September 2026
By Amy J. Nelson 

For the first time since the early 1970s, the United States and Russia have no legally binding limits on their strategic nuclear forces. Four months before the New Strategic Arms Reduction Treaty (New START) expired, Russian President Vladimir Putin offered to observe the treaty’s central ceilings informally for one more year while the two sides worked toward a successor.¹ U.S. President Donald Trump declined. Instead, on the day the treaty lapsed—February 5, 2026—he posted on social media that “we should have our Nuclear Experts work on a new, improved, and modernized Treaty that can last long into the future.”² Trump administration officials quickly supplied the missing specifics: not a bilateral extension, but “multilateral nuclear arms control and strategic stability talks” including both Russia and China.³

It is an ambitious answer to a real problem. Judged against 80 years of arms control history, it is also the approach that is most likely to fail, for a reason that runs deeper than great-power rivalry. The conditions that make a comprehensive settlement most necessary are the same ones that make it least attainable. That paradox has governed this field since 1945, and it points to a way through.

Not a Pause

New START’s verification architecture—data exchanges, on-site inspections, and telemetry sharing—was built for a bipolar world: two comparably sized arsenals, one shared interest in predictability. That world is gone, and it went in three directions at once. Russia curtailed inspections and data declarations in 2023, well before the treaty formally lapsed. China’s arsenal has grown entirely outside any transparency regime, from roughly 260 warheads a decade ago to an estimated 620 today,⁴ with the Pentagon projecting more than 1,000 by 2030.⁵ And commercial satellite imagery and open-source analysis now do quietly what treaty inspectors used to do openly.⁶

Of the three changes, open-source visibility is the least appreciated and most consequential. Arms control’s core trade was visibility for restraint: Each side accepted intrusive disclosure because the disclosure went to a single, symmetrical adversary bound by the same limits. In a world with three main competing powers, that logic inverts. What Washington reveals to reassure Moscow is also read in Beijing, by a state bound by nothing. Verification has become a channel through which information reaches a party that never signed, and the governing question is no longer how much to disclose but to whom, about what, and in what sequence.

The institutional evidence arrived on schedule, and at the venue where the Trump administration was making its own case. Opening the nuclear Nonproliferation Treaty (NPT) 2026 Review Conference in New York in late April, U.S. Assistant Secretary of State Christopher Yeaw argued that all five nuclear-weapon states recognized under the NPT bear Article VI responsibilities, and that this was precisely why Washington has proposed multilateral strategic stability and arms control talks.⁷ China’s director-general for disarmament, Sun Xiaobo, answered that the proposal was “in essence an attempt [by the United States] to shift its special and primary responsibilities as a country with the largest nuclear arsenal onto others.” China, he said, “has no interest in them.”⁸ In May, the United States, the United Kingdom, and France issued a joint statement urging all five NPT nuclear-weapon states to engage. Russia and China were not party to that statement.⁹

What the five managed to do together is more revealing. Divided on nearly everything else, they worked in concert to strip the draft review conference outcome document of the specific commitments that would have obliged them to reduce the number and salience of their weapons.¹⁰ The conference closed May 22 without consensus—the third consecutive failure, and 16 years since states-parties last agreed on a final document.^{11 12} This outcome proved that these five states can coordinate, but what they coordinate on is the avoidance of specific obligations. None of this reflects a pause. A pause implies something that can be resumed, and the conditions under which New START was written are not returning.

Is There Anything New Here?

The instinct to go multilateral is not new, as the record shows. British Defense Secretary Des Browne proposed five-party talks at the Conference on Disarmament in Geneva in 2008, and the P5 process, a consultative mechanism initiated in 2009 to facilitate cooperation among the NPT’s five nuclear-weapon states, has met, on and off, ever since.¹³ Eighteen years in, the process has produced joint statements on strategic stability and support for the Comprehensive Test Ban Treaty.¹⁴ It has not produced a single limit on anyone’s nuclear forces. What happened in New York this year was not an aberration. It was the process working as it always has.

Alexey Arbatov, a Russian nuclear expert, laid out the structural reason for that record in June 2026: The United States, the UK, and France are allies with no interest in constraining one another, while Russia and China have voiced no mutual concern about each other’s arsenals since the late 1980s.¹⁵ Mutual deterrence and rough parity are what made the Strategic Arms Limitation Talks (SALT I) and the Intermediate-Range Nuclear Forces (INF) Treaty, both between the United States and the Soviet Union, tractable. Without them, a shared ceiling has nothing to anchor it.

The Trump administration’s declared ambition—a new treaty that would mean “taking into account for all Russian nuclear weapons, both novel and existing strategic systems, and addressing the breakout growth of Chinese nuclear weapons stockpiles”¹⁶—asks five states with five different threat perceptions to accept a single common framework. The one state whose growth supplies the rationale (China) has already declined and pointed the United States back toward Russia.

Arbatov is right, and his argument can be carried further. Parity of weapons stockpiles explains who can get to the table. It does not explain what happens once they sit down. Across negotiations—from the 1946 Baruch Plan through the 2019 Hanoi summit to the collapse of the 2015 Iran nuclear deal, formally known as the Joint Comprehensive Plan of Action (JCPOA), and the maximalist demands that followed—the variable that mattered most was not whether talks were bilateral or multilateral, nor even whether the parties held comparable arsenals; it was how each side chose to handle irreducible uncertainty about the other’s intentions and capabilities.¹⁷ States that tried to eliminate that uncertainty in one comprehensive settlement, requiring the other side to resolve every open question before any cooperation could begin, reliably failed. States that managed uncertainty incrementally, trading narrow and verifiable steps that left each side’s core deterrent intact, reliably succeeded. SALT I capped select delivery systems, not entire arsenals. The INF Treaty eliminated one symmetrically threatening class of missiles, not nuclear weapons generally. Even the JCPOA, an unusually ambitious deal, constrained Iran’s program rather than eliminating it outright, and it held as long as it did precisely because it left the harder questions of Iranian intentions unresolved rather than pretending to answer them.

The intuition to seek narrower solutions is correct. Any durable future framework must now account for more sources of uncertainty than New START ever did: Russian modernization and novel systems on one side, Chinese nuclear stockpile growth on the other. Although the diagnosis is right, the implementation repeats the nuclear arms control field’s oldest error: bundling every source of uncertainty into a single comprehensive settlement, restated for a five-party table instead of a two-party one.

A genuinely new idea follows from my forthcoming book’s central argument: that arms control has succeeded when it managed irreducible uncertainty and failed when it tried to eliminate it.¹⁸ That logic persists. In the current new treaty-thin environment, where no verification regime remains to convert disclosure into assurance, managing uncertainty means governing it directly. This is what I have called “disciplined ambiguity”—the deliberate, bounded management of uncertainty to signal intent without exposing operational detail.¹⁹ Its premise is that not all uncertainty is alike. Uncertainty about political commitments and escalation thresholds breeds alarm and forces inference under pressure. Uncertainty about operational specifics—patrol patterns, deployment configurations, and response options—complicates an adversary without obscuring intent. The discipline lies in telling the two apart: clarifying commitments while preserving indeterminacy precisely where precision would create vulnerability.

In practice, this means controlled disclosure of new systems without deployment timelines or operating parameters, reciprocal notification of major exercises and tests, and communicated red lines, identifying which actions would be treated as destabilizing. This may seem like a downshift from arms control. It is, instead, what is achievable today. It does not require five capitals to agree on a shared ceiling. Instead, disciplined ambiguity requires that each dyad of nuclear powers decides separately and sequentially what to reveal, what to withhold, and how to signal intent. It is not a substitute for agreement but a practice for the interval before one—arguably a harder discipline to sustain than signing a treaty, but one that states can begin practicing now without waiting for China to change its mind about parity.

Who Leads, and Toward What?

Who, if anyone, is positioned to lead arms control’s next era, and toward what end? The Trump administration’s messaging to date provides no answers to these questions. In July 2025, Trump himself reportedly expressed interest in maintaining only New START’s central limits; by February 2026, he had rejected that course in favor of a treaty that no one in his government could yet describe.²⁰ One senior U.S. State Department official has said that a new treaty must capture the full scope of Russian and Chinese arsenals.²¹ Another official, asked directly whether an informal understanding with Russia to hold force levels steady already existed, said only, “I know of no such agreement.”²² Trump told an interviewer in January that he would “do a deal with Russia without the Chinese,”²³ if he had to—a bilateral instinct dressed in multilateral language. When Trump and Chinese President Xi Jinping met in Beijing in May, a White House fact sheet released afterward claimed a shared commitment to denuclearizing North Korea; it said nothing about China’s own arsenal because Chinese negotiators declined to discuss it.²⁴ Washington has announced a negotiation it cannot yet describe with a partner that has not agreed to come.

That vacuum is precisely where the hardest question lives: Does the world end up with most nuclear weapons managed by a handful of strongmen rather than by institutions? In one narrow sense, this has always been the case: The authority to launch nuclear weapons everywhere rests with a small number of executives, elected or not. What is new is the source of the constraint on those executives. For five decades, it was institutional: negotiated by professionals, ratified by legislatures, verified by inspectorates. The risk now is that constraint becomes personal, negotiated leader to leader, deal to deal, with no institutional commitment or memory that survives a change of government.

A version of this scenario is already playing out. The June 2026 memorandum intended to end hostilities between the United States and Iran was negotiated bilaterally,²⁵ outside the P5+1 structure (China, France, Germany, Russia, the UK, and the United States) that produced the JCPOA, and its nuclear provisions run to a single vague paragraph with no verified baseline and no snapback mechanism comparable to the one built into the 2015 Iran nuclear deal. It postponed the nuclear question rather than answered it.²⁶ If the multilateral nuclear talks that Trump has proposed follow the same model—summit-driven, thinly staffed, dependent on the goodwill of whoever occupies the Kremlin, Zhongnanhai, and the White House—then, at a given moment, frameworks will indeed have given way to personalities. Institutions that once absorbed and outlasted individual leaders are being replaced by arrangements that will not survive a single election.

What Should Happen Instead

None of this means multilateral engagement is worthless, only that it is being asked to do the wrong job. The P5 process has an 18-year record of producing shared vocabulary and doctrinal exchange, but no numerical limits; it should be tasked with exactly that kind of norm-building, including work on crisis communication and testing moratoria, rather than the kind of treaty text that it has never been able to deliver. Alongside that process, Washington should pursue narrow, sequenced, bilateral tracks. With Moscow, the priority should be an informal understanding to hold deployed strategic force levels steady while negotiators work toward a successor arrangement, paired with reciprocal transparency on nonstrategic nuclear weapons, long-range conventional strike systems, and novel strategic capabilities such as nuclear-powered cruise missiles and undersea autonomous systems.

With Beijing, the emphasis should remain on risk reduction rather than parity. This approach would include advance notification of ballistic missile launches and major strategic exercises; protocols governing dual-capable missile operations; exchanges on launch-on-warning and command-and-control practices; crisis communication procedures; and transparency measures regarding the expansion of China’s missile silo fields. Both tracks should apply the incremental logic to which my research points: narrow, verifiable, sequenced steps focused on the capabilities most likely to lead to miscalculation, rather than comprehensive settlements that ask each side to resolve every outstanding issue before either side gains anything.

Trump is right that the New START era cannot simply be extended by another five years; the treaty itself forecloses that option, and the strategic environment for which it was designed no longer exists. He is also right that China’s arsenal cannot be indefinitely excluded from any conversation about strategic stability, but recognizing the problem is not equivalent to identifying the solution. The solution on offer, a single comprehensive multilateral treaty, repeats the very error that has frustrated most major arms control initiatives since 1945. A framework rebuilt instead on incremental, disciplined management of uncertainty, pursued through parallel bilateral channels rather than all-or-nothing negotiation, has a far stronger historical record.

Whether this administration or the next chooses the harder and less photogenic path will determine whether the coming years mark the start of a new arms control era or simply the interval between frameworks, filled, as such intervals always are, with force and the private judgment of whoever holds the launch codes.

ENDNOTES

1. Vladimir Putin, “Meeting with permanent members of the Security Council,” press release, September 22, 2025.

2. Donald Trump, “The United States is the most powerful Country in the World. I completely rebuilt its Military in my First Term,” Truth Social, February 5, 2026.

3. Congressional Research Service, “U.S.-Russian Nuclear Arms Control: Overview and Potential Considerations for Congress,” February 26, 2026.

4. Hans M. Kristensen, Matt Korda, Eliana Johns, and Mackenzie Knight-Boyle, “Chinese nuclear weapons, 2026,” Bulletin of the Atomic Scientists, Vol. 82, No. 4 (2026), pp. 293-322.

5. U.S. Department of Defense, “Annual Report to Congress: Military and Security Developments Involving the People’s Republic of China,” December 23, 2025.

6. Hans M. Kristensen, and Matt Korda, “China Is Building A Second Nuclear Missile Silo Field,” Federation of American Scientists, July 26, 2021.

7. Christopher Yeaw, “Statement by the United States to the NPT Review Conference,” U.S. Mission to International Organizations in Geneva, April 29, 2026.

8. Sun Xiaobo, “Statement by Sun Xiaobo, Director-General of the Department of Arms Control of the Foreign Ministry of China at the General Debate of the Eleventh Review Conference of the Parties to the Treaty on the Non-Proliferation of Nuclear Weapons (NPT),”Ministry of Foreign Affairs of the People’s Republic of China, April 29, 2026.

9. “P3 Joint Statement for Main Committee I 2026 Review Conference of Treaty on the Non-Proliferation of Nuclear Weapons,” Permanent representation of France to the Conference on Disarmament, May 1, 2026.

10. Libby Flatoff, and Daryl G. Kimball, “2026 NPT Review Conference Stymied by Disputes,” Arms Control Today, Vol. 56 (June 2026).

11. United Nations, “Review Conference Ends without Consensus Outcome amid Rising Nuclear Risks,” meetings coverage, May 22, 2026.

12. United Nations, “(Informal Consultations) 2026 Review Conference of the Parties to the Treaty on the Non-Proliferation of Nuclear Weapons,” March 3, 2026.

13. Maximilian Hoell, and Andreas Persbo, “Overcoming disunity: Reinvigorating the P5 Process a decade on,” report, European Leadership Network, July 30, 2020.

14. Arms Control Association, “P5 Reiterate Commitment to CTBT,” July 1, 2011.

15. Alexey Arbatov, “The Puzzle of Multilateral Arms Control,” Arms Control Today, Vol. 56 (June 2026).

16. Thomas G. DiNanno, “Statement to the Conference on Disarmament,” U.S. Department of State, February 6, 2026.

17. Amy J. Nelson, The Arms Control Paradox (Stanford University Press, 2026).

18. Ibid.

19. Amy J. Nelson, “After New START: The Case for Disciplined Ambiguity,” New America, February 23, 2026.

20. Trump, “The United States is the most powerful Country in the World.”

21. Olivia Le Poidevin, “US meeting Russian and Chinese delegations for nuclear arms control talks, official says,” Reuters, February 24, 2026.

22. Alex Raufoglu, “New START: No ‘Gentleman’s Agreement’ Between US, Russia On Expired Nuclear Treaty,” Radio Free Europe/Radio Liberty, February 17, 2026.

23. Donald Trump, “Two Hours, Scores of Questions, 23,000 Words: Our Interview With President Trump,” interview by Zolan Kanno-Youngs, Tyler Pager, Katie Rogers, and David E. Sanger, The New York Times, January 11, 2026.

24. The White House, “Fact Sheet: President Donald J. Trump Secures Historic Deals with China, Delivering for American Workers, Farmers, and Industry,” May 17, 2026.

25. Abigail Williams, “Text of the Iran-U.S. memorandum of understanding,” NBC News, June 17, 2026.

26.  Amy J. Nelson, “What the Iran MOU Gets Right, Gets Wrong, and Leaves Unresolved,” New America, June 25, 2026.

September 2026
By Zohaib Altaf

In April 2026, the U.S. Air Force paired three nuclear energy reactor developers with Buckley Space Force Base in Colorado, Malmstrom Air Force Base in Montana, and Joint Base San Antonio in Texas. The projects remain subject to siting, environmental, licensing, and contracting decisions, but the service expects deployment by 2030 or earlier. A separate Air Force pilot project at Eielson Air Force Base in Alaska is also advancing under a commercial ownership model.¹ Military microreactors have moved beyond studies and concept papers. They now have prospective sites, vendors, and target dates.

The broader policy is even more ambitious. A May 2025 executive order directed the Defense Department, acting through the Army, to establish a formal program for nuclear power both at permanent military installations and in support of forces and weapon platforms during military operations. Separately, it required the Army to begin operating an Army-regulated reactor at a domestic base by September 30, 2028. The Army’s Janus Program is intended to supply power to installations and mission-critical systems including communications networks, weapons systems, and command nodes. On August 26, the army announced the selection of five nuclear energy vendors and five initial installations under the Janus program, with up to $2.2 billion mile-stone based government funding over fiscal year 2027-2031. The prototypes will be contractor- owned and operated and the army expects more than 20 microreactors across military installations supported by the government and private investment.² Project Pele, the Pentagon’s transportable microreactor prototype, is designed to generate at least 1.5-megawatts and received its full core of high-assay low-enriched uranium fuel in late 2025.³

None of these programs is a nuclear weapon project, and the reactors now being considered for U.S. bases are not forward-deployed battlefield systems. Their immediate purpose is energy resilience. Yet the distinction between an ordinary power plant and military capability begins to narrow when a reactor supplies mission-critical communications, sensors, command systems, or weapons infrastructure. If other states follow the U.S. lead, a familiar piece of military infrastructure may become a nuclear site as well.

The case for microreactors is credible. So is the need to govern them before deployment becomes routine. Existing safeguards, nuclear security rules, and the law of armed conflict address parts of the problem, but they were developed around civilian power, naval propulsion, and large nuclear facilities. They do not map neatly onto small, land-based reactors that may be commercially operated, use fuel enriched to higher levels than conventional power-reactor fuel, connect to military networks, and face attack in wartime. The answer is not a ban; it is a set of rules that preserves the operational value of these systems without allowing military status to become a blanket justification for secrecy.

From Energy Resilience to Military Dependence

The military demand is easy to understand. The Pentagon’s 2023 Operational Energy Strategy warns that the U.S. homeland can no longer be treated as a sanctuary and that logistics will be contested. Fuel depots, ports, pipelines, power grids, and convoys are vulnerable to long-range strikes, cyber operations, and supply disruptions.⁴ At the same time, radar, data processing, air defense, electronic warfare, and prospective directed-energy systems are increasing the demand for reliable electricity.

Microreactors offer steady power without daily fuel deliveries. Many designs are factory fabricated, can operate independently of a large grid, and are intended to run for years before refueling. Project Pele, for example, is designed to fit into four standard shipping containers and generate at least 1.5 megawatts of electricity for up to three years. Its demonstration at Idaho National Laboratory is meant to test whether a reactor can be assembled, operated, shut down, and moved safely.⁵ Fixed reactors under the Air Force’s installation program follow a different model: Commercial firms may build, own, operate, and decommission them while selling power to the military base.

These differences are not administrative detail. A commercially owned reactor, licensed by a civilian regulator and supplying a domestic military base through a microgrid, presents one governance problem. An Army-regulated reactor assigned to operational energy and integrated with military systems presents another. A transportable reactor sent abroad would mean additional complications, including host-state consent, nuclear transport, emergency response, and wartime protection. Public policy still tends to group all three of these reactors under the single label of “military microreactor.”

The Safeguards Question, Properly Framed

The safeguards issue needs precision. The five nuclear-weapon states recognized by the nuclear Nonproliferation Treaty (NPT), including the United States, have voluntary-offer agreements with the International Atomic Energy Agency (IAEA). Non-nuclear-weapon states-parties to the treaty generally have comprehensive agreements requiring IAEA safeguards on nuclear materials used in peaceful nuclear activities.⁶ Paragraph 14 of the IAEA’s model comprehensive safeguards agreement allows a non-nuclear weapon state to seek temporary non-application of safeguards to nuclear material used in a non-explosive military activity. Because the United States operates under a voluntary-offer agreement, its current reactor projects do not trigger this procedure. The potential verification gap would arise if a non-nuclear-weapon state developed or imported a military reactor and sought to remove its fuel from safeguards under paragraph 14.

Paragraph 14 of INFCIRC/153, the standard blueprint for an IAEA safeguards agreement, provides a procedure for such a case. A state is required to inform the IAEA of the activity, affirm that the material will not be used to produce a nuclear weapon or other nuclear explosive device, and conclude an arrangement defining the period or circumstances in which safeguards will not apply. The arrangement requires approval by the IAEA Board of Governors, and safeguards must resume when the material returns to peaceful use.⁷

Military use does not automatically require non-application of safeguards. A state may leave the material under safeguards. For a fixed land-based reactor with sealed fuel and a known location, continued safeguards may be far more practical than for a submarine whose movements, design, and operating schedule are classified. This should be the starting presumption. A state seeking a paragraph-14 arrangement for a land-based reactor should have to identify the specific operational information that ordinary safeguards would expose and why managed access or safeguards by design cannot protect it.

The IAEA’s negotiations with Australia over naval nuclear propulsion, under a deal in which the United Kingdom and the United States plan to supply Australia with nuclear-powered submarines, provide a useful, although incomplete, experience. As of November 2025, the agency and Australia were still discussing advance notification, reporting, verification before material entered a paragraph-14 arrangement, voluntary transparency during the excluded period, and the point at which safeguards would resume.⁸ Land-based reactors should not simply inherit a naval exception. Their fixed location and different operating models may permit stronger verification.

Fuel makes the question more consequential. High-assay low-enriched uranium (HALEU) contains more than 5 percent but less than 20 percent uranium-235. It is not highly enriched or weapons-grade uranium, but less additional enrichment work is required to reach weapons-grade levels than from the fuel used in most commercial reactors. Many advanced reactor designs use HALEU to obtain smaller cores and longer operating cycles; Project Pele’s environmental impact statement analyzed a core containing up to 400 kilograms of HALEU in tri-structural isotropic (TRISO) fuel.⁹ Wider use would create larger inventories moving through enrichment, fabrication, transport, military, and contractor-controlled facilities.

Safeguards are only part of the gap. The Convention on Nuclear Safety applies to land-based civil nuclear power plants. The amended Convention on the Physical Protection of Nuclear Material covers material and facilities used for peaceful purposes and expressly excludes nuclear material used or retained for military purposes and facilities containing such material.¹⁰ National regulation and IAEA guidance can still provide rigorous protection, but international assurance will depend heavily on how each state classifies the reactor. A system should not lose transparency, peer review, or security discipline merely because its electricity serves a defense mission.

A Nuclear Site in the Targeting Process

Wartime status is the second major problem. Article 56 of Additional Protocol I gives special protection to dams, dykes, and nuclear electrical generating stations when an attack may release dangerous forces and cause severe civilian losses. For a nuclear station, the protection may cease only if it supplies electricity in regular, significant, and direct support of military operations and an attack is the only feasible way to end that support. Even then, the protocol requires all practical precautions to avoid releasing dangerous forces.¹¹

The provision leaves difficult questions for microreactors. Its text contains no generating-capacity threshold, for instance, but a small, sealed reactor may not resemble the large station the negotiators had in mind. A fixed reactor serving an entire base could fit the ordinary meaning more readily than a transportable unit attached to a radar or command post. Mixed civilian and military loads would further complicate the assessment.

Article 56 also is not the whole law. It binds states-parties to Additional Protocol I; the United States is not a state-party and does not accept the article’s special protection as customary international law. The U.S. position is that such installations may be attacked if they are military objectives, subject to distinction, proportionality, and precautions. The risk that an attack could release radiation or radioactive material and cause severe civilian harm would weigh heavily in those assessments.¹² Any international policy must acknowledge this legal divergence rather than present Article 56 as a universal prohibition.

Reactor design can reduce the expected consequences of attack. The use of TRISO fuel, an actual passive shutdown or the capacity for a passive shutdown, underground cabling, and physical separation may all limit risk, but they do not settle target status. An attacker might strike the substation, control link, cooling equipment, or supported military system instead of the reactor vessel. Damage could still cause radioactive release, interrupt monitoring, force evacuation, or create uncertainty about the reactor’s condition. Even when civilian harm is limited, reports of a strike on “nuclear infrastructure” may generate political pressure and escalation before reliable technical information is available.

Rules Before Reactors Become Routine

Four measures would address the most immediate risks without blocking microreactor deployment. First, states should publish a basic classification for every military reactor program: fixed or transportable; domestic installation or operational use; civilian-licensed or military-regulated; government- or contractor-operated; fuel type and enrichment band; and whether the reactor is intended to supply civilian loads, general base power, or specified mission systems. Exact vulnerabilities need not be public, but the regulatory authority and emergency responsibilities should be.

Second, the IAEA Board of Governors should commission a technical study of land-based reactors used in non-explosive military activities. The study should establish a presumption of continued safeguards for fixed reactors in non-nuclear-weapon states. Where paragraph 14 is genuinely necessary, arrangements should be material-specific and define entry and exit points, duration, reporting, reapplication of safeguards, and verification measures that protect classified information while maintaining confidence against diversion. The same work should develop safeguards-by-design guidance for sealed cores, remote monitoring, and long-life fuel.

Third, supplier states should make HALEU transfers for military reactors conditional on cradle-to-grave material accountancy, appropriate physical protection, prior consent for retransfers or changes in use, and credible plans for spent fuel return or disposition. Confidential reporting to the IAEA can protect commercially sensitive data. “National security” is not an adequate accounting category.

Fourth, siting and system design should anticipate war, not only accidents. When feasible, reactors should be physically separated from high-value command and weapons systems, and microgrids should permit military loads to be isolated without attacking the reactor. Reactor control and safety systems should be segregated from operational command networks, with independent shutdown and authenticated radiological reporting. Additional Protocol I already requires its parties to avoid locating military objectives near protected works where feasible and to take precautions for civilians under their control.¹³

Crisis communication completes the package. States operating reactors near potential conflict zones should establish channels for reporting shutdowns, attacks, cyber incidents, and radiation readings to the IAEA, host authorities, and adversaries where appropriate. India and Pakistan’s 1988 agreement prohibiting attacks on each other’s nuclear installations shows that rivals can exchange protected site information through confidential channels. They carried out their annual exchange again January 1, 2026.¹⁴ The agreement cannot simply be copied for every region, and states will hesitate to disclose military locations. It nevertheless offers a precedent for bounded notification without public exposure.

An Early Choice

Military microreactors could improve resilience, reduce dependence on vulnerable fuel deliveries, and support remote operations. Their benefits should not be dismissed because nuclear technology is involved. Nor should their small size obscure the institutional change they introduce. A base reactor can sit simultaneously inside an energy system, a nuclear fuel cycle, a commercial contract, and a military target complex.

Governments still have time to decide how those roles will be separated and supervised. The United States can set the first useful precedent by explaining which of its programs will be civilian-licensed, which will be Army-regulated, how fuel will be accounted for, and how reactor sites will be protected during conflict. The IAEA can begin the harder work before a non-nuclear-weapon state requests a paragraph-14 arrangement for a land-based military reactor.

Regulation will become more difficult once reactors are installed, supply contracts are locked in, and militaries depend on them for critical missions. The responsible course is neither prohibition nor silence. It is to establish safeguards, disclosure requirements, design standards, and crisis-management protocols while the technology is still taking shape.

ENDNOTES

1. U.S. Department of the Air Force, “DAF announces next steps in Advanced Nuclear Power for Installations initiative,” April 22, 2026; U.S. Space Force, “Buckley SFB, Malmstrom AFB selected for Advanced Nuclear Power for Installations program,” April 8, 2026; Air Force Civil Engineer Center, “DAF ignites energy innovation with microreactor pilot at Eielson AFB,” September 11, 2025.

2. U.S. Army Communication and Outreach Office, “Army Selects Vendors, Sites for Nuclear Microreactors,” August 27, 2026.

3. U.S. President Donald Trump, Executive Order 14299: “Deploying Advanced Nuclear Reactor Technologies for National Security,” May 23, 2025, Sec. 3; U.S. Army, “The Janus Program: Fueling the Army’s future with resilient, on-demand nuclear energy,” October 14, 2025; Idaho National Laboratory, “INL advances Department of War’s Project Pele demonstration microreactor with first TRISO fuel delivery,” December 2, 2025.

4. U.S. Department of Defense, “Department of Defense Operational Energy Strategy,” May 2023, pp. 1-3.

5. U.S. Department of Defense, “DoD Breaks Ground on Project Pele: A Mobile Nuclear Reactor for Energy Resiliency,” September 24, 2024; BWX Technologies, “Project Pele Begins Taking Shape With Start of Core Manufacturing,” July 24, 2025.

6. International Atomic Energy Agency, “Safeguards agreements,” accessed August 22, 2026.

7. Laura Rockwood, “The Legal Framework for IAEA Safeguards,” IAEA, 2013, p. 12; IAEA, “IAEA Safeguards Glossary,” 2022, p. 22.

8. IAEA Board of Governors, “Naval nuclear propulsion: Australia,” report by the director-general, GOV/INF/2025/12, November 12, 2025, pp. 3-5.

9. U.S. Department of Energy, “What Is High-Assay Low-Enriched Uranium (HALEU)?” December 3, 2024; National Academies of Sciences, Engineering, and Medicine, “Merits and Viability of Different Nuclear Fuel Cycles and Technology Options and the Waste Aspects of Advanced Nuclear Reactors,” 2023, pp. 181-182; U.S. Department of Energy, “Final Environmental Impact Statement for Construction and Demonstration of a Prototype Mobile Microreactor: Environmental Impact Statement,” February 2022, pp. S-9 to S-10.

10. IAEA, “Convention on Nuclear Safety,” July 5, 1994, p. 2; IAEA, “Amendment to the Convention on the Physical Protection of Nuclear Material,” p. 3, July 8, 2005.

11. International Committee of the Red Cross, “Protocol Additional to the Geneva Conventions of August 12, 1949, and Relating to the Protection of Victims of International Armed Conflicts (Protocol I),” June 8, 1977.

12. U.S. Department of Defense, “Law of War Manual,” July 2023, pp. 278-279; Office of the General Counsel, U.S. Department of Defense, “Official Treaty Documents Related to the Law of War,” accessed August 22, 2026.

13. International Committee of the Red Cross, Additional Protocol I, articles 56-58; Vasiliki Tafili “New CRP: Enhancing Computer Security of Small Modular Reactors and Microreactors,” IAEA Department of Nuclear Safety and Security, March 4, 2024.

14. Indian Ministry of External Affairs, “Agreement on the Prohibition of Attack Against Nuclear Installations and Facilities Between the Republic of India and the Islamic Republic of Pakistan, December 31, 1988; Indian Ministry of External Affairs, “India and Pakistan exchange list of Nuclear Installations,” January 1, 2026.