On the fourth day of the India AI Impact Summit in New Delhi on Thursday, Union IT Minister Ashwini Vaishnaw unveiled the ‘New Delhi Frontier AI Impact Commitments’, a voluntary set of guidelines pushing signatories toward measurable outcomes on inclusion and responsible development.
“We are shaping an AI future of the human, by the human, for the human…Together, these efforts mark an important step towards shaping AI that is not only powerful but also inclusive, development-oriented and globally relevant,” Vaishnaw said in his keynote as India sought to position itself at the center of AI governance, keeping in view the perspective of the Global South.
The framework landed as Prime Minister Narendra Modi laid out India’s core anxiety about the technology: that in the age of AI, humans risk becoming little more than raw material. “For AI, humans are just data points,” Modi warned, calling for AI to be democratised and made “a medium for inclusion and empowerment, especially in the Global South.” His message: “The direction in which we take AI today will determine our future.”
The event was attended by about 20 heads of state from across the globe, alongside policymakers, industry leaders, researchers, and startup founders. Day Four also marked keynote addresses from Alphabet and Google CEO Sundar Pichai; OpenAI CEO Sam Altman; Anthropic co-founder and CEO Dario Amodei; French President Emmanuel Macron; Reliance Industries chairman Mukesh Ambani, among others. Bill Gates, billionaire and co-founder of Microsoft, however, pulled out of the summit hours before the keynote address.
The document, at the India AI Impact Summit, outlines specific outcomes that signatories are expected to work towards, with two commitments detailed on the first day.
data transparency push
The framework’s first commitment is on data and transparency. Signatories are expected to publish statistical insights on global AI adoption drawn from anonymized, aggregated usage data before the next AI summit. The goal is to map where and how AI is actually being deployed across the global economy, and feed that evidence into policymaking on workforce transition and education. It’s also meant to set a baseline for tracking how adoption shifts across regions over time.
However, this might disproportionately burden smaller Indian AI startups as they “already face ₹₹2-10 lakh for even small dataset preparation, with data quality work often exceeding model development costs. Creating anonymised, aggregated, taxonomised usage data requires significant engineering resources that smaller companies lack,” said Abhivardhan, president, Indian Society of Artificial Intelligence and Law, an industry forum.
Language gap and new AI rules
The second commitment in the framework addresses the language gap — a focus that also dominated the agenda for global tech CEOs, including Altman, Amodei and Pichai.
“We need to make AI as effective in every language as it is in English, and today it is not,” said Brad Smith, president and vice chair of Microsoft, which has announced $50 billion to bridge the global AI divide. “Performance tests show that’s the case,” he added, calling for investment in better data, tools and measurement systems for languages beyond English.
“Across India, creators use our AI to automatically translate their reels into the language of the person watching,” said Alexander Wang, chief AI officer at Meta — a signal that multilingual AI is already finding real-world traction on the ground, even as the policy frameworks around it are still being built.
The framework itself points to this: “Participating organizations recognize that cross-lingual support is helpful for democratizing AI and aspire to improve AI performance and high-quality experiences for users across the globe.”
AI systems have long performed unevenly across languages, with most benchmarks skewed toward English. Signatories are expected to assess multilingual capabilities across languages and cultural contexts, and work with governments, researchers and civil society to build datasets suited to local use cases.
Though the framework preserves flexibility over which tools, benchmarks and languages participants prioritize, “which lets companies self-select the easiest evaluations, as [these benchmarks] by design do not address deeper technical approaches due to trade secret risks & systemic limits of large language models,” said Abhivardhan. Current AI evaluations are too narrow, measuring speed and efficiency while missing what actually matters: how well these systems work for real people in real contexts, he said.
funding push
For context, under the IndiaAI mission, the government has allocated ₹199.55 crore for the datasets platform, accounting for 1.92% of the total ₹10,372 crores.
Earlier this month, India notified the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, a significant regulatory move that brings AI-generated content under a formal legal framework for the first time.
The rules formally define “synthetically generated information,” covering audio, visual or audio-visual content created or modified using a computer resource in a way that appears real. Platforms are required to remove flagged illegal content, including deepfakes, within three hours of a government or court order, and must label AI-generated content.
