From nuclear weapons to chips and now AI models: US ban on Anthropic's Fable 5 marks a new era of AI controls

In one of the most memorable exchanges from Game of Thrones, Petyr Baelish declares that knowledge is power.

Cersei Lannister responds with a blunt correction: "Power is power," before turning the king's guards on Littlefinger to demonstrate that authority and the ability to command force matter more than information alone.

History suggests that both were right.

The most consequential technological breakthroughs have often blurred the line between knowledge and power.

Scientific discoveries gave rise to nuclear weapons, industrial innovation helped build empires, and advances in computing reshaped economies and militaries.

In each case, knowledge evolved from being merely an asset into a source of power itself.Ever since rulers first realised that knowledge could be converted into a strategic advantage, they have sought to keep their most valuable discoveries out of the hands of rivals.

Whether it was military strategy, navigation techniques, industrial secrets, or scientific breakthroughs, technological superiority has rarely been viewed as something to be shared freely.

By the twentieth century, this instinct reached its most consequential expression in the Manhattan Project, during which the United States marshalled some of the world's brightest scientists to build the first atomic bomb while keeping the underlying knowledge shrouded in secrecy.The decades that followed witnessed similar efforts to control access to strategically important technologies.

Strong cryptography was tightly regulated, advanced jet-engine technology was closely guarded, and the Cold War-era Coordinating Committee for Multilateral Export Controls, better known as CoCom, sought to restrict the transfer of critical technologies to geopolitical adversaries.

More recently, advanced semiconductors and chipmaking equipment have become the focus of export controls aimed at preserving technological advantages and limiting rivals' access to cutting-edge capabilities.Today, that same logic is increasingly being applied to artificial intelligence.

The Anthropic momentThe shift became impossible to ignore in June when Anthropic, one of the world's leading AI companies, disclosed that it had received a US government directive affecting access to some of its most advanced AI systems for foreign nationals.

What began as a dispute over a specific model and a specific security concern quickly evolved into a broader debate over whether frontier AI should be treated less like software and more like a strategic asset.According to Anthropic, government officials raised concerns that some of its most advanced models could potentially be manipulated to assist in identifying software vulnerabilities.

The company argued that the vulnerabilities in question were limited and comparable to capabilities already available elsewhere.In a public statement, Anthropic warned of the wider implications of the government's position."If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers."The dispute may appear technical, but it sits at the centre of a far larger question.

Who should have access to the world's most powerful AI systems? Why Washington is concernedAccording to administration officials, the restrictions were driven by concerns that increasingly powerful AI models could potentially be manipulated to identify software vulnerabilities and support malicious cyber activity.

Officials reportedly feared that sufficiently sophisticated users might bypass built-in safeguards and exploit the systems in unintended ways.The issue prompted discussions between technology executives and senior administration officials about the cybersecurity implications of increasingly capable AI systems.

Administration officials maintained that export controls were imposed only after other efforts to address those concerns had failed.As one White House official reportedly put it:"Export controls were a last resort after begging them for hours to work with us."Anthropic, however, disputed the government's assessment.

The company argued that the vulnerabilities in question were limited and comparable to capabilities already available elsewhere.

In a public statement, Anthropic warned:"If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers."The disagreement highlights the growing tension between national security concerns and the openness that has traditionally characterised AI research and development.Understanding computeTo understand why governments care so deeply about frontier AI, it is necessary to understand a concept that sits at the center of the modern AI race: compute.Compute is the industrial machinery of artificial intelligence.Training frontier models requires enormous GPU clusters, massive datasets, extraordinary amounts of electricity, and billions of dollars in investment.

Developing a leading AI model increasingly resembles building national infrastructure rather than creating a traditional software product.This concentration of resources matters because it transforms AI from a purely commercial technology into a strategic one.A handful of countries possess the capital, semiconductor supply chains, energy resources, and research ecosystems necessary to train frontier models at scale.

As a result, AI development is becoming increasingly intertwined with geopolitical competition.That reality helps explain why advanced semiconductors have become a focal point of US policy.

Washington has spent years restricting exports of advanced chips and chipmaking equipment in an effort to slow competitors' progress in strategically sensitive areas.Those policies rest on a straightforward assumption: if access to the machinery of AI can be limited, the pace of frontier AI development can be influenced as well.

From chip controls to model controlsYet frontier AI introduces a new challenge.If governments restrict access to advanced chips but allow unrestricted access to the powerful models trained on those chips, some policymakers argue that export controls become less effective.Put differently, if the factory is tightly controlled but the finished product remains universally available, the strategic advantage may prove difficult to preserve.Whether that argument ultimately holds up remains contested.

But it increasingly shapes policy discussions in Washington and other capitals.The Anthropic episode may represent one of the clearest signs yet that governments are beginning to shift their focus from controlling hardware to controlling capabilities.For decades, technology policy concentrated on physical assets: uranium, missiles, supercomputers, advanced semiconductors, and manufacturing equipment.Now the attention is gradually shifting toward the intelligence those assets make possible.

The restrictions have also drawn criticism from several influential figures in the AI community.

Meta chief AI scientist Yann LeCun accused Anthropic of helping create the conditions for such measures through years of emphasizing existential AI risks, remarking that "one reaps what one sows." Former White House AI adviser David Sacks similarly warned that fear-driven narratives risk producing policies that ultimately undermine innovation.

Anthropic itself argued that the government's concerns were too narrow to justify restrictions broad enough to affect legitimate researchers and users, warning that applying the same standard across the industry could disrupt the deployment of frontier models.

Critics in Europe and cybersecurity circles have also questioned whether nationality-based restrictions can meaningfully improve security, arguing that they may instead encourage researchers and companies to shift toward alternative ecosystems and non-US AI providers.The talent dilemmaThe debate becomes even more complicated when talent enters the picture.

For decades, America's technological leadership has rested on a simple formula.

Attract the world's best scientists, engineers and entrepreneurs, then give them access to capital, universities and research laboratories.

Silicon Valley itself is a product of that model.

According to the National Foundation for American Policy, immigrants have founded or co-founded more than half of America's billion-dollar startups.Many of the leaders who shaped the modern technology industry were born elsewhere.

Alphabet chief executive Sundar Pichai grew up in India before studying at Stanford and Wharton.

Microsoft chief executive Satya Nadella arrived from Hyderabad as a graduate student.

Nvidia founder Jensen Huang emigrated from Taiwan as a child and went on to build one of the world's most valuable companies.

OpenAI co-founder Ilya Sutskever was born in Russia and raised in Israel before moving to North America.

Their stories reflect a broader pattern.

American dominance in technology has been reinforced by its ability to absorb talent from abroad.That pattern extends far beyond chief executives.

More than half of doctorate degrees awarded by US universities in computer science and engineering go to international students.

Georgetown University's Center for Security and Emerging Technology has estimated that a majority of top AI researchers working in the United States were born overseas.

Immigrants account for roughly one-fifth of the country's STEM workforce despite representing a much smaller share of the population.Critics of broad restrictions argue that excessive controls could undermine one of America's greatest advantages.

They warn that treating foreign nationals as security risks may discourage the researchers and entrepreneurs who have helped build US leadership in semiconductors, software and artificial intelligence.

Anthropic itself argued that the government's concerns did not justify measures broad enough to affect legitimate researchers and users.Several technology leaders share those concerns.

Former Google chief executive Eric Schmidt has argued that America's edge comes from attracting exceptional people from around the world.

Researchers have similarly warned that tighter restrictions could push talent towards Europe, Canada, Singapore and increasingly China, where governments are investing heavily to recruit scientists and engineers.Supporters of tighter controls see the issue differently.

Their argument is that frontier AI could eventually acquire strategic importance comparable to nuclear technology or advanced cryptography.

In that world, governments cannot afford to wait for problems to emerge before acting.

Some loss of openness, they argue, is a reasonable price for reducing long-term risks.The dilemma cuts to the heart of America's technological success.

The country became a global innovation powerhouse because it remained open to talent.

As artificial intelligence becomes increasingly intertwined with national security, Washington faces a difficult balancing act.

What history suggestsHistory shows that export controls can slow the spread of strategically important technologies, but rarely stop it for long.

Nuclear weapons, advanced computing and strong encryption were all subject to tight restrictions at different points, yet each eventually diffused beyond the original centres of control.

The pattern is familiar: governments can delay access, raise costs and buy time, but they struggle to contain technologies indefinitely once the commercial, scientific and geopolitical incentives to spread them become strong.That is why the current debate matters.

If frontier AI follows the same path, then today’s controls may shape the pace of diffusion, but not its eventual direction.

The real question is not whether AI can be contained forever, but how much leverage states can preserve while the technology is still young.Why frontier AI is differentWhat makes frontier AI harder to regulate is that the strategic asset is no longer just the hardware.

Chips still matter, but the bottleneck has shifted towards compute, model access and the ability to train systems at scale.

In practical terms, that means a model can be treated less like ordinary software and more like an industrial capability, built on scarce resources, massive capital and concentrated expertise.That is also why policymakers are paying closer attention to access itself.

If a model can help identify vulnerabilities, assist cyber operations or amplify high-end capabilities, then the state is no longer looking only at the machine running it.

It is looking at the intelligence the machine produces.

In that sense, frontier AI sits somewhere between software, infrastructure and strategic power.

Can AI really be contained?For all the comparisons with nuclear weapons and Cold War export controls, the Anthropic episode raises a narrower question.

Can access to frontier AI models be restricted in a meaningful way, or will capabilities inevitably spread? According to Reuters, the US government's directive was driven by concerns that a jailbreak technique could allow Anthropic's latest models to identify software vulnerabilities.

Anthropic disputed the severity of the issue, saying the vulnerabilities were minor and comparable to capabilities already available elsewhere.The dispute has exposed divisions within the AI community.

Former Meta chief AI scientist Yann LeCun criticised what he described as excessive fear surrounding advanced AI and accused Anthropic chief executive Dario Amodei of "fear-mongering".

Former White House AI and crypto adviser David Sacks similarly accused Anthropic of pursuing a regulatory strategy that could benefit large incumbents.

Anthropic, meanwhile, warned that applying the same standard across the industry could disrupt frontier-model deployment.History offers evidence for both sides.

Export controls have often bought time and raised costs for rivals, but they have rarely stopped transformative technologies from spreading indefinitely.

Whether frontier AI proves different remains an open question.What it means for Anthropic and the worldFor Anthropic, the episode suggests that frontier AI companies may increasingly find themselves operating under the kind of national-security scrutiny once associated with semiconductors and other strategic technologies.

The dispute also illustrates how closely AI development is becoming intertwined with geopolitics.Beyond a single company, the issue raises a broader question for governments.

Should frontier AI be treated primarily as a commercial technology, or as a strategic capability with national-security implications? Reuters reported that G7 leaders discussed frameworks for granting trusted partners access to advanced US AI systems, indicating that questions of access and control are already becoming matters of international policy.For countries such as India, the lesson is particularly important.

Access to foreign models can provide immediate benefits, but long-term competitiveness may depend on domestic compute infrastructure, semiconductor capacity, research institutions and talent development.

As AI becomes increasingly tied to economic and strategic power, dependence on external capabilities could emerge as a vulnerability.The significance of the Anthropic episode extends beyond a single model or a single government directive.

It reflects a larger shift in how states view artificial intelligence.

For decades, governments focused on controlling the hardware that enabled technological advantage.

Increasingly, attention is shifting towards the capabilities themselves.Whether that approach proves effective remains uncertain.

History suggests that transformative technologies eventually spread, even when governments attempt to restrict them.

Yet history also shows that states rarely surrender strategic advantages willingly.The modern frontier is no longer defined by geography.

It is increasingly shaped by access to intelligence itself.Get the latest technology news and updates.

Download the TOI App.