For most of my childhood, my father kept the whisky in a cabinet with a small brass lock he was inordinately proud of. The lock stopped no one; the key hung on a hook three feet away, in plain sight of every guest. Its job was not to deny access but to announce a rule, and the announcing was the whole point. I have been thinking about that cabinet all week, because the governments that spent two generations building locks around dangerous technology are quietly conceding that their locks were always more announcement than wall.

Consider the week's ledger. Washington has reportedly agreed to send American nuclear technology, material, and equipment to Saudi Arabia, a deal the Atlantic Council says openly raises the issue of nonproliferation. The same institution describes the underlying strategy as a 'Plan C': pursuing nonproliferation goals through deep political and commercial integration with partners' nuclear programs. In June, two of America's most capable artificial-intelligence models went dark worldwide because the government ordered it. And Europe, trying to build a wall of drones along its eastern flank, is discovering that the ally who can actually build it sits not in Washington but in Ankara.

Here is my thesis, and I will defend it: the age of technology denial is ending. For decades, the reflex of powerful states was to keep the crown jewels — enriched uranium, frontier code, precision weapons — behind a lock and dole out access as leverage. That reflex is now failing on three fronts at once, and the replacement is not a better lock. It is integration and sovereign capacity: shape the technology by being inside it, or build your own. The security this buys is real. So is the price — a more fragmented, more self-reliant, less governable world.

Plan C means Plans A and B failed

Start with Riyadh. The reported deal would ship American nuclear technology, material, and equipment to a kingdom that has never hidden its interest in the full fuel cycle, which is precisely why the Atlantic Council frames it as raising the issue of nonproliferation. The revealing detail is the label its analysts have attached to the approach — 'Plan C'. A Plan C implies a Plan A and a Plan B that did not hold. The old plans were denial plans: withhold the sensitive technology, and you withhold the bomb. Plan C inverts the logic. If you cannot stop a determined partner from acquiring nuclear capacity, the reasoning goes, then bind yourself to its program so tightly — commercially, politically, technically — that you shape it from the inside rather than watch it from the outside.

It is a serious bet, and not an obviously foolish one. Integration gives inspectors, engineers, and lobbyists a seat at a table they would otherwise be locked out of. But it concedes the central premise of half a century of arms control: that the safest number of hands on a dangerous technology is as few as possible. Plan C says the hands are coming anyway.

The off switch nobody can use

The artificial-intelligence story is the same story in a different costume. For eighteen days in June, Anthropic's Fable 5 and Mythos 5 models went dark for everyone on earth — not for a technical fault or a business dispute, but because on June 12, 2026 the Commerce Department told the company the models could no longer be supplied to any foreign person without a license, as War on the Rocks reported. Faced with an export rule it could not selectively enforce, the company concluded that the only compliant move was to switch the models off for everyone; public access to Fable 5 returned on June 30.

That episode is what a denial policy looks like when the technology refuses to respect borders. You cannot ship enriched uranium by accident, but you can serve a language model to a foreign national through a browser tab in Prague or Lagos without either of you knowing it crossed a line. The Atlantic Council puts it bluntly: country-based AI controls are failing because verifying who is really on the other end of an account is a classic, unsolved spyware-industry problem. When you cannot tell who your user is, a per-country lock has only two settings — open to all or closed to all — and closed-to-all is exactly the blunt instrument June exposed. War on the Rocks argues America still lacks a trusted, standardized process for judging a model's security risk or applying a proportionate remedy, and the Council on Foreign Relations notes that Washington is only now moving to design an AI regulator at all.

Capacity is the new sovereignty

The third thread runs through Ankara. Europe's ambition to build a 'drone wall' along its eastern border is, the Atlantic Council reports, constrained less by money or will than by industrial capacity — and the NATO ally best placed to close that gap is Turkey, not the United States. Read the three stories together and the pattern is hard to miss. When denial stops working, states do not surrender; they reindustrialize. If you cannot be reliably given the technology, you build the capacity to make it — or you buy it from whoever can, even if that whoever sits outside your usual circle of trust.

Hypothesis: these are not separate news items but one policy turn — from controlling access to controlling capacity. Supporting it: in each case the actor abandons a denial tool (the withheld reactor, the export license, the American-only supply chain) for an integration-or-build tool (Plan C, a domestic AI regulator, a Turkish drone line). Against it: the cases differ in maturity — the Saudi deal is reported, not signed; the AI regulator is a proposal, not an institution; the drone wall is an aspiration. The through-line is real, but it is a direction of travel, not a finished map.

The case for keeping the locks

The strongest objection deserves a fair hearing, because it is the one I felt myself. Denial worked. For fifty years the world had far fewer nuclear-armed states than the pessimists predicted, and the reason was not good manners; it was that the fuel cycle stayed hard to get. Integration, on this view, is a rationalization of surrender dressed as strategy: once American technology, engineers, and money are woven through the Saudi program, Washington's leverage to walk away evaporates, and 'shaping from the inside' becomes a euphemism for legitimizing what you failed to prevent. The same critic would add that the June blackout, however clumsy, at least proved the state can still pull a lever — and that abandoning the lever because it is blunt is how you end up with no lever at all.

I take the point, and I do not think it is wrong so much as nostalgic. The lock worked when the thing behind it was heavy, rare, and slow — a centrifuge cascade, a reprocessing line. It works far less well when the thing behind it is a set of model weights that can be copied, a browser session that hides its user's nationality, or a drone whose supply chain has a dozen substitutes. Denial's success was always contingent on genuine scarcity. Strip the scarcity away and the lock reverts to my father's brass one: a rule announced, not enforced. The honest response is not to polish the lock but to decide what you do once everyone knows where the key hangs.

What to watch

  • Whether the reported US-Saudi deal is actually signed, and on what nonproliferation terms — the gap between 'reported' and 'binding' is where the real policy lives.
  • Whether Washington builds the graduated 'off-ramp' War on the Rocks and the regulator the Council on Foreign Relations describe, or whether the next model shutdown is again all-or-nothing.
  • Whether Europe's drone wall is sourced from Turkey at scale — the first concrete test of capacity-over-denial on European soil.
  • Whether any of this arrives with new verification — the ability to tell who is on the other end of the account — because without it, integration is just denial with the lock left open.

My father eventually stopped locking the cabinet. Not because he trusted his guests more, but because he had worked out what the lock was really for and decided the pretense cost him more than the honesty would. The great powers are arriving, thread by thread, at the same conclusion: the age of the announced-but-unenforced lock is closing. What replaces it — integration that shapes, or capacity that fragments — will decide whether the key on the hook was a failure of security or the beginning of a more honest one.