When Does Alliance Anxiety Become Nuclear Proliferation Risk?
A familiar worry in foreign-policy circles goes like this: if American allies stop believing the United States will defend them, the most capable of them will start working toward nuclear weapons of their own. South Korea and Japan are the test cases. Both live next to nuclear-armed rivals, both rely on the US nuclear umbrella, and both could build a bomb on a short timeline if they decided to.
For most of the nuclear age this stayed an abstract fear. It is less abstract now. In January 2026 the US National Defense Strategy dropped any explicit mention of "extended deterrence," the decades-old promise to defend allies with the full weight of American power, nuclear weapons included, and instead told allies to take "primary responsibility" for their own defense, with "critical but more limited" US support. Allied governments noticed. They had also watched Ukraine relinquish inherited Soviet nuclear weapons and later be invaded, and Iran’s nuclear facilities be struck before Iran had fielded a nuclear weapon. Seoul, under progressive President Lee Jae Myung, used an October 2025 summit in Gyeongju to win US backing for a path toward enriching its own uranium and reprocessing spent fuel, plus approval to build nuclear-powered submarines. In Tokyo, Prime Minister Sanae Takaichi spent late 2025 hedging on whether Japan's Three Non-Nuclear Principles would survive a coming review of its strategy documents, before reaffirming them in February 2026.
So the question is no longer whether allies are anxious. They plainly are. The question is what that anxiety becomes. This piece tries to turn the vague fear into something testable: the specific events that would count as a breaking point, and the evidence that would change my mind.
Two ways to read the same anxiety
A note on method before the two camps: I've built "the pessimists" and "the optimists" as two stylized positions and assigned each the probabilities it would plausibly hold. They aren't summaries of named scholars or polled forecasters; they're a device for reasoning through what the evidence might mean. The events and figures cited throughout are real; the camps and their numbers are mine.
The pessimists, structural realists, argue that the decline in US credibility is durable and self-reinforcing. Each year of erosion makes the next harder to reverse. They point to the 2026 strategy, the burden-sharing demands, and the Ukraine and Iran lessons, and they read the allied response as hedging that will harden into real movement toward a weapons option. Their strongest argument is about concealment: every past instance of allied weapons activity was secret. South Korea ran a covert program under Park Chung-hee in the 1970s; Taiwan did the same into the 1980s; Japan has quietly studied the question more than once. States that fear
the costs of going nuclear have every reason to keep genuine movement off the public record, which means the public signals always lag the real intent.
The pessimists are careful about one thing, and it's worth dwelling on because it's the most disciplined part of their case. They do not lean on opinion polls. South Korean majority support for the bomb is old news. It has sat above 70% for years while the state did nothing, hitting a record 80% in 2026 (up from 76.2% in 2025). High public support has coexisted with state restraint for over a decade, so by itself it proves nothing. What the pessimists treat as new is elite and state behavior: the enrichment and submarine track, Japan's strategy review, the debate among senior officials. The level of public opinion is background; the movement in state conduct is the signal.
That discipline matters this year in particular, because the opinion data cut against the alarmist read. The 2026 Asan poll found that even as desire for the bomb climbed to a record 80%, South Korean confidence in the US commitment rose about ten points to 59%, reversing a dip the year before. The "collapsing trust" a loose version of the pessimist story might lean on did not, in fact, collapse. The serious pessimist case doesn't need it to (it was always about what governments do, not what publics say), but it's a useful reminder of how easily the two get blurred.
The optimists, institutionalists, don't deny the anxiety. They argue the system absorbs it. The Non-Proliferation Treaty has constrained allied behavior since 1970, and the historical record cuts their way: South Korea and Taiwan both started weapons programs and both abandoned them under US pressure. In fact, the covert cases the pessimists cite are cases where concealed programs were caught and rolled back, evidence that detection works. The optimists read today's behavior as channeled into sub-threshold paths, not weapons pursuit: Seoul's enrichment work proceeds bilaterally with US oversight, and Korea's foreign minister has said the planned submarines will run on low-enriched fuel, 20 percent or less, well below weapons grade. Japan's review is broad, not nuclear-specific. France's March 2026 initiative, which drew eight European partners into a nuclear-cooperation dialogue while Paris kept sole control of any launch decision, shows how allied nerves get absorbed into consultation rather than independent arsenals. And the rising confidence number, they argue, is exactly what you'd expect if 2026 is a passing shock rather than a structural break. The costs of a formal break remain brutal: sanctions, isolation, near-certain Chinese and North Korean reactions, and the loss of the very US support the ally was worried about.
Notice that nobody disputes the anxiety itself. The fight is over what the anxiety means. Is it the start of something, or noise that existing institutions will soak up?
The finish line
A forecast needs an unambiguous outcome, so I'll set a deliberately narrow one.
The outcome (U): Before January 1, 2036, South Korea or Japan either formally files notice of withdrawal from the NPT under Article X, or becomes the subject of a formal IAEA Board of Governors finding of safeguards non-compliance, referred to the UN Security Council, in connection with suspected weapons activity.
Read that carefully, because it's narrower than the headline fear. It is not "builds a bomb." It is a formal break with the nonproliferation system: a treaty exit notice or a referral to the Security Council. A country could debate nuclear weapons openly, expand its enrichment program, grow its plutonium stocks, push for nuclear sharing, and inch right up to the technical threshold without ever tripping this question. All of that might be evidence the pessimists are onto something. But it doesn't resolve the forecast unless someone crosses a formal institutional line.
There's a tradeoff here and I want to be honest about it. A broader question would capture more of the real concern but would be nearly impossible to judge cleanly. A narrow question is judgeable but can miss meaningful movement below the threshold. I took the narrow version because a forecast you can't score isn't worth much. The cost is that a NO in 2036 wouldn't prove eroding US credibility had no consequences. It would only mean the consequences stopped short of a formal rupture.
The five-year crux
2036 is a long way off. If the goal is to think better now, you need a nearer-term question whose answer would actually move the forecast. Forecasters call this a crux. Mine asks whether, before January 1, 2031, South Korea or Japan does any one of four things in either country:
1. Launches a formal public weapons review: an official commission, task force, or legislative process whose mandate explicitly includes evaluating whether to acquire nuclear weapons. Not a politician musing on TV.
2. Conducts a credibly reported classified review: at least two major news organizations, each with a named official source or government documents, reporting a secret study to that effect. This one matters because the historical record says serious weapons work usually starts in secret; a crux that only watched for public announcements would miss the most plausible pathway.
3. Crosses a hard technical threshold: enriching uranium past 20 percent, or separating plutonium, outside an agreed, safeguarded civilian arrangement. That's categorically different from normal civilian nuclear activity, and beyond even the low-enriched submarine-fuel work Seoul says it intends.
4. Formally threatens or invokes Article X, the NPT's withdrawal clause, on "supreme interests" grounds.
Just as important is what doesn't count. Polls don't count. Think-tank debates don't count. Opposition politicians sounding off doesn't count. A general defense review with no nuclear mandate doesn't count. Conventional buildups, nuclear-sharing talks, "all options are on the table" rhetoric: none of it counts. Those things might be politically interesting, but they can't distinguish between a nervous ally hedging and a state actually moving toward a weapon. The whole point of the crux is to draw that line.
Why four indicators, not one? Because the two camps don't disagree about US declaratory policy; they disagree about whether anxiety becomes state action. A single public-review indicator fails that test twice. A public review can be a pressure-release valve, launched to
placate voters and then shelved, so a YES wouldn't clearly favor the pessimists. And since real movement has historically been covert, a NO on a public-only indicator would be weak evidence that the brakes held, since it could just mean the program went underground. The basket fixes both. Indicators 2 through 4 reach the concealed and physical pathways a state can't easily stage for show. That makes a clean NO (no public review, no reported secret review, no technical crossing, no treaty threat, across both countries, over five years) genuinely strong evidence for the optimists, in a way a single-indicator version never could be.
The numbers
| Question | Pessimists | Optimists |
|---|---|---|
| Probability of the outcome by 2036 | 20% | 7% |
| Probability the crux fires by 2031 | 50% | 20% |
| Probability of the outcome if the crux fires | 36% | 25% |
| Probability of the outcome if it doesn’t | 4% | 2.5% |
These aren't free-floating. Each side's overall estimate has to be the weighted average of its two conditional paths, and both reconcile exactly: the pessimists' (0.36 × 0.50) + (0.04 × 0.50) = 20%, and the optimists' (0.25 × 0.20) + (0.025 × 0.80) = 7%. The check doesn't prove either side right; coherent and wrong is a common combination. It proves each side is internally consistent: the headline number isn't smuggling in optimism or alarm the conditional reasoning doesn't support.
What jumps out is how much the two sides agree about the conditionals. Both think a fired indicator would sharply raise the risk. Both think five quiet years would push it toward negligible. The real fight is over whether the indicator shows up at all: a coin flip for the pessimists, a one-in-five shot for the optimists.
One more thing worth saying plainly: the 2026 confidence rebound doesn't move the pessimist numbers, and that's defensible rather than evasive. Their case was never built on public opinion; it was built on what governments do, and 2026 brought continued state-level movement, the Gyeongju enrichment and submarine track and Japan's live security-document review. The optimists, already near the historical floor, read the rebound as confirmation of their transitory-shock view. So the new poll sharpens the disagreement instead of settling it.
The weight of history
A 20% estimate sounds modest until you set it against the record. In 55-plus years, the number of US-allied non-nuclear states that have formally broken with the NPT, whether by withdrawing
under Article X or being referred for non-compliance, is exactly zero. (Actually building a bomb is rarer still.) The only state ever to withdraw is North Korea. So the pessimists are forecasting a large departure from history, and they owe an account of why history stops applying.
Their answer is specific: that zero-event record was built entirely under an intact, explicitly declared US umbrella, the very thing the institutionalists credit for allied restraint. If the 2026 strategy, the burden-sharing pressure, and the Ukraine and Iran lessons amount to a real change in that underlying condition, the old base rate no longer governs.
Even so, the pessimists discount heavily for the breakout costs the optimists name, which is why their number is 20% and not higher, with most of the probability still sitting on sub-threshold outcomes. The optimists accept the logic but deny the premise: they read 2026 as a transient, administration-level shock rather than a structural shift, and point to the confidence rebound as evidence, so they stay near the historical floor at 7%. Whether you find that reassuring or complacent depends on how durable you think the credibility erosion really is. Which is, of course, the actual question.
What the crux would teach us
The value of the five-year window is that it forces both sides to move.
If an indicator fires before 2031, the debate changes immediately, because anxiety will have become state action. Both sides revise up, the pessimists toward 36%, the optimists toward 25%. They'd still disagree, but they'd agree it mattered.
If the window closes clean, five years, four indicator types, two countries, nothing, the two theories converge. The pessimists fall to 4%, the optimists sit at 2.5%, and the disagreement mostly dissolves. And because the basket was built to catch concealed and behavioral movement, that clean NO can't be waved away as "it was just hidden." Five years of silence across all four indicators means the brakes held.
That two-way sensitivity is what makes the crux worth having. A yes moves the forecast one direction, a no moves it the other. A warning sign that can only confirm your fears isn't a test of anything.
Where I come down
I come down closer to the optimists. If forced to a number, I'd put the chance that South Korea or Japan formally breaks with the nonproliferation regime by 2036 at roughly 10–12%, above the optimists' 7%, well below the pessimists' 20%.
The load-bearing consideration for me is the base rate, and specifically what it would take to overturn it. For more than half a century, US-allied non-nuclear states have had every security incentive to go nuclear, and none has crossed the formal institutional line this forecast uses. The pessimists have a real answer to that, namely that the record of restraint may depend on credible US deterrence, a condition now under strain, and it's strong enough that I won't sit at 7%. But it's a claim about a premise that hasn't yet been demonstrated, and the 2026 evidence cuts the other way: even as South Korean support for the bomb hit a record 80%, confidence in the US commitment rose ten points to 59%. The trust whose collapse the pessimist story needs is, for now, not collapsing. Absent that proof, the base rate keeps most of its weight.
I'd express the residual uncertainty through the crux. I put the chance that at least one serious indicator fires by 2031 at around 25–30%, higher than the optimists, because the concealment problem and state-level hedging are real, and lower than the pessimists, because the most probable path still runs through institutional absorption rather than rupture.
So the bottom line: the crux is live, but the finish line is hard to cross. Alliance anxiety is real and consequential, and it will likely show up as more hedging, more latency, more political debate. I just think it's more likely to be managed inside the existing system than to produce an NPT withdrawal or IAEA referral by 2036, and 2031 will tell us a great deal about whether that's right.
The point of the exercise
We started with a sprawling fear: that fading trust in American power might set off a nuclear chain reaction among its closest friends. Calling that fear "serious" or "plausible" gets us nowhere. The work is in drawing a clean line around the outcome, naming the rival explanations, picking evidence that can tell them apart, and checking that the probabilities reconcile.
The hardest part wasn't the arithmetic. It was keeping two categories separate that public debate constantly blurs: a nervous ally hedging, and a state moving toward a weapon. South Korea or Japan can debate nuclear options, demand stronger commitments, build up their militaries, and negotiate civilian nuclear flexibility without ever rupturing the regime. Those things matter, but they aren't the same thing, and a forecast that can't tell them apart will see proliferation risk everywhere or nowhere depending on the forecaster's mood.
By January 2031 we'll have a real answer to the near-term question, and either way, the forecast will have earned its conclusion instead of asserting it. Reasoning under uncertainty isn't mainly about sounding confident. It's about making your judgment inspectable, so that when the evidence arrives, you and everyone else can see whether it should change your mind.
Sources and notes
1. On the 2026 National Defense Strategy and extended deterrence: the official NDS says allies and partners should take “primary responsibility” for their own defense with “critical but more limited” U.S. support; a Congressional Research Service report summarized by USNI News notes that the 2026 NDS did not explicitly mention extended deterrence.
Official 2026 National Defense Strategy
2. On South Korean public opinion: the Asan Institute’s 2026 annual poll reported record 80% support for an indigenous nuclear-weapons capability, up from 76.2% in 2025, while confidence in the U.S. extended nuclear deterrence commitment rose to 59%.
3. On U.S.–ROK nuclear cooperation: the White House joint fact sheet from President Trump’s meeting with President Lee Jae Myung states that the United States supports a process leading to civil uranium enrichment and spent-fuel reprocessing for peaceful uses, and has given approval for the ROK to build nuclear-powered attack submarines.
4. On South Korea’s nuclear-powered submarine plans: Reuters reported that South Korea aims to launch its first nuclear-powered submarine by the mid-2030s using low-enriched uranium fuel; Reuters also reported that U.S.–ROK talks have addressed enrichment, reprocessing, and fuel-sourcing issues connected to the submarine program.
5. On Japan’s Three Non-Nuclear Principles: Carnegie summarizes the principles and notes that Prime Minister Sanae Takaichi’s 2025 comments created debate over whether they would remain in Japan’s reviewed strategic documents. The Diplomat reported in February 2026 that Takaichi reiterated her government would uphold the principles.
6. On South Korea and Taiwan’s past nuclear-weapons programs: the National Defense University’s WMD Center describes both as major Cold War rollback cases in which U.S. allies reversed nuclear-weapons programs under strong American pressure.
National Defense University WMD Center
7. On Ukraine’s nuclear disarmament: Arms Control Association summarizes Ukraine’s inherited Soviet nuclear arsenal, its return of nuclear warheads to Russia by 1996, and its accession to the NPT as a non-nuclear-weapon state.
8. On Iran: Arms Control Association states that Iran is not a nuclear-weapons state and separately summarizes the 2025 Israeli and U.S. strikes on Iranian nuclear facilities.
Arms Control Association, Iran profile
Arms Control Association, strikes on Iran’s nuclear program
9. On NPT withdrawal and North Korea: Arms Control Association’s NPT timeline explains Article X’s withdrawal standard and notes North Korea’s 2003 withdrawal from the treaty.
Arms Control Association NPT timeline
10. On IAEA safeguards non-compliance: the IAEA’s safeguards legal framework explains the Board of Governors’ role in reporting non-compliance to the UN Security Council and General Assembly; the IAEA’s 2003 DPRK resolution is an example of such a Board action.
IAEA safeguards legal framework
11. On France’s 2026 nuclear-cooperation initiative: AP and The Guardian reported that President Macron announced deeper nuclear-weapons cooperation with eight European allies while keeping final decision-making over French nuclear use under French presidential control.