US Orders Anthropic to Suspend Frontier AI Models for Foreign Nationals

TOI correspondent from Washington: In a move that could redefine the global artificial intelligence landscape, the Trump administration on Friday abruptly ordered the AI firm Anthropic to suspend access to its newest frontier models -- Fable 5 and Mythos 5 -- for all foreign nationals, regardless of where they are located, including non-American employees working in the US. The directive, issued under US export-control rules by authorities citing national security concerns, effectively forced Anthropic to pull the models for everyone overnight. The company, which like most tech firms employs many foreign nationals including Indians, said it could not reliably distinguish between American and foreign users quickly enough to comply with the order. "The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance," Anthropic said, while insisting it believed the government's action stemmed from a "misunderstanding" and expressing hope that access would be restored. The episode marks a dramatic escalation in Washington's efforts to treat cutting-edge AI capabilities much like advanced semiconductors -- technologies whose distribution is controlled because of their potential military value. At the centre of the controversy is Mythos 5, a powerful cybersecurity-focused model that Anthropic had previously restricted to a select group of trusted organisations, including some in India, through an initiative known as Project Glasswing. The model was designed to identify software vulnerabilities at a scale and speed beyond most human researchers, potentially helping governments and companies defend themselves against cyberattacks. But the same capability that can patch vulnerabilities can also help identify ways to exploit them. India’s inclusion in Project Glasswing, announced just two weeks ago, was an unusually significant gesture given that New Delhi is not a formal US treaty ally. It was the only major non-aligned power included in a group of 15 countries including Japan, Australia, Germany, Canada and South Korea. The decision was widely interpreted as recognition of India's vast software talent pool and growing strategic importance in the AI race. That access now appears uncertain, reminding New Delhi of the risks of relying on foreign-controlled AI infrastructure, and intensifying calls for "sovereign AI." In a social media post, Zoho founder Sridhar Vembu described the development as proof that "technology is the ultimate weapon" and argued that "globalization is dead." He urged Indian organisations to adopt smaller Indian and Chinese open-source models where appropriate while deepening indigenous research efforts. Meanwhile, the US government order also has an ironic twist: it affects some of the very people helping build these systems. Anthropic employs a significant number of Indian-origin engineers and researchers, including Chief Technology Officer Rahul Patil, formerly the CTO of Stripe. The company also maintains operations in Bengaluru as part of its broader international expansion. Under the directive's interpretation of "foreign nationals," non-US citizens within Anthropic itself could potentially lose access to the models they helped develop. Anthropic has pushed back forcefully against the government's action, arguing that the alleged security concerns surrounding Mythos involve narrow "jailbreak" scenarios and capabilities that comparable public models already possess. Company officials warn that if such findings become grounds for export restrictions, no frontier AI model may remain widely accessible. Other AI companies are watching closely. Executives across the industry fear the emergence of a fragmented AI ecosystem in which nationality determines access to the most capable systems. The concern extends beyond China, the traditional focus of US technology controls, to close American partners in Europe, Japan, Australia, and India. Analysts say the longer-term consequence could be the creation of a two-tier AI world: top-tier models reserved for Americans, while everyone else receives less capable versions or loses access altogether. Paradoxically, such restrictions could strengthen the hand of China, which, ironically for a so-called communist country, has offered open source access to its models. Beijing has invested heavily in indigenous AI development precisely to avoid strategic dependence on foreign technology. If American firms become unreliable suppliers of frontier AI, countries seeking technological independence may increasingly turn to Chinese open-source models or accelerate domestic alternatives. For ordinary users, the immediate disruption may be limited. Anthropic says access to its other models remains unaffected, while competing offerings from OpenAI and Google continue to operate normally. The bigger story, however, is that AI is increasingly becoming a geopolitical asset rather than merely a commercial product. For decades, software crossed borders effortlessly; frontier AI may not. And for countries such as India, the lesson is becoming difficult to ignore: in the age of artificial intelligence, technological self-reliance may no longer be an aspiration. It may be a strategic necessity.