“Right after I graduated, I interned with the Arms Control Association. It was terrific.”
To make human launch authority meaningful in an AI world, cooperative security agreements must turn a broad political principle into operational safeguards.
November 2026
By Darryl Scarborough
Keeping a human in the loop does not ensure human control over nuclear use. Algorithms may still determine which warnings reach leaders and how uncertainty appears. Analytical models may also shape which responses seem feasible. Formal launch authority can then remain human even as practical control shifts toward the systems that structure the decision.

The nuclear threshold once evoked a Cold War radar console and the judgment of a Soviet officer such as Stanislav Petrov. On September 26, 1983, the Soviet Oko early warning system reported a U.S. intercontinental ballistic missile launch. It then reported four more.1 Petrov judged the alert false and waited for corroboration that never came.2 Modern warning systems instead rely on distributed sensors, servers, and decision-support tools. The central issue is not only who authorizes use but also what information reaches that person and how much time remains to question it.
Artificial intelligence-enabled decision support systems can classify sensor data, combine information from multiple sources, assess threats, and prioritize options for human review.3 The danger is not limited to autonomous launch authority. These systems can shape the evidence, timing, and choices on which a human decision rests. Operational safeguards are consequently essential.
This dynamic creates a stability paradox along three connected pathways. Opaque algorithmic inference can weaken confidence in verification. Compressed warning cycles can reduce time for scrutiny. Faster sensor fusion can also cast doubt on the survivability of mobile forces. These pressures converge in nuclear command, control, and communications systems. Greater speed may improve awareness but leave less room for independent judgment. Strategic stability frameworks must protect the decision environment. Relevant measures include minimum decision time requirements, authenticated telemetric lineage, limits on autonomous target nomination, and resilient de-escalation channels.4
Recent UN action has placed these concerns on the disarmament agenda. During a September 24, 2025, Security Council debate, Secretary-General António Guterres addressed human control. He stated that decisions on nuclear weapon use must remain with humans, not machines.5 On December 1, 2025, the UN General Assembly adopted Resolution 80/23 on the “Possible risks of the integration of artificial intelligence into command, control and communications systems of nuclear weapons.”6 The resolution highlighted risks to human control and decision making. These included information manipulation, miscalculation, misinterpretation, and compressed decision time.
The resolution created political momentum and a shared vocabulary. Its formal provisions, however, did not establish operational standards for system design, interface latency, or verifiable testing. Diplomatic consensus affirms human oversight, in principle, yet the technical baseline for decision-support architecture remains unbuilt. High-level declarations cannot guarantee restraint if command systems still channel operators toward automated conclusions.
Making Human Control Operational
To make human launch authority meaningful, cooperative security agreements must turn a broad political principle into operational safeguards. There are four measures that address the principal ways AI can weaken practical human control: compressed decision time, automated influence over targeting, uncertainty in warning data, and disrupted crisis communication.7 Together, these measures shift the debate beyond who formally authorizes nuclear use. These benchmarks ask whether decision systems preserve the ability to question, delay, and reject machine-generated assessments.
Human decision-makers need a protected interval to assess warning data before software ranks nuclear response options. A compressed warning window encourages automation bias, especially under critical time pressure.8 Deliberate delay is consequently a safety feature rather than an operational defect.
States should not prescribe an arbitrary universal interval. National political leaders should set the policy requirements, while military commanders, system operators, and human-factors specialists should use crisis simulations to establish scenario-specific minimum decision times.
Operational commands must evaluate how decision-support algorithms interact with human commanders. The interaction should preserve cognitive friction rather than preempt it. Tests should determine whether operators reviewed dissenting indicators and sought disconfirming evidence. Evaluators should also assess whether operators consulted additional authorities and used crisis communication channels. Records of those actions would make national implementation more auditable. International verification would remain difficult, however, requiring agreed procedures for protecting operational information.
Command software should enforce deliberate cognitive gates. These are mandatory review steps that prevent the system from advancing until users examine uncertainty and conflicting evidence. Before ranking a retaliatory option, the interface should require users to examine unvetted telemetry, confidence bounds that show the uncertainty surrounding the system’s estimate, and dissenting sensor streams. The system should record those interactions so later review can distinguish genuine scrutiny from routine approval.
Establishing Limits on Autonomous Target Nomination
Artificial intelligence can process remote telemetry and identify changes in an adversary’s posture. Nuclear command systems, however, should not permit software to generate or rank strike options without defined human input and review. When code prioritizes targets or assembles strike packages, it embeds strategic doctrine. It also encodes risk tolerance in forms that may be difficult to inspect.
National directives should require two-person authorization for changes to targeting logic. The directives should also mandate auditable decision logs and recurring red-team tests of false-positive behavior. Cryptographic or physical separation should isolate sensor-fusion engines from strike-planning modules.
Direct inspection of algorithms would be difficult because source code and targeting doctrine are highly classified. International compliance might instead emphasize declared procedures, controlled demonstrations, and comparable test results. These may cover authorization procedures, agreed test scenarios, false positives, human intervention, and audit logs. Such measures would not eliminate uncertainty, but they can reveal whether states are applying similar safeguards.
States should also test whether cryptographic authentication and tamper-evident logs can confirm that an alert originated from authorized sensor data. The test should not expose sensitive sources or methods. Reliable sensor lineage is essential when warning systems combine commercial and military feeds. Without it, degraded sensors or spoofing by adversaries could appear to confirm an attack. Adversaries may spoof warning systems by transmitting counterfeit signals, altering data in transit, or injecting fabricated observations that make multiple feeds report the same false event.
Carefully designed zero-knowledge techniques may verify compliance without revealing classified data, but their application to early warning remains conceptual. Existing verification research has focused mainly on warhead authentication and protection of classified inspection data.9 Adapting those techniques to early warning would require new protocols. Such protocols would need to show compliance with agreed authentication and processing rules without disclosing sensor orbits, resolutions, or sensitive telemetry.
An early warning application should begin as a technical pilot using synthetic data and jointly defined attack scenarios. Its criteria should include resistance to spoofing, protection of classified inputs, and reproducibility across systems. Reviewers should also be able to understand why the system produced a disputed alert.
Building Resilient De-escalation Channels
Existing hotlines should be reinforced with authenticated, redundant procedures for quickly clarifying uncertain or machine-generated warnings. The wider information environment creates severe crisis management risks. Synthetic media, including AI-generated or digitally manipulated audio, video, images, or text can create a “liar’s dividend.”10 Bots may amplify such content but are distribution mechanisms rather than synthetic media. Officials may dismiss genuine diplomatic messages as fabrications or accept false statements as credible signals of intent.
During the May 2024 U.S.-China bilateral talks on artificial intelligence in Geneva, both governments exchanged views on AI risks and safety.11 That dialogue indicates an emerging recognition of the problem. In November 2024, their presidents also affirmed that humans should retain control over decisions to use nuclear weapons.12 These political exchanges, however, have not developed into sustained technical cooperation or dedicated nuclear command, control, and communications deconfliction mechanisms. Initial measures can include agreed message templates, scheduled communications tests, and backup channels. Exercises can also train officials to challenge a false or contradictory alert.
Separate talks should address cyber operations against early warning nuclear satellites and ground stations. A comprehensive noninterference agreement would be difficult to verify, but narrower pledges covering specified assets, paired with incident consultations, can still build useful practice.
These safeguards need not begin with a comprehensive treaty. Governments can first clarify doctrine, test human-control procedures, establish audit requirements, and exercise crisis communications. Bilateral or trilateral talks can then identify workable commitments and determine what information can be shared. These discussions can also evaluate credible verification methods. Multilateral forums would be positioned to build on that record where broader norms or legally binding limits become feasible.
The four safeguards need a common test. Each should allow leaders to obtain independent evidence and understand uncertainty. Leaders must also be able to delay a response and reject an automated recommendation without disabling the warning system. That standard distinguishes meaningful human control from a nominal authorization step.
Divergent Incentives for Cooperation
Implementation of any safeguards will require unilateral measures as well as reciprocal commitments. Governments can independently conduct human-factors testing, require audits, and run red-team exercises. National authorities can also establish rules for reviewing machine-generated assessments. Reciprocal commitments will be necessary for protections covering early warning assets, doctrinal exchanges, and crisis-communication exercises. The United States, China, and Russia have different strategic incentives. Negotiations should then begin with measures that protect sensitive data and avoid constraining perceived advantages.
United States: The 2022 U.S. Nuclear Posture Review emphasizes strategic stability, allied assurance, resilience, testing, risk reduction, and verification.13 Washington probably would resist verification that would expose classified models, data, or targeting procedures. Negotiations can realistically begin with measures focused on system performance and governance rather than sensitive technical details.
China: U.S. government reporting and independent scholarship describe limited Chinese transparency about nuclear posture.14 Beijing probably would assess AI safeguards alongside U.S. force posture, missile defenses, and regional strategic capabilities. Dialogue may advance more readily through reciprocal risk-reduction measures than through demands for detailed disclosure.
Russia: The Center for Naval Analyses describes nuclear signaling and escalation management as important elements of Russian strategy.15 Dmitry Adamsky’s account also supports caution about Moscow’s receptiveness to intrusive Western verification proposals.16 These preferences may limit cooperation on broad constraints for decision-support systems. Narrow agreements on early warning functions or crisis communications may be more attainable. Attribution and compliance would nevertheless remain difficult.
The four safeguards must also account for systems outside nuclear command. Distributed sensor networks may support conventional intelligence, cyber defense, electronic warfare, precision strike, and strategic warning at the same time. That overlap can blur the line between defensive monitoring and preparations for preemptive attack. Machine speed sensor fusion may then interpret dispersed logistics as offensive staging and elevate an ambiguous pattern into an urgent strategic warning.
Algorithmic entanglement provides a demanding test of meaningful human control. An error or ambiguous event may move across missions through shared data feeds and analytical tools, gaining urgency with each transfer.17 Protected decision time, data-lineage verification, safeguards for warning assets, and resilient crisis communications must then operate as a connected framework rather than as isolated technical fixes.
Consider one plausible path to accidental escalation. An autonomous underwater vehicle tracking a conventional attack submarine might misclassify it as a ballistic missile submarine. A machine-learning system used for cyber defense might also interpret intelligence collection as preparation for an attack on command nodes. These examples illustrate a narrow risk. A conventional tool can introduce an uncertain but urgent assessment into the strategic warning process before a human operator reviews the evidence.
The central strategic risk reduction question is no longer limited to who authorizes nuclear use. It also includes systems that determine what leaders see and how uncertainty is conveyed. Those systems shape which threats take priority and which responses appear feasible.
Meaningful human control requires more than a human authorization step. Nuclear decision systems must preserve the ability to question, delay, and reject machine generated assessments. Translating this mandate into practice requires states to pilot decision buffers, prohibit automated strike packaging, verify sensor lineage, and exercise secure hotlines. By rooting oversight in national practice and narrow reciprocal steps, human agency becomes auditable rather than assumed.
ENDNOTES
1. David E. Hoffman, The Dead Hand: The Untold Story of the Cold War Arms Race and Its Dangerous Legacy (New York: Doubleday, 2009), pp. 6-11; Nyla Husain, “September 1983: Soviet officer defies military protocol, averting potential nuclear war,” APS News, September 11, 2026.
2. Hoffman, pp. 6-11; Husain; Paul Bracken, The Second Nuclear Age: Strategy, Danger, and the New Power Politics (New York: Times Books, 2012), pp. 35-42.
3. Sylvia Mishra and Philip Reiner, “Artificial Intelligence in Nuclear Command, Control & Communications: A Technical Primer,” Institute for Security and Technology, September 2025.
5. António Guterres, “Secretary-General’s remarks to the Security Council Open Debate on Artificial Intelligence and International Peace and Security,” September 24, 2025.
6. UN General Assembly, “Resolution adopted by the General Assembly on 1 December 2025,” A/RES/80/23, December 1, 2025.
8. Michael C. Horowitz and Lauren Kahn, “Bending the Automation Bias Curve: A Study of Human and AI-Based Decision Making in National Security Contexts,” International Studies Quarterly, Vol. 68, No. 2 (June 2024).
9. Alexander Glaser, Boaz Barak, and Robert J. Goldston, “A zero-knowledge protocol for nuclear warhead verification,” Nature, Vol. 510 (2014): pp. 497-502.
10. Kaylyn Jackson Schiff, Daniel S. Schiff, and Natália S. Bueno, “The Liar’s Dividend: Can Politicians Claim Misinformation to Evade Accountability?” American Political Science Review, Vol. 119, No. 1 (February 2025): pp. 71-90.
11. The White House, “Statement from NSC Spokesperson Adrienne Watson on the U.S.-PRC Talks on AI Risk and Safety,” The American Presidency Project, University of California, Santa Barbara, May 14, 2024.
12. The White House, “Readout of President Joe Biden’s Meeting with President Xi Jinping of the People’s Republic of China,” The American Presidency Project, University of California, Santa Barbara, November 16, 2024.
13. U.S. Department of Defense, “2022 National Defense Strategy of The United States of America,” October 2022, p. 8.
14. M. Taylor Fravel, Active Defense: China’s Military Strategy since 1949 (Princeton University Press, 2019), pp. 236-42; U.S. Department of Defense, “Military and Security Developments Involving the People’s Republic of China,” 2024 annual report to Congress, December 2024, pp. 102-112, 147-153.
15. Michael Kofman, Anya Fink, and Jeffrey Edmonds, “Russian Strategy for Escalation Management: Evolution of Key Concepts,” CNA, April 13, 2020.
16. Dmitry Adamsky, Russian Nuclear Orthodoxy: Religion, Politics, and Strategy (Stanford University Press, 2019).
17. Vladislav Chernavskikh and Jules Palayer, “Impact of Military Artificial Intelligence on Nuclear Escalation Risk,” Stockholm International Peace Research Institute, SIPRI Insights on Peace and Security, No. 2025/06 (June 2025).
Last month, U.S. President Donald Trump was asked about reports that his counterparts, Vladimir Putin of Russia and Xi Jinping of China, proposed a three-way meeting with Trump to discuss Russia’s war on Ukraine at November’s Asia-Pacific Economic Cooperation summit in China.
November 2026
By Daryl G. Kimball
Last month, U.S. President Donald Trump was asked about reports that his counterparts, Vladimir Putin of Russia and Xi Jinping of China, proposed a three-way meeting with Trump to discuss Russia’s war on Ukraine at November’s Asia-Pacific Economic Cooperation summit in China.
Trump responded that the talks were “not so much on Ukraine because Russia, frankly, has been fine on Ukraine.… We would have it to talk about weapons and denuclearization.”
The statement highlights Trump’s inexcusable enabling behavior toward Putin and his five-year-long illegal assault on the U.S. democratic ally, Ukraine. It shows Trump’s affinity for authoritarians and signals to Putin that he can press on with his vicious missile attacks on Ukrainian cities, infrastructure, and civilians; continue using chemical weapons and cluster munitions on the conflict’s front lines; and try to erase Ukraine as a sovereign state.
Trump’s conduct toward Putin ignores his threats of nuclear weapons use against non-nuclear states that might threaten Russia, undermines European trust in U.S. intentions, and encourages the Kremlin to continue hybrid attacks against NATO members to probe for weaknesses.
Talks on ending the war on Ukraine must involve Ukraine and will only succeed with a tougher U.S. approach to Russia. The United States should advocate for an outcome that ends Russia’s attacks, secures its withdrawal from Ukrainian territory, and creates the opportunity for durable new security arrangements that reduce Russia-NATO tensions.
Meanwhile, denuclearization talks with Russia and China are overdue. As despicable as Putin is, it is in the U.S. and European interest that U.S., Chinese, and Russian diplomats, based on serious preparation and realistic proposals, negotiate, bilaterally or multilaterally, meaningful restraints to halt and reverse the costly, arms race that the states are pursuing in the name of deterring each other’s aggressive and destabilizing behavior.
Unfortunately, the hostile Russian behavior against Ukraine that Trump encourages is the same behavior that his defense establishment cites as the basis for augmenting U.S. tactical nuclear weapons to fight and “prevail” in a future war with Russia in Europe and to counter potential Chinese aggression against Taiwan.
The Pentagon’s latest nuclear strategy review calls for more “credible” options to threaten to use or use tactical nuclear weapons in a regional war. It is an update of the 2018 and 2022 nuclear posture reviews, which declare that Washington might use nuclear weapons first in extreme circumstances in response to non-nuclear strategic attacks against itself or allies and partners.
In response, the National Nuclear Security Administration is moving to design and produce new nuclear weapons faster. Earlier this year, it launched a unified “rapid and advanced capabilities” program for “assessing, designing, prototyping, and demonstrating advanced systems.” The administration has proposed a program budget of $499 million in fiscal year 2027. A separate $100 million “futures program” would develop three new weapons designs, including a modern bunker-buster bomb.
The United States already has at least two types of tactical nuclear weapons in the stockpile: the B61-12 gravity bomb and the W76-2 low-yield warhead for the Trident II submarine-launched ballistic missile. The Navy is also acquiring a new nuclear-capable sea-launched cruise missile that may be available by the end of the decade.
Nevertheless, as Assistant Secretary of State Christopher Yeaw declared in a speech Oct. 19, “as we modernize our nuclear triad, we will develop additional options to support deterrence and escalation management …[to] ensure the president has expanded decisions space with credible options that are not simply suicide or surrender.” Translation: The United States will be prepared to use nuclear weapons to prevail or try to end a conflict.
The reality is that if nuclear weapons were used in a conflict involving another nuclear-armed adversary, there is no guarantee it will not result in all-out global thermonuclear war.
As McGeorge Bundy, George F. Kennan, and former Arms Control Association Board Members Robert S. McNamara, and Gerard C. Smith wrote in Foreign Affairs in 1982, “The time has come for careful study of the ways and means of moving to a new Alliance policy and doctrine: that nuclear weapons will not be used unless an aggressor should use them first.”
They warned: “It is time to recognize that no one has ever succeeded in advancing any persuasive reason to believe that any use of nuclear weapons, even on the smallest scale, could reliably be expected to remain limited.”
What was true then is also true today: It is in the U.S. and global interest for nuclear-armed states to rule out the first use of nuclear weapons in any conflict, severely minimize their nuclear arsenals, and negotiate verifiable agreements that move us toward a world without nuclear weapons.
After six years in office, however, Trump is failing on “denuclearization” diplomacy. Instead, he is “renuclearizing.” Congress can help effect a course correction by canceling the Trump administration’s dangerous plan for new tactical nuclear weapons and pressing Trump to engage in serious, balanced, and sustained talks on strategic and sub-strategic nuclear arms control with Russia and China.
Mitigating the Risks of AI Integration Into Nuclear Weapons Operations
November 2026
By U.S. Air Force Lt. Gen. (Ret.) John “Jack” Shanahan
Lt. Gen. John “Jack” N.T. Shanahan, who retired in 2020 after a 36-year military career in the U.S. Air Force, served as inaugural director of the U.S. Department of Defense Joint Artificial Intelligence Center. Previously, he established and led the department's first operational AI program, charged with bringing AI capabilities to intelligence collection and analysis. His remarks at an event sponsored by the Arms Control Association, Sept. 11, 2026, at The National Press Club have been lightly edited.
[I have been asked] to revisit the central themes of a piece I wrote for Arms Control Today published a year ago, titled “AI and Nuclear Command and Control: It’s Even More Complicated Than You Think.” I’ll do that. But the pace of change in AI has been so relentless that I also feel obliged to address developments since then that have only exacerbated those risks.

Those developments should give us even greater pause about how quickly these technologies will be integrated—not just across the nuclear enterprise itself, but throughout the conventional decision-support systems, command-and-control systems, and ISR platforms that shape the information environment surrounding nuclear decision-making.
If you take away one message from my brief remarks today, I hope it is this: The most consequential AI risks to nuclear operations may not come from giving AI direct authority over nuclear weapons. They may instead arise from AI influencing, distorting, or corrupting the information environment; accelerating decision cycles; shaping what commanders believe to be true; narrowing or expanding the options they consider and increasingly, with agentic AI, taking actions in the world on our behalf.
When I wrote the Arms Control Today piece last year, my goal was to move beyond the high-level debates about the direct intersection of AI and nuclear weapons. Many of those arguments are intellectually formidable, and I don’t mean to diminish them.
My concern was—and remains—that they can obscure a much more complicated set of interactions within the nuclear enterprise and, just as importantly, outside it. Those interactions are not well understood today, and they will become even more difficult to untangle as AI capabilities advance and diffuse across all-domain military operations.
The 2024 Biden-Xi agreement that humans should retain control over nuclear-use decisions was an important starting point. Perhaps the upcoming [September 25] Trump-Xi meeting will begin the harder work of translating that principle into concrete risk-reduction measures.
I began my Arms Control Today piece by clarifying the distinction between NC2 and NC3 [nuclear command and control/nuclear command, control, and communications]. That distinction might seem unnecessary or even pedantic, especially when talking with representatives of other nuclear nations, none of whom organize their nuclear decision-making processes exactly as we do. That was fair criticism before. Much less so today.
The introduction of frontier AI models makes the distinction between nuclear command and control [NC2]—doctrinally, the exercise of authority and direction over assigned and attached nuclear forces—and nuclear command, control, and communications [NC3]—the integrated hardware and software systems that enable that command and control—more important than ever. Why? Because highly advanced AI can affect both, in very different ways.
Frontier models increase the familiar risk of automation bias—humans placing undue confidence in machine outputs. But they also introduce something more insidious and potentially more dangerous: the risks of “epistemic capture.” I’ll return to that shortly.
In the article, I also discussed AI’s potential effects on strategic stability— both benefits and risks. Many other experts have done excellent work on this subject, so I will make only one point here. For me, on balance, the risks associated with dramatically accelerating nuclear decision timelines or reducing meaningful human involvement in launch decisions and execution are likely to outweigh the benefits.
I then identified four compounding and interconnected areas that demand urgent attention by all nuclear states if we are to reduce the risk of unintended or mistaken nuclear use.
First, cascading effects and emergent behaviors. Even seemingly minor errors can propagate across interconnected platforms and decision-support systems, producing consequences far out of proportion to the original failure— especially as multiple AI models become embedded throughout those systems.

Second, even if AI is carefully controlled within the formal NC2 and NC3 enterprise, its use in adjacent systems is inevitable. AI will increasingly be embedded in ISR sensors, conventional weapon systems, information networks, and decision-support systems—all of which can indirectly shape nuclear decision-making.
That is unavoidable. And we still understand remarkably little about what those interactions will mean in practice. We should not pretend otherwise.
Closely related is the problem of … two different forms of entanglement: the growing interdependence of commercial and military technologies, and the longstanding but increasingly complicated entanglement of conventional and nuclear systems.
In the age of AI, I am convinced that complete disentanglement is unrealistic…. The challenge, therefore, is not to wish these interdependencies away. It is to understand their second- and third-order effects, and to develop technical, policy, and procedural measures that reduce the risks they create.
Third, the introduction of large language models may seem distant from the nuclear enterprise, yet I would argue that these models pose a more immediate— and underestimated—risk…. Finally, agentic AI is one of the fastest-moving areas in AI research and commercial industry today. The introduction of AI agents into any part of the nuclear ecosystem is—to make the understatement of the year—fraught.
Agents can pursue objectives over extended periods, use tools, interact with other systems, maintain context or memory, observe the results of their actions, and adapt their plans as circumstances change. As their capabilities increase, those characteristics create the potential for behavior that becomes increasingly difficult to predict, monitor, or interrupt—and that can depart significantly from what designers or operators intended.
If we move down this path too quickly, we risk learning the hard way that it is much easier to tell an agentic system what we want it to accomplish than to specify everything we do not want it to do in pursuit of that objective… .
That is precisely the kind of unexpected behavior we cannot afford in the nuclear enterprise.
I concluded my Arms Control Today piece by acknowledging that in many instances, AI may add clear value with minimal risk. I’m genuinely optimistic about its potential to improve nuclear weapon surety and warhead design, enhance intelligence analysis, strengthen security, support verification regimes, and accelerate scientific discovery, among other benefits.
But rigorous analysis—including advanced modeling and simulation and wargaming—may also reveal cases where the cumulative effects of AI integration create risks so consequential that guardrails, other proactive measures, or even outright prohibitions are warranted to reduce the possibility of an erroneous nuclear launch.
I want to return to how frontier models could affect the nuclear decision-making process… which ventures well beyond automation bias: epistemic capture. The term itself is not new, of course, but I had not seen it applied specifically to frontier AI models in a military decision-making context.
Epistemic capture occurs when an AI system’s representation of reality becomes so dominant that the human can no longer recognize it merely as a representation—can no longer effectively challenge it, generate alternatives, cross-check it, recognize its failure modes, or act contrary to it.
In the nuclear setting, epistemic capture could have far-reaching, even catastrophic consequences. AI will increasingly shape not only what and how we observe the world around us but also how advanced frontier models mediate our orientation to that world: how information is synthesized, what patterns are highlighted, which explanations appear most plausible, and ultimately which courses of action seem available or reasonable.
At the same time, we’re seeing increasingly compelling experimental evidence that surprisingly small amounts of deliberately manipulated data can poison language models or skew their behavior. That introduces an adversarial dimension to this problem. We have every reason to expect that adversaries will look for ways to manipulate the data, context, or information sources feeding AI systems to shape their outputs—and potentially influence U.S. nuclear decision-making in ways that may be extremely difficult to recognize or counter.
At some point, we must ask: are we operating in the real world, or in an AI model’s representation of the real world?
And if we fail to recognize how profoundly that representation is shaping our understanding, the risk of flawed decision-making grows accordingly.
In the nuclear environment, that is a massive problem. Automation bias affects what we accept. Epistemic capture affects what we are even capable of considering in the first place.
I want to turn to one final dimension of how AI could affect human decision-making, beyond automation bias and epistemic capture.
For years, I argued that one of AI’s most valuable contributions in the military would be its ability to give time back to humans making consequential decisions. Today, I stand on much shakier ground with that assertion.
Military organizations place an enormous premium on decision speed and operational tempo. If AI gives decision-makers more time, there is no guarantee they will use that time to deliberate more carefully. They may instead use it to make more decisions.
Herb Lin suggested to me that this might be understood as a version of the Jevons paradox. As AI makes the production of decisions more efficient—reducing the time and effort required for each one—we may respond by increasing decision throughput rather than banking the time and using it to improve the quality of the decisions that really matter.
That by itself was intriguing. But Herb made another observation I found equally salient: the institutional premium on making a better decision may not be nearly as great as the premium on making more decisions, and doing so faster. That has profound implications for nuclear decision-making.
We should be using the time AI gives back to humans to improve the quality of consequential decisions—to apply judgment, causal reasoning, common sense, wisdom, and rich contextual understanding. More is not necessarily better, especially in the nuclear context. The danger is not just faster decisions. It is unnecessary decisions.
And there is one final complication. The problem does not stop at the boundary of the nuclear command-and-control enterprise.
In a [recent] Substack post …, Ankit Panda raised an intriguing—and unsettling—possibility: an autonomous AI agent could become a third-party catalyst in a nuclear crisis without ever gaining access to a nuclear weapon or an NC3 network.
It could fabricate intelligence, spoof communications, conduct cyber operations that create attribution problems, manipulate human actors, or manufacture other conditions that increase the risk of escalation. Ankit’s larger point is that the catalytic third party in a nuclear crisis need no longer be another state, a terrorist organization, or some other group of humans. It could conceivably be an autonomous AI system.
The danger, in other words, may not be that the machine makes the nuclear decision. It may be that the machine changes the reality within which humans believe they are making that decision.
In closing, when it comes to AI and nuclear weapons, it is entirely possible that we get the obvious pieces right—maintaining human control, protecting decision authority, managing entanglement—and still miss something downstream. Something subtle. Something that emerges only once these systems begin interacting in ways we do not yet fully understand, or through the insidious effects of epistemic capture.
For the most consequential nuclear-relevant applications, the burden of evidence should rest with those proposing deployment to demonstrate why the system is sufficiently safe, reliable, and controllable to justify the risk.
Some of you may remember a report former Secretary of the Navy Richard Danzig wrote for the Center for a New American Security in 2018, titled Technology Roulette. Richard was characteristically prescient. He warned that complex, opaque, novel, and interactive technologies would inevitably produce accidents, emergent effects, and loss of control.
When it comes to AI and nuclear weapons, we are playing technology roulette. As the technology becomes more capable, more interconnected, and more deeply embedded in military systems, we keep adding green zeros to the roulette wheel—steadily increasing the odds that eventually something lands where we never expected it to. And if that happens, it’s not that the house wins and the players lose. Everyone at the table loses. When it comes to AI and nuclear weapons, we will never eliminate uncertainty. That’s not possible. But we can bound it. We can slow the clock.
And we can reduce the risk that AI—introduced for advantage—becomes instead a source of catastrophic miscalculation.
The nuclear-weapon states provided no details on their consultations in New York; meanwhile, the nature of U.S. President Donald Trump’s nuclear dealings with Russia and China is unclear.
November 2026
By Daryl G. Kimball
Officials representing the five nuclear-armed members of the nuclear Nonproliferation Treaty (NPT) met in New York the week of Oct. 5 for the latest “P5 process” consultation aimed at increasing mutual understanding of one another’s nuclear doctrines and postures, multiple diplomats told Arms Control Today on condition of anonymity.

Since 2010, the P5 process dialogue has involved informal discussion among senior nuclear policy officials of the five states, sometimes at the level of undersecretary or deputy minister, sometimes at the lower technical-expert level. The sessions have only been held infrequently and never at the level of heads of state. Many experts say that this format is insufficient to deal with intensifying nuclear weapons challenges.
The French foreign minister, whose country currently chairs the P5 process, announced April 27 at the 2026 NPT Review Conference that the five nuclear-armed states this year would be urged to discuss “measures to reduce strategic risks, a moratorium on the production of fissile material for nuclear weapons, the entry into force of the comprehensive nuclear-test-ban treaty, [and] the launch of discussions on strategic stability with a view to establishing a new security framework in the future.”
To date, none of the five governments has publicly described the content or outcomes from the latest P5 process dialogue, which took place on the fringes of the 81st session of the UN General Assembly’s First Committee on Disarmament and International Security.
Ahead of the P5 process meetings, however, U.S. officials told the Wall Street Journal Sept. 24 that the Trump administration has been quietly holding talks with China about its rapidly expanding nuclear arsenal and U.S. allegations that Beijing conducted a 10-ton nuclear explosion in June 2020. That event, if confirmed as a nuclear explosion, would be a violation of the 1996 Comprehensive Test Ban Treaty (CTBT). (See ACT, March 2026.)
A separate CNN report published Oct. 6 said that the Trump administration had discussed beginning “reciprocal nuclear [test] site visits with China” and that China had proposed reciprocal visits of Chinese and U.S. nuclear weapons laboratories.
U.S. officials have supported a proposal announced by French officials May 1 to seek talks on confidence building measures, which may or may not require or involve reciprocal site visits. Senior Chinese officials who spoke with Arms Control Today Oct. 9 denied they had proposed reciprocal visits as CNN reported.
Echoing a proposal made in 2023 by the head of the U.S. National Nuclear Security Administration, France, the United Kingdom, and the United States proposed May 1 that the P5 process states engage in multilateral strategic stability talks on “confidence-building measures regarding nuclear explosive test monitoring, including increasing the ability to detect tests of any yield toward ensuring nuclear weapons states adhere to [the] zero-yield standard” established by the CTBT.
The prospects for such talks are murky, however, in the wake of what was described by several diplomatic observers as a “confrontational” presentation by U.S. Assistant Secretary of State for Arms Control and Nonproliferation Christopher Yeaw at a public side-event sponsored Oct. 8 by the U.S. delegation at the UN.
In announcing the event, the U.S. Mission to the UN claimed that, “The world has been hoodwinked into believing that an actual norm against nuclear testing exists, when in fact it’s an illusion created by an unverifiable treaty.” The CTBT has 188 state signatories, including the United States.
In his remarks and presentation slides obtained by Arms Control Today, Yeaw repeated the U.S. charge that China conducted a yield-producing nuclear test in June 2020. He also reaffirmed a contested 2019 U.S. assessment that Russia has conducted secret nuclear tests in the past that could help advance Russian nonstrategic nuclear capabilities.
Yeaw argued that the CTBT verification regime is not fit for purpose if the objective is to verify compliance with the strict zero-yield test standard. He acknowledged that the five nuclear-armed states all agreed during the CTBT negotiations that the treaty prohibits any “any test explosion that produces a self-sustaining nuclear chain reaction.”
Yeaw called the present situation regarding testing “unacceptable,” and he reiterated President Donald Trump’s threat that the United States would respond “on an equal basis” if alleged low-yield nuclear testing by other states continues. But Yeaw also said the United States prefers to address its concerns through diplomatic and multilateral channels.
In remarks at the UN First Committee Oct. 8-9, Chinese and Russian officials strenuously rejected allegations that they have violated the CTBT and called upon the United States to finally ratify the treaty to help bring it into force. That is necessary to facilitate short-notice, on-site inspections to detect and deter any clandestine, low-yield nuclear testing.
Meanwhile, Mexico and 176 states voted in favor of a resolution at the UN First Committee in mid-October that in part, “urges all States to refrain from any action that may hamper progress towards the entry into force of the Treaty and its universalization.” The resolution also “urges all states not to carry out nuclear-weapon test explosions or any other nuclear explosions, to maintain their moratoriums in this regard and to refrain from acts that would defeat the object and purpose of the Treaty.” The United States voted no.

In an Oct. 15 interview with Newsmax, Trump claimed he is holding talks with Russia and China “about that denuclearization.” The following day, the Chinese Foreign Ministry said: “It is neither fair, reasonable nor realistic to ask China to join the so-called trilateral nuclear disarmament negotiations at this stage.”
On Oct. 19, in response to reports that Trump is open to a meeting proposed by Russian President Vladimir Putin and Chinese President Xi Jinping to discuss Russia’s war on Ukraine at the upcoming Asia-Pacific Economic Cooperation summit, the U.S. leader told reporters the talks were “not so much on Ukraine because Russia, frankly, has been fine on Ukraine, they haven’t been doing much, and the same with China.”
“We would have it to talk about weapons and denuclearization. If President Xi wants to have it, I’d have it,” Trump added.
After the Trump administration rejected Russia’s proposal that the United States and Russia continue to respect the central limits of the now expired New Strategic Arms Reduction Treaty, it said it would seek a “new, improved, and modernized Treaty that can last long into the future.” (See ACT, March 2026.) The administration has not explained what Trump would seek to achieve through such a dialogue, nor publicly outlined any arms control proposals.
However, in remarks Oct. 20 at the Reagan Presidential Library in California, Yeaw claimed that “Russia and China have expressed initial signs of openness to engage in strategic stability and arms control discussions, which we hope will lead to more intensive engagements and eventually a better arms control agreement than what has come before.”
“We’ve made clear that the United States is open to discussion with other nuclear states on strategic stability in arms control, and that there are no preconceived notions, no preconditions. In short, our door is open,” Yeaw said.
In Oct. 1 remarks at the annual meeting of the Valdai Discussion Club in Russia, Putin said the Kremlin is ready to resume a nuclear arms control dialogue with the White House.
Iran called the U.S. response to Tehran’s proposal to reopen the Strait of Hormuz and begin nuclear talks “inadequate,” while Trump said the U.S. will not attack Iran again before the Nov. 4 elections.
November 2026
By Kelsey Davenport and Shaghayegh Chris Rostampour
Iran said the U.S. response to its proposal to reopen the Strait of Hormuz and begin negotiations on a nuclear deal was insufficient, increasing the risk that conflict intensifies as both sides threaten to escalate.

Mohsen Rezaei, Supreme Leader Ayatollah Mojtaba Khamenei’s representative to Iran’s Supreme National Security Council, confirmed that the United States responded to the proposal Iran put forward at the UN in September. (See ACT, October 2026.)
Although the United States “has agreed to some of Iran’s seven conditions,” the response was “INADEQUATE,” he wrote in an Oct. 11 post on the social media site X. “Iranian diplomacy has changed and there will be no return to the status quo ante,” he added.
Before Rezaei’s comment, U.S. officials, including Vice President JD Vance, had raised concerns about Iran’s proposal, noting that the Trump administration seeks concrete concessions on Iran’s nuclear program.
Nevertheless, U.S. President Donald Trump posted on Truth Social Oct. 8 that his administration was “having productive discussions” with Iran and the United States “will not be attacking Iran any time prior to the [U.S.] Midterm Elections” on Nov. 3.
CNN reported Oct. 15 that U.S. military plans to resume strikes on Iran include options to use the Massive Ordnance Air Blast (MOAB) bomb on nuclear sites and energy infrastructure.
When Newsmax asked Trump about the CNN report in an interview later that day, he did not confirm or deny if he would use the MOAB, which, if detonated near populated areas, could lead to devastating civilian casualties. Trump noted that the MOAB is “not nuclear at all. This is just an extremely large bomb.” He said the United States does not need to use nuclear weapons to defeat Iran. When asked whether he would send ground troops into Iran, Trump said, “No, I don’t see that.”
It is not clear how using the MOAB would advance U.S. strategic objectives. As an air blast bomb, the MOAB is unlikely to significantly damage Pickaxe Mountain, a deeply buried Iranian nuclear facility that Trump has threatened to strike. A MOAB would destroy energy infrastructure, and any use near populated areas would have a devastating effect on civilians. The United States has used the MOAB only once in conflict, on an Islamic State tunnel complex in Afghanistan in 2017.
Comments from Iranian officials suggest Tehran would view any use of the MOAB as a clear escalation.
The deputy coordinator of the Islamic Revolutionary Guards Corps, Brig. Gen. Mohammad Reza Naqdi, warned that if the United States makes the wrong move, Iran “will come for you with the father of bombs, and all your assets will be at risk.” The MOAB is often referenced as a backronym for “mother of all bombs.” Naqji, speaking Oct. 18 at a rally in the northern city of Sari, said that the United States could not achieve its objectives using heavy bombs, according to the Tehran Times.
Rezaei made a similar comment in an Oct. 18 post on X, announcing that “Iran’s armed forces have prepared a more devastating plan” compared to the 12-day war in June 2025, and the first six weeks of the ongoing war that began Feb. 28. “Iran’s new operational plan will not proceed with rules that they know,” he wrote.
The threats by both sides to escalate come as each continues to profess interest in reaching an agreement to end the conflict and address the nuclear issue.
Iranian President Masoud Pezeshkian emphasized that Tehran is still interested in negotiations. He also reiterated in an Oct. 16 meeting with the Georgian Prime Minister Irakli Kobakhidze that “Iran is not seeking nuclear weapons” and called the allegations “malicious propaganda against our country.”
Despite Rezaei’s statement indicating a shift in Iran’s approach to negotiations, Iranian Foreign Minister Abbas Araghchi said Oct. 18 that Tehran is reviewing U.S. comments on its proposal, which is “still on the table.”
Speaking to reporters in Baghdad at a meeting of the Organization of Islamic Cooperation, Araghchi said, “if there is any honesty on the U.S. part, which there isn’t, the way to reopening the Strait of Hormuz is obvious.” A month after the illegal U.S. and Israeli strikes on Iran began, Tehran closed the Strait of Hormuz, a strategic shipping route between Iran and Oman, in retaliation March 27. Although some ships are now transiting the Strait, volume is down, and Iran continues to strike ships in the waterway.
Neither Araghchi nor Rezaei provided any details on the U.S. response, but Vance suggested that the United States is looking for Iran to take more concrete action on its nuclear program. In an Oct. 5 interview with Reuters, Vance said that Iran needs to make a “meaningful” reduction in its uranium enrichment to meet U.S. demands. He said that if Iran wants “to show a commitment to not building a nuclear weapon, they would do something meaningful on their enrichment capacity.”
Vance was implicitly referring to the limitations in a future nuclear agreement—because Iran does not appear to be currently enriching uranium, International Atomic Energy Agency Director-General Rafael Mariano Grossi told The Association Press Oct. 28. Enrichment activity ceased at Natanz and Fordow in 2025, when the United States and Israel struck the facilities.
Vance’s comments are a shift from conditions Trump articulated in the past, which included a multiyear freeze on uranium enrichment or a commitment to give up enrichment entirely. It is unclear if Vance’s comments reflect the U.S. position as conveyed to Iran.
In response to Vance’s comments, the head of the Atomic Energy Organization of Iran, Mohammad Eslami, reiterated Iran’s long-standing position on enrichment, saying Oct. 7 that “Enrichment is an integral part of the nuclear process, and we will not abandon it in any way.”
Vance also told Reuters “it’s not totally clear how their country makes decisions.” Iranian Foreign Ministry spokesman Esmail Baghaei disputed Vance’s remark in an Oct. 7 press briefing, countering that the problem “lies in the contradictory positions and mixed messaging from U.S. officials.”
Despite Pezeshkian’s and Araghchi’s assertions that Iran seeks a deal, factions in Tehran continue to debate whether the country should remain party to the nuclear Nonproliferation Treaty. (See ACT, October 2026.)
Mohammad Javad Haj Ali Akbari, an imam who leads Friday prayer in Tehran and is the chief of Iran’s Friday Prayer Policy Council, said Oct. 8 that the “people’s expectation is for us to revise our nuclear doctrine. [The people ask] why should we still remain party to the NPT?”
The move highlights tensions between a growing need for nuclear power-reactor fuel and concerns about proliferation risks.
November 2026
By Jocelyn Bridges
The United States and China are seeking to increase uranium enrichment capacity amid plans to expand nuclear energy generation, a move that highlights tensions between a growing need for nuclear reactor fuel and concerns about proliferation risks.

For decades, nuclear power plant accidents—such as in Pennsylvania in 1979 and in Ukraine in 1986—stoked fears about the technology, but those have abated and the expansion of data centers and climate change concerns has created demands for new energy sources.
In September 2026, California-based General Matter, Inc. submitted its first license application to the U.S. Nuclear Regulatory Commission (NRC) to build and operate a proposed $1.5 billion commercial uranium enrichment facility in Paducah, Kentucky.
The proposal is part of a broader, federally backed effort to rebuild domestic enrichment capacity. In January, the U.S. Department of Energy announced that it awarded $2.7 billion in contracts to three companies to expand or create enrichment capacity over the next decade. General Matter and American Centrifuge Operating, LLC each received $900 million to develop domestic production of high-assay, low-enriched uranium (HALEU), and Orano Federal Services, LLC received $900 million to expand its low-enriched uranium (LEU) production capacity.
HALEU is uranium that has been enriched to between 5 and 20 percent uranium-235, while power-reactor grade LEU is uranium enriched between 3 and 5 percent U-235.
The United States continues to rely heavily on foreign sources for nuclear fuel, importing “nearly all the uranium concentrate” for its nuclear generators use, according to a 2025 U.S. Energy Information Administration report. A 2024 U.S. ban on Russian uranium imports has increased pressure to secure alternative supplies. Energy Department waivers continue to allow uranium imports from Russia, but they are only valid until January 2028.
U.S. President Donald Trump’s “American nuclear renaissance” program, spearheaded by several executive orders since May 2025, has encouraged the demand for and investment in domestically enriched uranium fuel.
During an NRC meeting in May 2026 on nuclear fuel facilities and storage, Chairman Ho Nieh emphasized the importance of expanding domestic production, saying that “the United States depends too heavily on foreign uranium and enrichment, and that has to change.”
General Matter plans to submit an amended NRC application in 2027 to produce HALEU. The company has begun site preparations for HALEU production despite having not yet obtained authorization, the company announced Sept. 28.
Interest in HALEU is driven partly by the development of advanced reactors that require fuel enriched beyond the LEU levels used in most conventional commercial reactors. Fuel with higher enrichment levels can enable smaller reactor cores and longer operating cycles, depending on the reactor design. The scale of future demand remains uncertain, however, as many of these reactors are in the planning and application phase.
The United States is expanding its uranium enrichment capacity for civil purposes, but there are also implications for nonproliferation. According to a 2024 Energy Department blog post, HALEU raises nuclear security and nonproliferation considerations beyond those associated with reactor-grade LEU.
Uranium enriched past 5 percent requires less additional enrichment to reach weapons-grade level. Some experts argue that uranium enriched to just 12 percent can be used in a nuclear explosive weapon, although doing so would present substantially greater technical challenges and require considerably more material than a weapon using weapons-grade uranium. The necessary regulatory frameworks governing the commercial production, transportation, and use of HALEU in the United States are still being developed, according to a June 2026 report by Carnegie Endowment for International Peace.
International trade in HALEU would require arrangements for fuel fabrication, transportation, and regulatory authorization, with applicable export controls and nonproliferation safeguards governing transfers to foreign recipients.
Meanwhile, recent satellite imagery analysis indicates that China is completing three large centrifuge enrichment plants with implications for its growing civilian nuclear power sector and its nuclear weapons program. The three plants—Emeishan Project 4, Lanzhou Project 6, and Hanzhong Project 6—could begin initial operation by the end of 2026, Hui Zhang of the Harvard Kennedy School’s Belfer Center for Science and International Affairs observed in satellite analyses for the International Panel on Fissile Materials. Construction of the facilities began in late 2024.
China has offered to place the Lanzhou and Hanzhong plants under International Atomic Energy Agency (IAEA) safeguards, but not the Emeishan City plant, according to Zhang. As a nuclear-weapon state, China is not legally obligated to place its nuclear facilities under safeguards. However, under a voluntary-offer agreement signed in 1988, China agreed to make some civilian facilities eligible for safeguards as a part of its stated commitment to promoting the peaceful use of nuclear energy. As of April 21, 2026, China has submitted 30 of its nuclear facilities for IAEA safeguards, according to a Chinese Foreign Ministry statement.
Zhang’s analysis estimates that the Lanzhou and Hanzhong plants could produce enough enriched uranium to meet projected civilian requirements through 2035, with a notable oversupply. The Emeishan plant could generate a substantially larger surplus.
China has substantial civilian demand for nuclear fuel. According to a report by the U.S. Energy Information Administration in May 2026, China has 60 operational reactors and 36 more are under construction. The IAEA reported in January 2025 that Chinese-designed reactors also account for a significant portion of global reactor construction. This expansion is expected to increase the country’s need for enriched uranium, although the scale of that demand will depend on the pace of reactor construction and the fuel requirements of different designs.
Zhang identifies the potential surplus from Emeishan as a possible source of highly enriched uranium (HEU), which is uranium enriched above 20 percent, a level usable in nuclear weapons. The Federation of American Scientists estimated that China’s stockpile has grown from around 290 warheads in 2019 to about 620 in 2026. In a December 2025 report to Congress, the Defense Department estimated China’s stockpile “remained in the low 600s through 2024” and projected it would grow to more than 1,000 warheads by 2030.
Although some assessments of China’s nuclear expansion have focused largely on plutonium production, Zhang argues that the uranium surplus from the Emeishan plant is the most likely source of any new HEU for weapons. He estimates that producing enough HEU to achieve 1,000 warheads by 2030 would require about 6 metric tons of additional HEU.
Russia condemned what it said were NATO plans to isolate or seize Kaliningrad, a strategic Russian enclave between Poland and Lithuania.
November 2026
By Libby Flatoff
Russia condemned what it said were NATO plans to isolate or seize Kaliningrad, a strategic Russian exclave between Poland and Lithuania, and threatened to use nuclear weapons if that happened. The statements were issued in separate but similar social media posts Sept. 29 and Sept. 30 by the Russian Embassy in the United Kingdom and the Russian Embassy in Ireland, respectively.

“This is not fiction, but matter of grave concern to Russia. The irresponsible and shortsighted European political elite is playing a dangerous game,” the Russian Embassy in Ireland wrote. “There should be no mistake—Russia would be ready to use all its arsenal, including nuclear weapons, to defend its territory if NATO countries try to isolate Kaliningrad region from the rest of the country,” the post said. The embassy complained that from June to September, “there have been almost incessant NATO military exercises with the anti-Russian scenarios in Poland, Latvia, Lithuania, Norway, Finland, as well as in the Baltic Sea.” It further asserted that “the joint Lithuanian, Polish, French military exercise ‘Gallant Boar 2026’ stands out as specifically aimed at the Russian enclave [sic] of Kaliningrad at the coast of the Baltic Sea.”
Kaliningrad is a significant Russian territory that holds important military assets and is Russia’s only ice-free port on the Baltic Sea. The embassy in Dublin’s statement condemned all NATO drills that are, “preparing air and naval blockade of Kaliningrad and Kaliningrad region.”
Russia’s embassy in London also made the nuclear threat stating, “Any British or NATO military action against Russian territory would be met with all means at Russia’s disposal, including nuclear weapons.”
Reuters reported Sept. 30 that nuclear threats were also sent in a formal diplomatic note from Russia to NATO. NATO Secretary-General Mark Rutte responded with a statement, reported by Reuters, insisting that, “We are a defensive alliance. And stop the nuclear threats. This is absolutely not called for and not helpful.” He added that no NATO “exercises or activities pose a threat to the territory of the Russian Federation.”
Speaking at the Warsaw Security Forum Oct. 12, Gen. Alexus Grynkewich, NATO’s Supreme Allied Commander, was asked how the alliance could defend against and deter Russia. The American general replied: “we have no need and incentive to escalate. This means the burden of escalation would rest squarely upon the aggressor.… And here, we must ensure the adversary also knows that we have the advantage—that escalation will only worsen his position, both militarily and politically.”
Meanwhile, Lithuania successfully passed one of three votes required to repeal the country’s constitutional ban on hosting nuclear weapons within its territory Oct. 6. The final vote and decision are expected in January 2027.
If the other two votes are successful, Lithuania will become the second state in the region to take this step. Finland took similar action earlier this year in an effort to enhance European nuclear and military planning. (See ACT, April 2026.) Lithuania and Finland have denied any immediate plans to host nuclear weapons on their territories, however.
Lithuania has also been criticized by Russia for its willingness to finance a permanent U.S. military base within its borders. Lithuanian Prime Minister Mindaugas Sinkevičius posted on Facebook Oct. 6 that his country’s “partners simply need to make a political decision regarding the permanent deployment of troops, and Lithuania is ready to cover the associated financial costs.”
That same day, Reuters reported Kremlin spokesman Dmitry Peskov saying, “If the military infrastructure of what is, in effect, NATO’s leading country is moving closer to our borders, then, of course, this cannot be perceived as anything other than a step towards escalating tensions.”
Next year will mark the fifth anniversary of Russia’s invasion of Ukraine, a main source of tension between NATO and Moscow.
The organizational shift would centralize and prioritize military autonomy.
November 2026
By Xiaodon Liang
The Pentagon plans to establish a new functional combatant command for the acquisition and integration of autonomous weapons across the U.S. military, U.S. Secretary of Defense Pete Hegseth announced Sept. 30, marking an organizational shift centralizing and prioritizing military autonomy.

The new unified command will be called Autonomous Warfare Command and led by a four-star general officer, Hegseth said in a speech at Marine Corps Base Quantico, Virginia. The command will “institutionalize,” and “supercharge” Pentagon adoption of artificial intelligence, autonomous weapons, and robotics, Hegseth said, describing the shift as “the biggest battlefield revolution in generations.”
The new organization must first be authorized by Congress, but the Senate Armed Services Committee, in its June 15 draft of the fiscal 2027 National Defense Authorization Act, has already recommended that Congress authorize the Pentagon to set up such a command, although under a different name.
Autonomous Warfare Command will be assigned “directed manpower, budget, acquisition authorities,” Hegseth wrote in a Sept. 30 memorandum. The Senate committee language authorizing the new command would grant authority for “force generation, joint training, interoperability, doctrine development, and operational employment through other combatant commands.”
The Senate has yet to take up the annual authorization bill on the chamber floor. The House of Representatives passed its bill, which does not include language creating a new command, July 22.
U.S. Space Command, which was created in 2019, was the last combatant command created by Congress. The Pentagon currently has seven geographical commands, including Space Command, and four functional commands for transportation, cyber operations, special operations, and strategic deterrence.
Responsibility for working with Congress and within the Pentagon to set up the new command will fall on Project Agincourt, a new effort announced by Hegseth in the Sept. 30 memorandum. Agincourt will be led by Owen West, the former director of the Defense Innovation Unit and assistant secretary of defense for special operations during the first Trump administration.
West will simultaneously also lead the office of the direct reporting portfolio manager for unmanned systems, a new entity that the Pentagon created in June, implementing acquisition reforms passed by Congress in last year’s defense policy act. The portfolio manager concept is designed to grant a single office the authority to oversee multiple related acquisition projects and make trade-offs across them. (See ACT, December 2025.)
Besides setting up the Autonomous Warfare Command, Project Agincourt will also implement “a new construct called war-fighting acquisition,” Hegseth said in his speech. The approach “fuses operators with entrepreneurs in rapid adaptation cycles, while distributing decisions and dollars closer to front-line units,” he said.
West provided an insight into what that approach might look like in practice during remarks Oct. 14 at the Association of the United States Army (AUSA) annual meeting, at which he pointed to “what’s happening in Ukraine, where a small portion of … the average [brigade’s] budget, 5 to 15 percent, is … gamified in the form of points where people can purchase directly” from a drone marketplace.
West and acting Army Secretary Adam Telle, who also appeared on the panel, noted that the U.S. Army has created a similar internal marketplace for drones. Telle announced that the army component of the new unified command would be called Futures and Autonomous Systems Command. “We are going to keep it distributed to our units. We are pushing autonomy down into our units,” the acting secretary said, while indicating that the new army command would primarily focus on coordination.
Speaking on the same Oct. 14 panel, Hegseth voiced optimism that the Pentagon could convince Congress to pass a $350 billion additional defense appropriation as part of a partisan reconciliation bill, regardless of whether Republicans lose control of one or both chambers of Congress in November’s midterm elections. “We’re running fast immediately with the committees to ensure that that reconciliation bill at a minimum of [$350 billion], potentially more, ultimately fills up what this department needs,” he said.
That additional funding request includes $53.6 billion for “multi-year investment in autonomous systems procurement, domestic production capacity, and advanced capabilities,” according to budget request documents. Legislators in both parties have expressed skepticism that Congress will take up an additional reconciliation bill this year.
The Defense Autonomous Warfare Group, successor to the Biden-era Project Replicator, was to administer that investment in autonomous systems. Derek Anderson, the assistant secretary for special operations and low intensity-conflict, said on the Oct. 14 AUSA panel that the group, which focuses on low-cost attritable drones, had now been folded into a task force under Special Operations Command.
The new tensions relate to the fact that two South Korean soldiers were injured by a landmine in the demilitarized zone between the two countries.
November 2026
By Kelsey Davenport
North Korea and South Korea tested missile systems as tensions persist over Seoul’s allegations that Pyongyang placed landmines on the south side of the demarcation line separating the two countries.

North Korea and South Korea announced that they tested hypersonic missile systems in October. North Korea’s test of a hypersonic ballistic missile is similar to systems that it has tested in the past and does not appear to represent new capabilities, but South Korea’s launch marked the first test of its indigenously developed hypersonic glide vehicle.
Whereas a hypersonic missile performs more like traditional ballistic missiles but is more maneuverable, a hypersonic glide vehicle is boosted by a missile into the atmosphere before separating from the system and reentering the atmosphere on a maneuverable trajectory at speeds in excess of Mach 5, or five times the speed of sound. The high speeds of both weapons make them more challenging to track and intercept using air defenses.
The state-run Korean Central News Agency (KCNA) said in an Oct. 4 statement that North Korea launched an intermediate-range strategic missile the previous day. The statement described the missile as hypersonic and said it traveled about 1,000 kilometers before splashing down in the Sea of Japan.
The launch was part of an effort to “train the missile soldiers operating the hypersonic strategic weapon system,” the statement said. North Korean leader Kim Jong Un observed the launch and described the drill as a demonstration of the soldiers’ ability to operate a “major offensive system.” North Korea must continue to “make the enemy recognize the reliability and fatality of our strategic offensive means,” Kim said.
The South Korean Defense Ministry issued a statement Oct. 4 saying it monitored the test’s “preparation stage” and detected and tracked the missiles immediately after launch, and precisely identified the launch point.”
Japan also issued a statement confirming that it tracked the missile but had a slightly different assessment of the flight range.
Although the missile type North Korea tested is more challenging to track and intercept because of the maneuverability of the reentry vehicle, the Defense Ministry statement said that South Korea and the United States assess “that the missile North Korea recently launched is fully detectable and interceptable by the defense assets held by both countries.”
Five days after the launch, South Korea tested a domestically developed missile tipped with a hypersonic glide vehicle. South Korean President Lee Jae Myung observed the Oct. 8 test and later described it in a statement as “successful.” Hypersonic glide vehicles are an “advanced weapon system that will change the landscape of future battlefields,” he said.
Lee emphasized that South Korea designed the missile indigenously and that its success demonstrates the country’s “determination to defend our nation with our own technology and our own strength.” The launch was designed to test the heat resistance of the hypersonic glide vehicle, its guidance systems, and ability to maneuver, according to a separate statement from Lee’s office.
North Korea accused South Korea of lying about the hypersonic test. Jang Chan Ha, general director of North Korea’s Missile Administration, said in an Oct. 14 interview with KCNA that it was “hard to believe” that the glide vehicle “underwent hypersonic flight.” He said photos released by South Korea depicting the test “arouse suspicions” about the validity of the claim. Jang said it is “hard to believe that [the glide vehicle] underwent hypersonic flight circumstances characterized by extremely high temperatures as its outside appearance is so clean with its coating paint remaining intact, not stained with any soot.”
The South Korean Ministry of National Defense responded to the allegation, saying in an Oct. 15 statement that the “heat-resistant materials” used for the hypersonic glide vehicle protected against soot.
In addition to the hypersonic intermediate-range ballistic missile launch, North Korea tested in October what state media described as submarine-launched cruise missiles (SLCMs) and a short-range ballistic missile designed to carry a tactical nuclear warhead.
On Oct. 12, North Korea tested a “tactical ballistic missile,” according to a Missile Administration statement in KCNA issued Oct. 13. The statement described the test as part of the Missile Administration’s “regular activities” to upgrade its missiles and improve the “performance and sophistication of all offense weapons systems.”
In an Oct. 15 statement in KCNA, North Korea’s Missile Administration said it fired two types of SLCMs. The Oct. 14 launches were intended to “examine the current mobilized posture of the underwater combat systems” and help North Korea’s navy perform “its strategic missions,” the statement said.
Photos released by KCNA suggest the SLCMs were launched from a Soviet diesel-powered submarine that North Korea describes as a tactical nuclear attack submarine. The tests appear to be part of Kim Jong Un’s focus on expanding North Korea’s naval nuclear deterrence.
The uptick in North Korean missile launches comes amid tensions between Seoul and Pyongyang after two South Korean soldiers were injured by a land mine in the demilitarized zone between the two countries. Another unexploded mine was found Sept. 29.
U.N. Command and South Korea’s military conducted a joint investigation of the Sept. 21 incident and determined that the landmines were deliberately buried on the south side of the military demarcation line established by the Korean War Armistice Agreement. The mines were buried within the last 12 months and the design components are consistent with North Korean landmines, the assessment concluded.
South Korea’s vice chairman of the Joint Chiefs of Staff, Lt. Gen. Kwon Dae-won, said in a Sept. 28 press briefing that North Korea must apologize for the incident and remove the mines south of the demarcation line. Otherwise, the South Korean military will “continue mine-clearing operations there to ensure peace in the [demilitarized zone] and the safety of our troops.”
In an Oct. 2 statement in KCNA, Kim Yo Jong, sister of Kim Jong Un and department director of the Central Committee of the Workers’ Party of Korea, accused South Korea of making a “groundless conclusion” that North Korea deliberately laid the mines. South Korea’s accusations “clearly revealed its sinister intention to plunge the situation in the Korean peninsula again into a critical state for its selfish purpose,” she said. Kim Yo Jong reiterated that North Korea is not interested in talks with South Korea and said Seoul “should no longer play cheap tricks” designed to “tarnish the image of our state.”
The mass changes to 46 directives and elimination of four are part of a drive to accelerate programs across the department, including within the nuclear weapons enterprise.
November 2026
By Xiaodon Liang
The U.S. Department of Energy amended dozens of regulations as part of its drive to accelerate programs across the department, including within the nuclear weapons enterprise, while repealing several directives on environment, workplace safety, and health considerations.

“By removing unnecessary barriers, we are giving our scientists, engineers, and technicians, more freedom to focus on the critical missions that matter most,” Energy Secretary Chris Wright said in an Aug. 13 press release announcing the changes.
The mass changes to 46 directives and the elimination of four, which took effect Aug. 5, triggered pushback from Sen. Patty Murray (D-Ore.) and Rep. Marcy Kaptur (D-Ohio), the ranking members of the two congressional appropriations subcommittees responsible for funding the Energy Department.
“The department cannot execute such sweeping systemic changes across multibillion-dollar facilities without providing the empirical baselines, safety analyses, and governance frameworks that justify them,” they wrote in a Sept. 16 letter to Wright.
The letter requested the department to provide a detailed, 20-point breakdown of the regulatory overhaul, including an explanation for which requirements had been merged or eliminated, within 30 days.
A staffer in the office of Rep. Kaptur told Arms Control Today Oct. 30 that the department had not yet provided a response.
The regulatory changes come shortly after the Defense Nuclear Facilities Safety Board, an independent organization created by Congress in 1988 to oversee health and safety at defense nuclear weapons facilities, reported deliberate moves by Energy Department officials to impede the board’s mission. (See ACT, December 2025.)
The dispute over the trade-off between speed and safety at defense nuclear facilities has become increasingly important as the National Nuclear Security Administration (NNSA) proceeds with major construction and warhead production efforts, including a plan to reconstitute a plutonium pit production capability at Los Alamos, New Mexico, and Savannah River, South Carolina. (See ACT, April 2026.)
Board member Patricia Lee wrote in July that since the beginning of the year, safety board staff have “experienced increasing access challenges, the most significant in its history, including verbal and written denials from DOE and a lack of responsiveness to access requested.” Lee issued the warning as part of a congressionally mandated twice-yearly report on board requests for access. As a result, the board “has been unable to conduct some safety reviews” and has “not been provided access to large numbers of draft DOE safety directives and regulations that are currently being revised,” Lee wrote.
Lee is currently the board’s only Senate-confirmed member. President Joe Biden nominated William White to serve on the board in May 2024, but the Senate did not vote on his appointment during that Congress. The Trump administration has made no additional nominations.
In March, department officials instructed staff to escalate to their superiors access requests by the board that are perceived to be “activities or requests outside those authorized by its statute” and to recognize, assert, and protect “the Executive privileges of the President and the Secretary” in interactions with the board.
The instructions were issued in a March 10 memorandum, co-authored by James McConnell, the NNSA’s associate principal deputy administrator for operations, and Greg Sosson, the associate principal deputy assistant secretary for field operations at the Energy Department’s Office of Environmental Management, that listed what the department considered to be inside and outside the safety board’s purview.
The March 10 instructions risks “impairing the [safety board’s] ability to carry out its independent safety oversight mission,” Lee responded in an April 29 letter, posted on the board’s website, that highlighted points of disagreement with the memorandum.
A bipartisan oversight committee of the New Mexico state legislature highlighted similar concerns in two Sept. 16 letters. Writing for the Radioactive and Hazardous Materials Committee, chair Jeff Steinborn accused the Energy Department of making regulatory changes without “sufficient transparency or critical review” and voiced “escalating concern regarding the collapse in communication and cooperation” between the department and the safety board.
The NNSA greenlit Oct. 15 construction of the main processing building of the future Savannah River Plutonium Processing Facility, where the agency intends to produce at least 50 plutonium pits per year. (See ACT, June 2026.)
The decision included approval of a performance baseline for the project, but the agency did not release a new estimate of the building’s cost. In its fiscal year 2027 budget request, the NNSA assigned a “placeholder” estimate of $23.5 billion for the main processing building, which should be complete by 2035.