The U.S.’s abrupt decision to restrict access to AI firm Anthropic’s Mythos and Fable large language models is leading an “I told you so” moment for national security hawks within the government, who have dealt for months with skepticism and sparse resources in supporting sovereign efforts to develop an Indian AI stack.
“Globalisation is dead, and Bharat must find its own way ahead,” declared Sridhar Vembu, the founder of Zoho and a member of India’s National Security Advisory Board (NSAB), on X shortly after Anthropic announces the export controls. “What can our government do right now? Ensure that organisations in India embrace smaller models, both Indian and Chinese open-source ones. With a bit of effort, we can make them work. Anyway, why pay money to people who don’t even want to sell to you?”
The U.S. government ordered Anthropic to disable Mythos and Fable access for non-U.S. nationals, even within its own company, on Friday evening (early Saturday morning in India). Mythos is a model that Anthropic claims has been highly accomplished in finding and patching cybersecurity vulnerabilities that even human researchers have missed over decades.
Amid concerns that AI-generated cyberattacks may be a threat to Indian companies and government entities, India sought access to Mythos, and some entities joined the so-called Project Glasswing to gain some access earlier this month. That access may now be disrupted.
The Ministry of Electronics and Information Technology (MeitY) and the Ministry of External Affairs, which have been in touch with Anthropic and the U.S. government regarding Project Glasswing, did not respond to queries from The Hindu. Additionally, the Indian Cybercrime Coordination Centre (I4C), which reportedly gained access to Mythos this month, along with the cybersecurity agency CERT-in, which operates under MeitY, also did not respond.
Even some less cybersecurity-focused projects are experiencing disruptions due to the U.S. move. The restriction of Fable, which has been available for all paying users of Claude, had immediate fallout on Saturday. For instance, Vikram Chandra, an entrepreneur and journalist, said on X, “I have projects that were to run on Fable today – and they will come to a grinding halt… Yes, guardrails for frontier AI are essential – and Anthropic itself has argued for them. But creating national barriers isn’t the solution.”
India’s capabilities to train a frontier AI model lag behind those of China, which itself is a few steps behind the U.S. While Beijing is home to firms like DeepSeek, which use slightly older graphics processing units (GPUs) in large quantities, with abundant access to data centre capacity and electricity, to try to catch up to the more efficient U.S. models made by firms like Anthropic and ChatGPT, India has a relatively limited capacity of such resources to train large language models (LLMs) like Claude’s Opus, Mythos, and Fable.
While a hypothetical Indian alternative to Mythos wouldn’t be subject to abrupt geopolitical interruptions, actually creating such an LLM is dependent on the availability of abundant quantities of expensive AI chips from firms like Nvidia, data centre capacity, and electricity availability to boot. Since the costs run into the tens of billions of dollars, sovereign AI proponents’ ambitions have been limited. Even Mr. Vembu, who has been critical of over-reliance on foreign tech and whose firm has been appointed to run email inboxes of Union Government employees, struck a realist note on the potential.
“We must deepen our R&D,” Mr. Vembu said. “Sarvam has been on it and we have been on it, but remember that the latest models cost not only huge GPU budgets to train; the GPUs themselves are restricted. So we can’t afford the scale of money (of the order of $100+ billion to even get in the game!), and even if we could come up with the money, we can’t get all the GPUs. I would not like to ask the government to fund tens of billions of dollars on this anyway — the money has far better uses.” He advocated lower-cost research like what Zoho was undertaking. The firm recently announced an indigenously developed server.
“We are way behind and need a national mission to get going quickly,” T.V. Mohandas Pai, a former CFO of Infosys who consults closely with the government on technology-related issues, said on X. “Existing government programs are too slow, way too small to make any large impact. We need an annual ₹50,000 crore fund for deep tech and AI, a ₹200,000 crore ELGS [Emergency Credit Line] Guarantee Fund to build Hyper cloud, hardware and chips.”
At least one firm has seen some progress, however: Bengaluru-based Sarvam AI launched a 105 billion parameter LLM that is specifically trained with an Indian bias (to counter the U.S.-centricity in most LLMs). While this model is capable of coding tasks, it is far from a frontier-class model for ambitious cybersecurity work.
“A few Indian friends of mine who had access to the Mythos model told me genuinely how terrifying it was,” said C.S. 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 to anything and it does uncover vulnerabilities unlike anything they have seen before.” While responses to Mythos are occasionally tempered by less advanced LLMs discovering similar software bugs, its true potential remains gatekept to Project Glasswing members, of whom there are only a few dozen, most of whom are in the U.S.
Since these capabilities can be used by both defenders and attackers, Anthropic said it created Fable with “overbroad” protections that guarded against misuse. The U.S. government informed the firm that its export control was imposed after it was alerted to a potential “jailbreak” of these safeguards. The firm said that there was no “universal” way to deceive the model and considered the U.S. move a “mistake”.
Published – June 13, 2026 12:16 pm IST
Leave a Reply