India should build its own artificial intelligence (AI) safety framework suited to its languages, scale, and application-led AI ecosystem rather than replicate a voluntary pact signed by US President Donald Trump and top technology firms last week, industry experts said.They also called for a commitment that brings together all stakeholders and tackles risks such as deepfakes and cyberattacks, even as they differed on whether voluntary measures alone would be enough.
The comments came after the White House Accord on Super Intelligence, a voluntary AI safety pact signed on September 29 by Trump and top executives from Google, OpenAI, Anthropic, Meta, Nvidia and Elon Musk’s xAI.
Under the accord, the companies committed to internal controls to monitor their AI models, dedicated oversight teams, independent external audits and board-level oversight. They also agreed to meet regularly to set safety standards.
Nasscom Chairman and Fractal Group CEO Srikanth Velamakanni said India should not simply replicate the US. Instead, it should use a domestic commitment as a building block towards a global understanding of AI safety.
“As models become more capable and agentic, we are giving them greater access to systems and greater autonomy to make decisions. AI safety must be strengthened considerably and must stay ahead of capability to prevent catastrophic risks,” Velamakanni said.
He noted that the US accord is largely a push for self-policing by frontier AI companies. The European Union, by contrast, has taken a binding, risk-based approach, while China follows a state-directed control model.
India, the Nasscom Chairman said, could use its role in the Global Partnership on AI (GPAI) and the momentum from hosting the AI Impact Summit in February to prevent worst-case outcomes, without citizens of India and the Global South being denied access to frontier models.
“India doesn’t need to copy the US model, but it can take inspiration from it and expedite efforts to build a trusted AI ecosystem,” said Soumen Mandal, Principal Analyst at Counterpoint Research.
He added that such a commitment should cover testing AI systems before deployment, independent audits, reporting serious problems, and naming people responsible for AI-related risks.
Amit Khanna, Partner and Automation Ecosystem Leader at Grant Thornton Bharat, said any commitment should be suited to Indian market conditions. He noted that India’s focus is on building AI applications rather than major foundational models.
“An industry commitment in India should ensure AI works reliably across vernacular languages and unique cultural contexts and should not exploit AI knowledge asymmetry in the population,” Khanna said. Ganesh Gopalan, co-founder and CEO of Gnani.ai, said any framework must ensure a level playing field.
Global frontier models should be subject to the same rules as Indian ones, he said, since Indian AI companies compete with global firms for the same customers. The experts differed on whether self-regulation would be enough. “Voluntary, self-regulatory approaches are necessary but woefully insufficient,” Velamakanni said.
“Pharmaceuticals and aviation are great examples of industries where strong, technically competent regulators have strengthened safety and public trust without stopping innovation.” He said India needs binding rules on transparency, incident reporting, and accountability when serious harm occurs.
The IndiaAI Safety Institute should be strengthened to test newly released models, but without creating a bottleneck by requiring pre-release certification of every model, he added. Mandal backed a hybrid approach. Mandatory rules would apply to potential harms in defence, healthcare, banking, and government use cases, while voluntary standards would cover fast-moving technical areas where regulators may lack technical expertise.
“This type of hybrid approach will provide flexibility to AI companies instead of restricting every line of code while ensuring safety and privacy are not compromised,” he said.
Khanna also called for a hybrid “techno-legal” approach. It would pair a strong voluntary layer with binding mandates, such as the provisions on synthetically generated information in the IT Amendment Rules, 2026. He said voluntary compliance alone leaves significant gaps, particularly in civil rights and public safety.
Gopalan, however, favoured self-regulation. “Voluntary self-regulation is key given the technology is new. Having mandatory regulation can stifle competition and lead to concerns about a few companies getting preferential treatment.
“India needs more than 100 frontier AI lab companies, and that’s the need of the hour to prevent India from being colonised, which is inevitable if we lag behind in the AI race,” Gopalan said.
The Indian government is in the process of bringing in its own AI regulation. Earlier this year, IT Secretary S Krishnan noted that while existing legal provisions have so far been adequate in addressing initial concerns on issues like deepfakes and AI-generated synthetic content, an “additional regulation or law may be needed”.
“It is a conversation which has commenced, and my minister (IT Minister Ashwini Vaishnaw) and I have both been on record earlier that we will look at AI regulation when the time is right, and it appears that the time is getting right, and we will start looking at it,” he had said.
Vaishnaw had also said that the current information technology law was framed much before the rapid emergence of Artificial Intelligence (AI) and that a new legal framework may be required to deal with the changing landscape. The industry agrees that any pact should go beyond AI companies.
“This cannot be an agreement among AI companies alone,” Velamakanni said. He said the government, frontier model developers, large deployers, including enterprises and public agencies, academia, and independent safety researchers should all have defined roles. Workers, users, and civil society should have formal channels to report harms.
He added that safety evaluations of frontier models should be carried out by independent, expert institutions and published, with narrow exceptions where disclosure itself creates a security risk. Mandal said the agreement should bring together the entire AI ecosystem, from AI companies and their business users to regulators, universities, researchers, workers, and users.
Khanna argued that it should include government and public sector bodies, which would be major users of AI, as well as civil society, worker unions, and technology companies. Meanwhile, Gopalan called for broader self-regulation across the AI ecosystem to ensure impartial access to limited compute resources. Velamakanni said the top concern must be preventing catastrophic and systemic harms.
He listed large-scale cyberattacks, electoral interference, biological or chemical weapons, attacks on critical infrastructure, and loss of control over highly autonomous systems. Individual harms such as deepfakes, fraud, impersonation, privacy breaches, and self-harm come next.
“Autonomy must lag behind assurance to provide a margin of safety. As AI systems gain the ability to act in the world (use tools, move money, change systems, make consequential decisions), the evidence that they are safe and controllable must grow faster than the autonomy we give them,” he said.
Gopalan said that given India’s large population and democratic framework, the bigger concerns are deepfakes and cyberattacks targeting financial institutions. Khanna also named deepfakes as the top immediate priority. He described misinformation, deepfakes, and political manipulation as “existential threats to India’s social fabric”.
He said rules on liability when AI systems fail, such as misdiagnosing a patient or wrongly denying welfare benefits, remain a critical gap. Mandal placed independent audits first, followed by accountability, data protection, deepfakes, and job impact. “Without independent testing and audits, the system will not be complete, as companies may sometimes overlook AI-related concerns,” he said.
Disclaimer : This story is auto aggregated by a computer programme and has not been created or edited by DOWNTHENEWS. Publisher: deccanchronicle.com
