Several months ago, on Shimmy Says and in an earlier column, I argued that digital sovereignty is national sovereignty and digital sovereignty is personal sovereignty. At the time, the discussion focused primarily on data residency, cloud infrastructure and the increasingly complicated question of whose laws govern information stored in a world without obvious borders. Those conversations remain important, but recent events suggest we have been looking at only part of the problem.

The strategic asset in the age of artificial intelligence is no longer simply data. It is access to intelligence itself.

That may sound like an abstract distinction, but the implications are profound. A nation may insist that its data remain within its borders, an enterprise may invest billions securing its intellectual property and an individual may carefully protect personal information, yet all three can still become dependent on an artificial intelligence system that exists entirely outside their control. If access to that system can be altered by a government directive, a corporate policy decision or a geopolitical dispute, then sovereignty extends well beyond data. It extends to cognition.

The events surrounding what I have come to call the Anthropic Affair provide perhaps the clearest example yet of why this matters.

Over the past year, Anthropic cultivated the image of the responsible frontier AI company. It spoke openly about safety, constitutional AI and the need for guardrails around increasingly capable systems. The company also drew ethical boundaries around certain uses of its technology, particularly applications involving mass domestic surveillance and autonomous lethal weapons. Whether one agrees with those positions is almost beside the point. Anthropic was asserting that the creators of frontier AI retain some responsibility for how those systems are ultimately deployed.

The U.S. government viewed that relationship differently. National security policy has traditionally been the domain of elected governments rather than private corporations, and tensions between the two became increasingly visible. The disagreement highlighted a question that will likely define the next decade: When artificial intelligence becomes essential national infrastructure, who gets the final say over how it is used?

The dispute became even more complicated when Anthropic publicly suggested that frontier AI development might be accelerating too quickly and that the industry should consider some form of coordinated pause. It was a thoughtful proposal, but one that immediately encountered the same obstacle that has frustrated every discussion of AI restraint. No company can realistically stop while competitors continue advancing, and there is no credible mechanism for persuading every major AI power to move in lockstep. Any unilateral pause risks becoming a strategic surrender.

That debate was soon overtaken by another. Anthropic introduced Fable, a model reportedly designed to retain much of the capability of its more advanced Mythos system while limiting certain cybersecurity functions associated with vulnerability discovery. According to multiple reports, Amazon, one of Anthropic’s largest investors and strategic partners, demonstrated a jailbreak capable of bypassing some of those limitations and alerted government officials. The irony was difficult to miss. A company deeply invested in Anthropic’s success became part of the chain of events that ultimately triggered government intervention.

The response reportedly required that foreign nationals, including those employed by Anthropic itself, could no longer access the affected models. Anthropic concluded that enforcing nationality restrictions across a global platform was operationally impractical and instead withdrew access much more broadly.

Reasonable people can disagree about every step in that sequence. They can debate whether the government’s concerns were justified, whether Anthropic’s safeguards were sufficient or whether the company should have approached the issue differently. Those are legitimate questions.

They are also secondary.

The lasting significance of the Anthropic Affair is that it demonstrated, in unmistakable fashion, that access to frontier intelligence can become a matter of state policy. A capability that thousands of organizations expected to be available as a commercial service suddenly became subject to geopolitical considerations.

That should force a rethinking of what Sovereign AI actually means.

Until now, discussions of AI sovereignty have largely focused on infrastructure. Nations invested in sovereign clouds, sovereign datacenters and sovereign compute. Enterprises sought assurances that sensitive data would remain within approved jurisdictions. Regulators concentrated on privacy, residency and compliance. Those efforts reflected the assumption that controlling data meant controlling the future.

Artificial intelligence changes that equation.

The value no longer resides solely in the information being processed but increasingly in the intelligence doing the processing. A country may possess enormous computing infrastructure and keep every byte of information within its borders while still depending on a model owned, operated and ultimately controlled elsewhere. An enterprise may spend years building proprietary datasets only to discover that the intelligence transforming those datasets into competitive advantage is subject to export controls or policy decisions entirely beyond its influence.

The same observation applies at the individual level.

The entrepreneur building a startup, the physician researching treatments, the engineer designing products and the student learning new skills increasingly rely on AI as a cognitive partner. Their productivity, creativity and competitiveness become intertwined with access to advanced models. If that access can be withdrawn because of events occurring thousands of miles away, personal sovereignty becomes conditional as well.

This is why I believe Sovereign AI must be understood at three distinct levels.

• Nations require Sovereign AI because dependence on foreign intelligence creates strategic vulnerability.
• Enterprises require Sovereign AI because intellectual property, innovation and operational continuity cannot rest entirely on technologies controlled by external parties.
• Individuals require Sovereign AI because personal agency increasingly depends upon access to intelligence augmentation.

One component of that strategy deserves considerably more attention than it currently receives.

Open source.

For decades, advocates of open source software reminded the industry that “free” meant freedom rather than price. The value of open source was never simply that software could be downloaded without cost. Its deeper value lay in reducing dependence on any single vendor and preserving the ability of users to inspect, modify and continue using software regardless of the fortunes or policies of the original creator.

That philosophy becomes newly relevant in the AI era.

Open source models will not eliminate safety concerns or remove the need for governance. They are not a substitute for responsible development. They do, however, make it substantially more difficult for any single company or government to become the exclusive gatekeeper of intelligence. A model whose weights are broadly available cannot simply disappear because one provider changes its acceptable use policy or one government issues an order. It can continue evolving through the efforts of universities, enterprises, nations and open communities around the world.

For that reason alone, open source should be viewed as an essential pillar of Sovereign AI rather than merely another licensing model.

None of this should be interpreted as an argument against the United States or in defense of Anthropic. Any nation possessing strategic technological leverage will be tempted to exercise it. History provides abundant examples of governments using control over critical resources to advance national interests. The United States happens to occupy that position today in frontier AI. Another country may occupy it tomorrow.

The larger issue is whether humanity benefits from concentrating so much influence over intelligence in the hands of so few actors.

The technology industry spent three decades promoting the vision of a borderless digital world in which information flowed freely across continents. The AI era may be taking us in the opposite direction. Data is becoming sovereign. Compute is becoming sovereign. Models are becoming sovereign. Intelligence itself is becoming sovereign.

The Anthropic Affair did not create that trend. It simply exposed it.

The lesson extends far beyond one company or one administration. It is a reminder that dependence is often invisible until the day someone else exercises control. Nations have learned that lesson about energy. Enterprises have learned it about supply chains. The AI era will teach it again about intelligence.

The defining question of this new era may not be who builds the smartest model. It may be whether governments, enterprises and individuals are willing to invest in the sovereignty required to ensure that no one else can decide when they no longer get to use it.