Key Takeaway
AI is changing the balance of power. Organisations need to preserve agency.
The Bottom Line Upfront (BLUF)
Access to frontier AI is a dependency and not an asset. In June 2026 the United States withdrew its most capable models from every foreign national on the planet — allies included — then restored access three weeks later on terms it set alone. That exposure is structural, and it did not end when the ban did. While the risk this poses cannot be competed away, it can be narrowed.
The fundamental challenge of the AI age isn’t access to AI. It’s maintaining organisational sovereignty while adopting AI. Every organisation possesses exactly three AI assets no vendor can create for them: their data, their people, and their capacity to deploy. Every successful AI program is ultimately an exercise in strengthening those three assets and improving how they interact with one another.
“Don’t Bother”
A few years ago, I had the opportunity to pose a question to a leading figure within the Asian AI community. “If the United States and China are going to dominate the world of AI…” I asked, “What can other countries do to survive and thrive in the age of AI?” His answer was blunt.
“Don’t bother.”
Words said without a nanosecond of hesitation. The room fell silent for several seconds, tension building until someone else found a question to bridge the awkward silence and the session moved on. I have never forgotten those words or the absolute confidence with which they were said.
Then, relatively early in my professional detour into the world of public sector data science and AI, I found those words to be grating.
The two AI superpowers, the US and China, were and remain the clear leaders in the AI domain with unassailable access to compute, data, and talent. However, everyone else had the opportunity to use, customise and adapt these new technologies to create commercial value and public good.
The problems I worked on in the public sector weren’t commercially attractive. Creating public good comes with cost, not ledger sheet profit or cash shareholder dividends, while creating value and opportunity for many. The data I worked on was unique and protected, never to be released publicly or to a commercial vendor under some exclusive arrangement. The open-source data science community was collegial. Data science, AI, and machine learning code libraries enabled anyone in the world to download, test, develop and deploy new capabilities once applied to a dataset. All this activity did not fit well with the sentiment of “don’t bother”.
Cut to June 2026
The United States government moved to block Anthropic’s most capable models, Fable 5 and Mythos 5, from being used by “any foreign national, whether inside or outside the United States, including foreign national Anthropic employees”. Not just adversaries either. Friends and allies. The export control directive reached America's closest partners and the foreign nationals working inside the very company that built the models. The reason was narrow and technical, a disputed security concern. Access was removed simultaneously with no prior notice or consultation.
The ban was lifted a little less than three weeks later. Fable 5 was made available to everyone; Mythos 5 access was restored to vetted US partners participating in Anthropic’s Project Glasswing.
However, a message was sent, and it was just as blunt as those words stated all those years ago by an Asian AI titan to a group of foreigners. The other AI superpower delivered the same message to its friends and allies. This time, not with words, but with export controls. This is our capability, not yours.
Despite the ban being lifted, access to critical technology can be withdrawn and reinstated without consultation on terms set and executed by Washington. Three weeks is a short outage but a stark lesson in who holds the power and is willing to use it.
What is frightening isn’t that two of the nearly 200 countries in the world have an unquestionable lead on a general-purpose, paradigm-shaping technology critically important to all. It’s the narrow concentration of power, influence, and effect that spans economic, societal, and political domains and reaches virtually every sector.
Lessons from history – The Gutenberg Printing Press
Large-scale technological disruption is not new. There are lessons from our past that can illustrate what is at stake.
Years ago, I sat in a class on digital government with Professor David Eaves. Citing arguments from books The Gutenberg Galaxy: The Making of Typographic Man (Marshall McLuhan), and Imagined Communities (Benedict Anderson), Eaves talked about the printing press as the foundation of the nation-state. He argued that before the press, the modern institutional nation-state was impossible. You could not deliver language, knowledge, and information at the scale a state requires to hold itself together. The press made the state possible.
But Eaves talked about the other side of it, too. The press standardised language, and standardisation has casualties. The France that emerged as a modern nation-state is not the country people belonged to before the printing press. Before it, you might identify with your tribe, your village, your region, your own language group. Eaves asked a question I have never forgotten: how many dialects, how many language groups simply went extinct as the printing press fixed one structure of language in place? How much of our unique human culture was culled so the technology could scale?
Fast forward to now. We have built a technology that can become more intimate than the printing press ever was. Intimate enough that we can cede to it not only our language, but our intellect and our identity. The standardising forces that came with the press are not weaker today. They are far stronger. And in the age of AI, they radiate from two centres of power and a handful of companies.
If we choose to adopt only technologies from those two centres without agency, we are ceding more than convenience. We are ceding our sovereignty, our identity, our prosperity, our agency, and ultimately our future to organisations and decision-makers that are not our own.
Safeguarding sovereign assets
In the face of this overwhelming concentration of power and influence in the AI domain, what can a country, a company or an institution do to protect itself?
The organisations positioned to survive and thrive in the age of AI already hold three things that no superpower can issue or revoke.
The first is data. They are the custodians of immense and diverse data holdings, uniquely positioned to collect, curate, and create high-value data assets — particularly in the domains that are theirs and no one else's. That is the raw material, and they already own it.
The second is people. People with deep and varied human expertise, across a range of domains an AI agent simply will not have. An organisation might employ people with more diverse experiences than what is contained within a standard resume. The value of that diversity of expertise is not easily measured. And because these people sustain collaborative relationships with allies and partners, the cross-pollination continues. Innovation has always happened exactly when people from different domains, different experiences, and different skills come together and collide.
The third is the ability to deploy, execute and scale. Having the ability to bring life to insights and outputs derived in the digital world into the real world. The resources, the structure, the culture and capacity to put capability into production rather than admire it in a demo. This is the asset most often overlooked, and it is the one that turns the first two into something real.
But the three things hold at every scale. A country without compute, without data centres, without data scientists and effective universities is renting its future from someone else. A company that hands its data, its people, and its structure to a handful of vendors is doing the same. Whoever stewards their data and people, and maintains control of their capacity to deploy, keeps custody of themselves.
Taking responsibility
Custody carries an obligation. To hold these assets requires assuming responsibility. A country, an institution, a company has a duty not to surrender them: not to a hyperscaler headquartered in one of the two centres of power, and not to a vendor who will take your data, your intellectual property, your people, and your capacity to deliver in exchange for a short-term saving and a long-term dependency. Careless outsourcing and poorly negotiated contract terms are not merely bad trades. They are the quiet ceding of your sovereignty, your autonomy, your agency, and your accountability for the outcomes you exist to deliver.
The organisations that survive and thrive in the age of AI will be the ones that do so with purpose. Not the ones who ask what the technology can automate, but the ones who ask what they intend to do with it — how they will wield it, what heights it lets them reach. They will safeguard their risks and their responsibilities precisely because they refuse to outsource the reasons they exist: their core IP, their organisational identity, and their ability to deliver the outcomes that make the world a little better for having been in it.
That is the gap that needs to be closed. Not to argue that anyone should refuse the technology but to insist that institutions can adopt it on their own terms, in their own interests, and keep hold of the things that make them worth backing in the first place.
The Asian AI titan was right about one thing. If your plan is to beat Washington and Beijing at their own game, on their scale, don't bother.
But that was never the game. The work is not to win the race. It is to refuse to disappear into it. Hold on to your data, your talent, and your power to deploy and to build on your own terms, something they cannot issue and cannot take back.