, some very tangible, pressing concerns weigh heavily on the minds of experts here in India. Deepfakes, for example, pose a monumental threat, especially when mixed with misinformation and political manipulation. We're also seeing a rising tide of cyberattacks specifically targeting financial institutions. And then there's the massive elephant in the room: liability. Who is responsible when an AI system fails? Imagine a misdiagnosis by an AI in healthcare, or an erroneous denial of benefits – these aren't theoretical problems; they’re very real scenarios where clear liability rules are, frankly, a critical gap that desperately needs to be filled.

So, what's the path forward? Experts are advocating for a truly collaborative and inclusive approach. They suggest that any Indian AI pact should bring together a diverse array of stakeholders: the government, naturally; the frontier model developers; large deployers (think big enterprises and public agencies); academia; independent safety researchers; workers; users; and critically, civil society organizations. Strengthening institutions like the IndiaAI Safety Institute, perhaps by empowering it to test models without creating unnecessary bottlenecks, could be a crucial step in this direction. It’s all about creating an ecosystem of trust and responsibility, uniquely suited to India's journey.

Ultimately, the message is loud and clear: India isn't looking to simply mimic foreign models. We're at a pivotal moment, poised to craft an AI safety framework that is deeply rooted in our own realities, addresses our specific vulnerabilities, and harnesses our potential responsibly. It’s a complex, challenging, but ultimately vital endeavor to ensure that AI serves all of India, safely and equitably, as we navigate this exciting new era.