New Delhi: India continues to tread cautiously on the issue of access to advanced artificial intelligence (AI) models, including Anthropic’s Fable 5 and Mythos 5, despite an assurance from the United States that trusted partners will not be cut off abruptly. A senior official of the Ministry of Electronics and Information Technology (MeITY) at an informal briefing Tuesday to mark 11 years of the Digital India programme declined to fully endorse the US assurance, instead laying out a fallback strategy built around open-source and homegrown models to shield Indian researchers from “arbitrary restrictions”.
The official was responding to a question on whether New Delhi trusted a US commitment made on the sidelines of the second Pax Silica summit in Washington last Thursday.
Asked whether the Fable model was discussed with India, US Under Secretary of State for Economic Growth, Energy and the Environment Jacob Helberg was quoted by ANI as saying, “We continue to have ongoing conversations about this topic with our Indian friends. These are very sensitive national security discussions that are not quite right for public consumption.”
The senior MeITY official, who did not want to be named, said that the wording needs to be read closely. “What they assured was, if access is given, it will not be [withdrawn]… It is when access is given. I’m saying—what did the US side say? This is what the US side said.” He declined to spell out the basis for the assurance.
The official’s comment comes in the backdrop of Washington suspending Anthropic’s Fable model under an export control order.
The Anthropic trigger
The US Commerce Department’s Bureau of Industry and Security on 12 June directed AI firm Anthropic to suspend access to its two most advanced models—Fable 5 and Mythos 5—for all foreign nationals, whether inside or outside the United States, citing national-security and cybersecurity concerns. Since the company could not reliably separate users by nationality, the practical effect was a worldwide shutdown of both models within hours.
This marked the first time that export-control law was used to restrict access to an AI model rather than to chips, and it landed hard on Indian infotech firms and startups that have built workflows atop foreign frontier models.
The episode has reignited a debate in India over sovereign AI, with industry voices split between those urging investment in homegrown frontier models and others—including Infosys co-founder Nandan Nilekani, who chaired the body that built the Aadhar identity system and Tata Sons Chairman N. Chandrasekaran—arguing that India’s edge lies in building AI applications and sector-specific solutions atop foreign models.
Asked what New Delhi is doing to insulate Indian researchers from such curbs in the future, the official described a two-part approach. While a frontier model may be fully capable, he said, open-source and Indian-built models were already some 60 to 80 per cent as capable—enough to do a fair amount of work.
The second leg, he said, was the India AI Mission — the government’s flagship AI programme, backed by a Rs 10,372-crore outlay, and offering subsidised compute to startups and researchers with a mandate is to back the development of Indian foundation models and push them ever closer to the frontier.
The official was sympathetic to Washington’s stated rationale, noting the technology’s reach. He said that because what such technology can do is far-reaching, it could cause serious problems if it fell into the wrong hands. The risk is especially acute in cybersecurity, where AI can be used to probe vulnerabilities in software code. The restrictions, he added, were “presumably within US law”.
Pax Silica backdrop
The caution sits awkwardly alongside India’s deepening technology engagement with the US. New Delhi is a member of Pax Silica, the American-led initiative aimed at building trusted, China-independent supply chains across semiconductors, AI and critical minerals.
India joined in February on the sidelines of the AI Impact Summit in New Delhi, and the framework’s second summit was held in Washington last week.
The Anthropic curbs have sharpened the tension running through that partnership: even as New Delhi signs up for trusted technology ecosystems with Washington, access to the most powerful American models can vanish on a single US directive.
On the related question of whether the government should take a sovereign stake in companies building strategically critical AI models — an idea being floated in the US — the official said that debate abroad was being driven by a different set of concerns, around AI displacing jobs and the need for wealth redistribution. “We are a long way from that discussion,” he said.
(Edited by Sugita Katyal)
Also Read: A human firmly in the loop: RBI proposes ‘kill switch’ for AI used by banks & NBFCs