US govt's curbs on Anthropic models reignite calls for sovereign AI

The US move to restrict access to Anthropic's latest AI models has intensified concerns over AI sovereignty and technology dependence in India

The United States (US) government’s decision to suspend access to artificial intelligence company Anthropic’s latest AI models, Fable 5 and Mythos 5, for all foreign nationals has reignited concerns around AI sovereignty in India. 

It has disrupted developers, cybersecurity firms and enterprises that had started building products and services around these models. 

The Donald Trump administration’s order, issued late last week, comes just weeks after India was included in Anthropic’s cyber defence initiative, Project Glasswing. Previously, Indian government and industry representatives had engaged with the company to secure broader access to its frontier AI models. 

A senior official said efforts to restore access to the latest Anthropic models for India and Indians would be taken up at a diplomatic level as well. 

Companies, however, would have to decide how they want to deal with revenue impact that US government curbs are likely to have on their revenues, the official added. 

“The loss of access for all users outside may temporarily impact the work of respective enterprises and clients of Anthropic. But the long term impact will be on the revenues of the company. So, it is up to the companies to see how they can get out of this,” the official said. 

This ban comes at a time when the government is looking to substantially increase the budget proposed under the India AI Mission 2.0 as it looks to on-board more startups to work on foundational models and small language models (SLM) rather than be completely dependent on models from companies such as OpenAI and Anthropic. Four foundation models were launched under this initiative during the India AI Summit earlier this year. 

Industry players have been left in flux, with many now needing to recalibrate product and technology strategies built around Anthropic's latest models. 

“The US export control directive forcing the immediate global shutdown of Claude Fable 5 and Mythos 5 is a structural watershed. It marks the precise moment where advanced algorithmic intelligence stopped being treated as a commercial software product and became weaponised as an instrument of state sovereignty,” said Neehar Pathare, managing director (MD), chief executive officer (CEO) and chief information officer (CIO) of 63SATS Cybertech, the cybersecurity arm of 63 Moons Technologies. 

The sudden withdrawal of Mythos-class models also imposes an immediate ‘latency tax’ on Indian enterprise defence, according to Pathare. India’s highly digitised banking and fintech companies, he said, were actively designing automated, asynchronous testing pipelines around these models. By being stripped of access to these models, Indian enterprises are forced to revert to legacy static testing tools and manual code reviews, drastically widening the window of vulnerability remediation from hours back to weeks. 

Only last week, the country’s largest IT services player Tata Consultancy Services (TCS) had announced a strategic Partnership with Anthropic, which included 50,000 licenses for its staff, and the Claude deployment to the information technology (IT) firm’s popular products and platforms. 

Pankit Desai, cofounder and CEO of cybersecurity company Sequretek, said while Fable 5 would hardly make an impact, losing access to Mythos 5 is a setback since companies have to recalibrate their strategies. 

“The larger problem for us is being dependent on these models from other countries. They can stop access at any point of time. This is weaponisation of technology and a wake-up call that for frontier technology, we cannot be completely dependent on these models.” 

Etisha Garg, developer relations engineer at open source coding agent platform Cline, said restricting access based on nationality creates uncertainty for builders who rely on these models to create products. “Developers need stable and predictable access, because sudden restrictions can disrupt years of work. I believe regulations should focus on misuse rather than where developers come from. An outcome of such moves is that more developers may increasingly turn towards open-source models in the future, as they offer greater transparency, control, and independence from decisions made by a handful of companies.” 

Anthropic said on Friday that while it was complying with the US government’s orders and removing access, it disagrees with the reasons that threat actors can bypass or “jailbreak” Fable 5. “We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.” 

America’s move to bar Anthropic’s services to any non-US resident user or enterprise comes only about 48 hours after the company’s Founder and Chief Executive Officer, Dario Amodei had written and published an extensive blog detailing the kind of regulations for frontier AI models. 

In the blog, Amodei had called for the “need to activate a slow and rickety policy apparatus to deal with risks and opportunities that are going to compound surprisingly quickly” after the global release of Anthropic and its other models.