Good afternoon, and thanks for having me as part of this event.  I hope you do not end up by concluding that it wasn’t a good idea to ask a mere lawyer for his thoughts on the ways in which emerging technology may affect the nuclear weapons arena.  Nevertheless, I am myself very much looking forward to hearing what you have to say on these topics – and, as always, I am happy to share my own thoughts, for whatever they may be worth.

Naturally, all I can offer are my own personal opinions here – and nothing I say should be taken to represent the views of anyone else.  That said, perhaps these comments can at least get some discussion going.

Verification and Monitoring 

Now, I know the focus here today is on emerging technology threats.  Yet I do think it’s worth noting at the outset how technological change has affected at least some aspects of the nuclear weapons world in recent years in perhaps more salutary ways.

The revolution in open-source overhead imagery from private sector collectors over the last decade or so, for instance – coupled with the emergence of quite powerful cyber traffic analysis and data analytics – has done impressive things for unclassified open source intelligence (OSINT), and has had real implications in the arms control arena.  When I first started working on arms control and nonproliferation verification and compliance assessment issues more than 20 years ago as Principal Deputy Assistant Secretary at what was then called the State Department’s Verification and Compliance (VC) Bureau, the idea that OSINT could provide the sorts of sophisticated imagery analysis, mensuration, and multispectral analysis that it does today – even including synthetic aperture radar (SAR) data – would have seemed almost fantastical.

It’s clearly also an extraordinary new world when, for instance, private-sector analysts drawing on unclassified data sources can do things such as trace the movements of the Russian FSB agents – in Russia, no less – who tracked and then poisoned the late Alexander Navalny with a novichok chemical weapon nerve agent in 2020.  Or when such analysts can use social media to help identify and locate a number of Russian soldiers then involved, despite Moscow’s initial denials, in invading Crimea in 2014.  Or when they can document Syrian use of chemical weapons, or identify perpetrators of the shoot-down of Malaysia Airlines flight MH17 by Russian-backed forces in Ukraine.  I’m not necessarily arguing that such excellent open-source work will provide a panacea for confusion and misinformation in the mass information space, but it is now a powerful and persuasive new tool at least among responsible expert audiences.

To be sure, it would seem that the promised revolution in which civil society social media activism would transform arms control verification by providing opportunities for citizens to help catch violators – which I heard officials talk about rather breezily early in the Obama Administration – has not quite panned out as advertised.  One recent study, for instance, noted the degree to which such assumptions tend to assume an “idealized societal-verification community” in which citizens play the role of “holding their governments to account.  This would seem a model most useful, however, in holding to account governments that are already accountable to their citizens – and who permit those citizens the opportunity to inform against their rulers.  In the absence of such an idealized community, “societal verification practices may promote mistrust or distrust among publics and governments, with destabilizing effects,” and it is very hard to imagine such approached having much utility vis-à-vis a closed society such as Russia or China uninterested in arms control in the first place – which is, of course, precisely where they would be most needed.  So perhaps all that was oversold, along with so much else about the other wonderful effects that social media was supposed to have.

Nevertheless, it’s clearly the case that thanks to certain technological advances, more can be known about more sensitive things than one would have expected years ago.  OSINT capabilities may not be a panacea, but they can often at least contribute to arms control and nonproliferation efforts in the future, complementing negotiated verification protocols (to the degree that we ever again have such things), traditional spying, and observation by what were once known as “national technical means” (i.e., satellite reconnaissance).

In terms of what may lie ahead, I imagine that it won’t be long before we see such things as results from Artificial Intelligence (AI) algorithms trained on reams of international scientific publications to spot interesting but subtle anomalies or trends in who is working where, with whom, and upon what dual-use technologies – or perhaps who has stopped publishing for reasons that cannot otherwise be explained.   Such pattern analysis could be useful to arms control and nonproliferation analysists in helping identify potential signs of secret nuclear, biological, or chemical weapons work, or spotting advances in advanced conventional weaponry.

When I was at State/VC, we sometimes contracted with humans to do such analysis, and I think at least some such work is done by International Atomic Energy Agency (IAEA) safeguards investigators today.  But I’d wager that AI could soon become very helpful in such work, if it isn’t being used already.  With advances in data analysis being coupled with various sorts of improved data collection – such as might be provided by the development of quantum sensors – the verification arena may well continue to evolve in significant ways.

Those are probably all good trends, especially for so long as we remain in geopolitical struggles against the predatory imperialist powers of Russia and China and continue to try to stymie the efforts of various lesser bad actors around the world to acquire weapons of mass destruction.   Such technologies can enrich both classified and unclassified analysis in ways that I certainly hope will help us continue to keep such threats at bay.

Nuclear Weapons Development and Production 

So that may represent some of the nice effects.  Somewhat more ambivalently, however, I would also imagine that emerging technologies will play important roles in changing how nuclear weapons systems are developed and produced.  I understand that Additive Manufacturing (AM) technology, for instance, is already being incorporated into some aspects of the U.S. nuclear weapons infrastructure, at least with respect to the production of certain non-nuclear components – such as at the Kansas City Plant, which manufactures around 80 percent of the non-nuclear components that go into the U.S. nuclear stockpile, and which uses AM for metals, polymers, pads and cushions, and even electronics. 

With facilities such as the Oak Ridge National Laboratory and the Idaho National Laboratory already working to use AM in the production of nuclear fuels, it stands to reason that such techniques might perhaps be useful in improving nuclear weapons manufacturing as well.  It may take time for such techniques to migrate into the core areas of a nuclear “physics package” – after all, some of you here may remember how tricky it was for U.S. experts to certify the switch from making wrought plutonium pits in the W-88 warhead (as manufactured originally at the Rocky Flats Plant) to casting pits as currently done at the Los Alamos National Laboratory – but such movement is hardly unimaginable.

I would also think that the world of nuclear weapons and delivery system design will follow longstanding broader defense and civil sector trends in the use of ever-more sophisticated “digital twin” software to allow systems to be increasingly effectively designed, tested, modified, and validated in silico, as it were.  This will likely speed up the design and developmental phases of any given program, pushing back the point at which systems have to be physically constructed, and will at some point even allow whole ecosystems of weapon systems to be tested and (digitally) “employed” together in more and more complex operational scenarios inside computers.  (Imagine, if you will, and entire functional battlespace modeled at high technical and operational fidelity, all inside a computational cloud.)

Such trends will presumably continue, and with further advances in computing power – including, perhaps, the eventual advent of quantum computing applications in this realm – I would imagine significant improvements might be possible in the weapons development and production process.   As the United States works on what the Biden Administration now terms a “Production-based Resilience Program” (PRP) – an effort specifically designed to “establish the capabilities and infrastructure that can efficiently produce weapons required in the near-term and beyond, and that are sufficiently resilient to adapt to additional or new requirements” – some combination of such capabilities may end up being very helpful. 

Weapons system development times might, for instance, be shortened, and the production infrastructure perhaps more decentralized, at least for some components, through more use of widely distributed manufacturing nodes in order to avoid “single-point-of-failure” bottlenecks.  Who knows?  It might even be possible to train classified AI algorithms on the “legacy code” data data from the United States’ more than 1,000 nuclear weapons tests – or perhaps even also those of our allies in the UK (45 tests) or even (one can dream, no?) the French (210 tests) – in order to support expedited or novel design work in response to geopolitical threats in the years ahead.

In all of these ways, the processes and techniques for nuclear weapons development may become considerably more sophisticated in the years ahead.  This might not necessarily herald a boom in proliferation problems, for the most likely nuclear-related “uptake” for such advanced techniques – at least for some time – will surely be among the most sophisticated powers that already have nuclear weapons production infrastructures. 

For those powers, these developments may ultimately be more evolutionary than revolutionary in their impact, and likely much less dramatic than in the nuclear arena than the potential role – in the biological warfare context – of using AI algorithms to predict the morphology of folded amino acids or using gene editing to create bespoke disease vectors, toxin-producing microbes, or entirely novel types of weaponized biology.  But they may prove very significant in the nuclear realm for all that.

Cyberattack

More worrying, however, are other possibilities.  Cybersecurity issues, for instance, will probably emerge as increasingly important concerns in the nuclear weapons arena.  To be clear, I myself have no particular insight into such matters, and was thankfully never party to any such deep secrets if there are any.  Nevertheless, it’s worth remembering that this is hardly a new idea.  In his doctoral dissertation a few years ago, for instance, a friend and former government colleague of mine drew upon declassified U.S. Government records suggesting that the idea of using cyberattack to impede the potential wartime effectiveness of a rival power’s National Nuclear Command and Control (NC3) systems goes back at least to the early 1980s.(1)

Let’s not forget that this seems to be a big time, right now, for countries to be working on NC3 issues.  Most prominently, the United States is currently modernizing its own NC3 for the first time since the first Reagan Administration.  Moreover, China is now building out a vastly-expanded nuclear force likely to approximately equal the size of our own by 2035, and apparently intends to “increase the peacetime readiness of its nuclear force by moving to a launch-on-warning (LOW) posture.”  All of this, of course, surely entails an expanded NC3 system.  And both India and Pakistan have presumably been building and expanding NC3 systems of their own as their nuclear arsenals have expanded since their overt weaponization in 1998. 

Whatever a modern NC3 system actually ends up looking like in any given nation, therefore, it seems a safe bet that cybersecurity will be a critical, first-order concern for its architects. By the same token, failing to provide adequate cybersecurity in the NC3 context could be a recipe not just for potential wartime failure, but also for highly problematic peacetime destabilization – such as if a would-be aggressor concluded that its adversary’s nuclear command-and-control system could be degraded enough through cyberattack to undermine that arsenal’s deterrent effect.

Quantum Computing 

Another potential way in which emerging technologies could have a destabilizing strategic effect is in the potential impact of a future revolution in highly capable quantum computing, creating what some have termed a “quantum apocalypse.”  In such a scenario of cryptographic catastrophe, a state making a major breakthrough in large-scale quantum computing might (in principle) be able to read essentially all of every other government’s most sensitive encrypted message traffic – perhaps drawing in this respect upon an archive of hitherto-unreadable intercepted data warehoused precisely for this purpose, perhaps for many years, while awaiting the final quantum breakthrough. 

The potential impact of such a scenario upon strategic stability might be significant, as it could render the strategic posture, plans, and intentions of one adversary in strategic competition – or in conflict – with another essentially an open book to its adversary.  The U.S. National Institute of Standards (NIST) is presently working to develop new “post-quantum” cryptographic standards for secure computing, and first started releasing candidate algorithms to this end in 2022, but we are still nowhere near a mass migration to a demonstrably secure new standard.  As things presently stand, “first mover” advantage in effective quantum computing for cryptography would be a significant advantage indeed.

Autonomy 

Finally, it would be hard to have a comprehensive discussion of possible risks from emerging technologies in the nuclear arena without mentioning autonomous weapons.  The broader subject of Lethal Autonomous Weapon Systems (LAWS) is far beyond the scope of what we can discuss here, but despite the frequently high level of civil society panic about autonomous weaponry, my own suspicion is that fully autonomous (i.e., “human out of the loop”) nuclear weapons systems are not terribly likely.

If there is any area in which national leaders would refuse to countenance handing over their own most existentially critical trigger-pulling authority to a computer, after all, this is presumably it.  In that sense, it may be that AI-type applications are more likely to take over human functionality, if they do at all, in areas merely supporting nuclear weapons release authority – such as time-urgent indications-and-warning (I&W) analysis, battle management decision-support tools, or other data-management and modeling applications supporting nuclear deterrence, war fighting, and force posture planning – than in the weapons-release decision itself.

Nevertheless, I’ll admit that there is at least one partial historical counterexample: the Perimeter (a.k.a. “Dead Hand”) system developed by the Soviets and apparently still maintained by Russia.  This is an automated system which, once activated, is reportedly designed to launch all of Moscow’s remaining nuclear arsenal if sensors confirmed nuclear strikes on Soviet territory and the system’s robot brain no longer had communications with the General Staff.  Perimeter doesn’t fit the usual stereotype of a single autonomous weapon cruising around and making “kill” decisions for itself, but it is in some sense a fully automated system for nuclear weapons release.  So even here, I suppose we can’t be entirely sanguine.

Conclusion

I will be very interested to hear your perspectives on all these issues as this conference proceeds.  I wish I had more clarity to offer about how emerging technologies might affect nuclear deterrence and strategic stability, but I hope at least that our discussions can help us identify issues of potential relevance for further inquiry.

Thanks for listening.

– Christopher Ford

Notes:

(1)    Specifically, in 1982-83, the U.S. Joint Chiefs of Staff (JCS) created a Joint Special Studies Group (JSSG) – supported by a storied National Security Agency (NSA) component then known as “K Group” – that (among other things) looked at the ways in which jamming and signals injection into adversary networks could be weaponized against an adversary’s strategic command-and-control (C2) systems as part of an overall strategy against, inter alia, Soviet nuclear powered ballistic missile submarines (SSBNs).  See Craig J. Wiener, “Penetrate, Exploit, Disrupt, Destroy: The Rise of Computer Network Operations as a Major Military Innovation,” dissertation submitted in partial fulfilment of the requirements for the degree of Doctor of Philosophy, School of Policy, Government, and International Affairs, George Mason University (October 26, 2016), at 80, 85-86, & 97.  (As noted in the document, sourcing for Dr. Wiener’s dissertation was declassified for publication.)