AI sovereignty hawks see red as US prepares to block Anthropic’s Mythos and Fable models

The US’s sudden decision to limit access to AI company AI’s large Mythos and Fable language models leads to an “I told you so” moment for national security hawks in the Indian government, who have struggled for months with skepticism and scarce resources to support sovereign efforts to develop India’s AI suite.

“Globalization is dead and Bharat must find its own way forward,” Sridhar Vembu, founder of Zoho and a member of India’s National Security Advisory Board (NSAB), said at X shortly after Anthropic announced the export controls. “What can our government do right now? Make sure organizations in India adopt smaller models, both Indian and Chinese open source. With a little effort, we can make them work. Anyway, why pay money to people who don’t even want to sell to you?”

This is a big one: all access to the Mythos and Fable AI models is off limits to anyone outside of America.

First thoughts:

1. Technology is the ultimate weapon. National sovereignty, national security, it’s all about technology now.

2. Globalization is dead and Bharat must find it… https://t.co/kCQpq93D3r

— Sridhar Vembu (@svembu) June 13, 2026

The US government ordered Anthropic to ban access to Mythos and Fable to non-US nationals, including within its own company, on Friday night (early Saturday morning in India). Mythos is a model that Anthropic claims has been highly successful in finding and patching cybersecurity vulnerabilities that have eluded even human researchers for decades.

Amid concerns that AI-generated cyberattacks could pose a threat to Indian companies and government entities, India has sought access to Mythos, with some entities joining the so-called Project Glasswing to gain some access earlier this month. This access may now be discontinued.

The Ministry of Electronics and Information Technology (MeitY) and the Ministry of External Affairs, which have been in touch with Anthropic and the US government regarding the Glasswing project, did not respond to The Hindu’s queries. In addition, India’s Cybercrime Coordination Center (I4C), which reportedly gained access to Mythos this month, along with cyber security agency CERT-in, which operates under MeitY, also did not respond.

Even some projects less focused on cyber security are experiencing disruption due to the US move. The Fable restriction, which was available to all paying Claude users, took immediate effect on Saturday. For example, Vikram Chandra, an entrepreneur and journalist, said in X: “I have projects that were supposed to run on Fable today – and they’re going to get stuck… Yes, barriers for border AI are essential – and Anthropic itself has argued for them. But creating national barriers is not the answer.”

India’s capabilities to train a frontier AI model lag behind those of China, which is itself several steps behind the US, while Beijing is home to firms such as DeepSeek, which use slightly older graphics processing units (GPUs) in large numbers, with abundant access to data center capacity and electricity, to try to catch up with the more efficient US models created by firms such as Anthropic and ChatGPT, with relatively limited training capacity like India’s language sources such as India. Claude’s opus, myth and fable.

While a hypothetical Indian alternative to Mythos would not be subject to sudden geopolitical disruptions, in reality the creation of such an LLM depends on the availability of large quantities of expensive AI chips from firms like Nvidia, data center capacity and electricity availability. With costs running into the tens of billions of dollars, the ambitions of sovereign AI advocates are limited. Even Mr Vembu, who has criticized over-reliance on foreign technology and whose firm has been tasked with running union government staff email boxes, struck a realistic note about the potential.

“We need to deepen our research and development,” Mr Vembu said. “Sarvam was on it, and we were on it, but remember that the latest models not only cost huge GPU budgets to train, but the GPUs themselves are limited. So we can’t afford that amount of money (on the order of $100+ billion to even get into the game!), and even if we could get that money, we can’t get all those GPUs on it. I wouldn’t want to ask the government for those ten billion for that. it has much better uses.” He advocated lower-cost research like what Zoho is doing. The company recently announced a self-developed server.

“We are way behind and we need a national mission to get going quickly,” TV Mohandas Pai, former CFO of Infosys, who consults closely with the government on technology-related issues, told X. “Existing government programs are too slow, too small to have any big impact. We need an annual fund of ₹ 50,000 crore for deep technology and AI (Ecroergency 2 Credit Line) ₹ 0.00 Guarantee Fund for building hyper cloud, hardware and chips.”

However, at least one firm has made some progress: Bengaluru-based Sarvam AI has launched an LLM with 105 billion parameters that is specifically trained with an Indian bias (to counter the American bias in most LLMs). While this model is capable of coding tasks, it is far from a frontier model for ambitious cybersecurity work.

“Several of my Indian friends who had access to the Mythos model actually told me how scary it was,” said CS Akshay, a cybersecurity researcher who has uncovered several vulnerabilities in Indian government and private websites in the past, in a LinkedIn post about Mythos. “You just point it at anything and it reveals a vulnerability unlike anything they’ve seen before.” While the backlash against Mythos is occasionally tempered by less advanced LLMs discovering similar software bugs, its real potential remains with Project Glasswing’s few dozen members, most of whom are in the US.

Since these abilities can be used by both defenders and attackers, Anthropic said it created Fable with “overbroad” protections to protect against abuse. The US government informed the company that its export controls had been imposed after it was alerted to a potential “jailbreak” of these safeguards. The firm said there was no “one-size-fits-all” way to fool the model and called the US move a “mistake”.

Published – 13 Jun 2026 12:16 IST