Fable 5, one of Anthropic’s most advanced AI systems released for broader use, disappeared almost as quickly as it arrived. What began as a controlled rollout quickly turned into a shutdown, and the reason had little to do with product decisions.
It had everything to do with who gets to control access to that capability.
What exactly happened
Fable 5 was introduced as a controlled release alongside Anthropic’s more powerful Mythos 5 model. While Mythos remained restricted to enterprise customers and vetted organisations under Project Glasswing, the Fable 5 was intended to bring a significant portion of that capability to a wider audience while retaining safeguards.
That rollout proved short-lived.
The US government issued a directive asking Anthropic to suspend access to both Fable 5 and Mythos 5 for foreign nationals, including those working within the US.
Anthropic complied, effectively shutting down access to the models globally.
As reported by TechCrunch, the move has had immediate implications across markets such as India, where developers and companies had only recently begun experimenting with the systems.
The decision did not come with a detailed public explanation, but multiple reports point to national-security concerns linked to the models’ capabilities.
How Anthropic has responded
Anthropic’s response has been measured, though the company has stopped short of endorsing the reasoning behind the move.
In its official statement, Anthropic said it was asked to suspend access without being provided detailed evidence of a severe threat. It also argued that the vulnerabilities cited were not unique to Fable 5 and could be replicated using other publicly available models.
The company pointed to safeguards already in place, including extensive red-teaming, tighter monitoring of model outputs and controlled deployment through its Glasswing framework.
There is also a broader argument embedded in Anthropic’s response. Frontier AI systems, by their nature, are not perfectly secure. Any model with meaningful capability will have edge cases and potential vulnerabilities. The question is not whether those risks exist, but how they are managed.
From Anthropic’s perspective, the shutdown shifts that decision away from the company and into the hands of the state.
Why this is happening now
The trigger appears to be a combination of technical and geopolitical concerns.
According to The Verge, US officials were concerned that Fable 5 could be used to identify software vulnerabilities, increasing the risk of misuse in cyberattacks. Reuters reported that Amazon had raised similar concerns, helping bring the issue to policymakers.
There is also a geopolitical dimension.
The Verge, citing Semafor, reported that the White House’s decision to impose restrictions on Anthropic’s Mythos programme was driven partly by concerns that the model had been accessed by a group linked to China. The report added that US officials were worried foreign actors could study or replicate the model’s capabilities through techniques such as model distillation.
Taken together, these concerns point to a broader shift in how advanced AI systems are viewed. They are no longer treated solely as commercial products. Their capabilities now intersect with cybersecurity, intelligence and strategic advantage.
That changes how governments respond to them.
What this changes
For years, software operated on a simple assumption: access would be global and continuous. As long as users followed the rules and paid for the service, they could rely on it.
That assumption, according to Ashish Tandon, founder and chief executive officer of Indusface, has now been challenged.
“A technology hundreds of millions already used, paid for and ran in production was revoked inside ninety minutes. What broke was the assumption behind two decades of digital globalisation: that the best technology, wherever built, stays available to anyone who licenses it,” he said in response to Business Standard queries sent to him over email.
The Fable 5 shutdown demonstrates that access to advanced AI systems can be shaped by policy rather than business decisions alone.
For companies, this changes how AI is viewed. It is no longer merely a capability layer. It is a dependency that carries geopolitical risk.
It also changes how these models are positioned. Increasingly, they are being treated as strategic assets, where access itself becomes a mechanism of control.
Most importantly, it draws a clearer distinction between using AI and having reliable access to it.
What this means for businesses in India
For Indian companies, particularly those embedding AI into critical workflows, the implications are immediate.
For enterprises, the disruption was not abstract.
“Teams lost access overnight, with no warning and no recourse, and found that one foreign government’s decision had become their outage,” Tandon said.
But the bigger issue is dependence.
Pratyush Kumar, co-founder and chief executive officer of Sarvam, pointed directly to this concern.
“You should not confuse access with ownership, or adoption itself as an advantage,” he said. If a company’s “most significant tech differentiator…has external control loops, then you have to accept you are vulnerable.”
Tandon argued that the risk extends beyond AI.
“The same concentration risk runs across the stack Indian business depends on, from cloud and data centres to payment rails…assume anything single-sourced can be withdrawn,” he said.
This changes how companies think about technology adoption. The conversation is no longer only about capability or cost. It is increasingly about control.
There is also a second-order effect on talent and alignment. Kumar noted that the episode sets a precedent where AI professionals may increasingly be “seen aligning to national interests more than company interests”, suggesting geopolitical considerations could influence talent flows over time.
Open, closed and sovereign AI systems
The shutdown also sharpens an ongoing debate within the industry: whether organisations should rely on closed, high-performance AI systems or invest in open-source and open-weight alternatives.
“The week forced this into the open, and it is not binary. Closed frontier models still lead on raw capability. Open-weight models you run yourself cannot be revoked by anyone’s directive,” Tandon said.
He argues that companies need to rethink architecture itself.
“The right architecture treats models as interchangeable components, not foundations…never let a capability you cannot control become load-bearing,” he said.
The latest developments add a third dimension to this trade-off. The choice is no longer simply between open and closed systems. Increasingly, it is about how much of the stack a company or a country can truly control.
Zoho founder Sridhar Vembu argued that Indian organisations should “embrace smaller models, both Indian and Chinese open-source ones”, even if they require more effort to deploy.
“Why pay money to people who don’t even want to sell to you?” he said.
This is where sovereign AI enters the discussion.
Sarvam’s Kumar described it as the need for “more countries and companies owning their own destinies”. In the context of AI, he argued, that means being able to “use and improve AI systems within their own perimeters”.
Eros Innovation founder Kishore Lulla described the moment as “inevitable”, arguing that dependence on “foreign platforms, foreign models and foreign infrastructure” creates a “structural dependency that India must urgently address”.
“Every time a foreign platform restricts access, changes terms, or simply decides that India is not a priority, we are reminded of what is at stake,” he said. “Sovereign AI is not a policy ambition. It is a national necessity.”
Policy implications
Earlier this year, more than 90 countries signed the New Delhi Declaration at the India AI Impact Summit, committing to expanding equitable access to compute, data infrastructure and frontier AI systems. The framework is built on the idea that AI should remain broadly accessible and that global cooperation is essential to unlocking its benefits.
A similar direction emerged at the Paris AI Action Summit 2025, where discussions focused on shared safety frameworks, collaborative governance and preventing excessive concentration of AI capabilities.
The Anthropic episode challenges those assumptions.
If frontier AI systems are increasingly treated as strategic assets subject to access restrictions, export controls and national-security considerations, global frameworks alone may not be sufficient to guarantee access.
In India, this thinking is already reflected in policy. The IndiaAI Mission is focused on building local compute capacity and supporting domestic AI models.
Tandon argues that the timeline for sovereign capability has effectively shifted.
“Sovereign capability has been an India ambition for the long term; this reframes it as near-term resilience policy,” he said. “The goal is not to out-build the frontier labs tomorrow, which is unrealistic, but to ensure critical sectors are never run on borrowed technology with no indigenous fallback.”
He also stressed that sovereign AI should not be viewed as a single-model or single-lab effort, but as a broader ecosystem spanning infrastructure, talent and capital.
“The state’s job is to enable, not operate. The government is poorly suited to picking architectures or running labs; what it does well is provide the conditions: compute, energy, talent and data access, plus sovereign capital routed through people who can underwrite frontier risk,” he said.
Drawing a parallel with India’s digital public infrastructure efforts, Tandon noted that the country previously built strategically important systems such as Aadhaar and UPI under significant constraints when it considered them essential to autonomy.
The lesson from Fable 5 may be similar. In an era where access to advanced AI can be altered by geopolitical decisions, capability alone is no longer enough. Control, resilience and sovereign alternatives are becoming part of the conversation.
