Project Glasswing melts: US government suspends early access to Anthropic's Fable 5, Mythos 5 within days of rollout

India’s access to one of the world’s most advanced artificial intelligence (AI) models appears to have lasted barely a few days.

After Anthropic expanded access to its closely guarded Claude Mythos system to a handful of Indian organisations, a US government directive led the company to suspend local access to Mythos 5 and Fable 5, raising questions about whether frontier AI models are becoming geopolitical assets similar to advanced semiconductors that remain the exclusive preserve of a chosen few.

The Centre, an official told ET,would continue to engage with both Anthropic and Washington to understand the scope of the restrictions and whether they are limited to Mythos 5 and Fable 5 or could extend to other models in the Mythos family.

“It’s to be seen if the latest directive is only for Mythos 5 and Fable 5, or will it also impact the other models in the Mythos class, going forward. We hope that’s not the case,” said the official cited above.

The significance of this suspension lies in how Anthropic rolled out the models. Mythos 5, among the company’s most advanced AI systems, was initially kept under restricted access and made available only to a small group of researchers, cybersecurity experts and trusted partners because of its powerful cyber and reasoning capabilities.

Through its Project Glasswing cybersecurity initiative, Anthropic gradually expanded access to select organisations. It then launched Fable 5, a more widely accessible version, just a few days ago.

Now access to both these models has been suspended following the US directive.

The development is particularly significant for India. Several organisations across cybersecurity, telecom, finance and banking had secured early access to Mythos under Project Glasswing, ET had reported last week.

Also Read: Anthropic’s Fable 5 takedown triggers India Inc push for AI self-reliance

Made the cut

While the number of Indian entities granted access was in the single digits, officials described it as a key win in India’s continued engagement on technology policy. Public-sector entities, including the Indian Computer Emergency Response Team (CERT-In), were also expected to receive access.

Beyond the immediate impact on those organisations, experts say the episode could mark the beginning of a broader shift in how governments approach advanced AI.

“The development signals the emergence of a new AI export-control regime. And the pace is striking,” said Subimal Bhattacharjee, an independent tech policy analyst. “Washington moved from chip controls to controlling the models themselves. What took years to build for semiconductors is being compressed into months for AI.”

According to Bhattacharjee, the Mythos-to-Fable-to-suspension sequence shows that governments increasingly view frontier AI systems as strategic assets rather than commercial products.

“The lesson is that states and labs urgently need a harmonised, predictable framework of enablement rather than abrupt, ambiguous shutdowns that disrupt access globally,” he said.

Access Control

While cybersecurity concerns have been cited as the trigger, the move reflects a broader geopolitical strategy, Bhattacharjee argued.

“The order asserts a broader power, not just to scrutinise model safety, but to restrict who may access frontier AI at all when officials judge the risk to be national-security related,” he said. “Combined with the chip-tier system and model-weight licensing, it fits a deliberate geopolitical strategy to control who can reach the technological frontier.”

Cybersecurity experts say concerns around Fable 5 are less about entirely new capabilities and more about the speed and scale at which such systems can operate.

"Fable 5 introduced long-horizon autonomy and proactive self-verification. When an AI can compress months of human penetration testing into a matter of hours, the sheer velocity of vulnerability discovery alters the risk landscape,” said Neehar Pathare, chief executive of 63SATS Cybertech, a cybersecurity company.

However, Pathare cautioned that restricting access may not be enough to contain such capabilities. “The belief that a centralized government directive can permanently restrict access to advanced AI capabilities is a regulatory fantasy,” he said, highlighting the rapid progress of open-source models.

For India, this episode raises questions that extend beyond access to a single AI model. Bhattacharjee said the restrictions are an early signal that access to frontier AI could become as strategically important as semiconductors. As governments increasingly seek to shape who can access the world’s most advanced AI systems, the development is likely to strengthen the case for sovereign AI and domestic capability building.