The New Kalashnikov: Governing the Poor Man’s Autonomous Drone
Abstract
Cheap commercial drones, fitted with readily available guidance software, are giving militias and proxies a form of precision strike once reserved for States. Ukraine has already shown the method, which China’s hold on civilian airframes and components makes hard to shut down. UN talks in Geneva can set rules for high-end state systems, but they were not built to police dual-use parts that become weapons only after conversion. The practical response is therefore supply-side: tighter controls on the hardware that makes onboard autonomy work, obligations on manufacturers whose platforms are routinely militarized, and a small coalition of component-producing allies that can move faster than a treaty.
The Proliferation Problem
Popular belief notwithstanding, the most dangerous autonomous weapons are not necessarily the flashy ones one sees in the headlines. Small systems built today from commercial parts in unassuming sheds and deployed tomorrow against US forces anywhere around the globe might very well be far more dangerous. Meetings in Geneva continue to grapple with the ethical and strategic implications of high-end autonomous weapons; however, the CCW was not designed as a mechanism for policing the commercial supply chains that can turn a quadcopter into a weapon.
Militias and insurgent groups are already adapting commercially available, primarily Chinese-made quadcopters to carry explosives. This increasingly involves adapting open-source or commercially available computer vision and navigation software for more autonomous or semi-autonomous functions, yielding ever more effective attack capabilities.
The strategic significance is difficult to miss. Cheap, AI-enabled kamikaze drones are rapidly lowering the cost of precision strikes and giving non-state actors capabilities once reserved for conventional militaries. While threats like Chinese autonomous stealth bombers make convenient copy for a hitherto unproven story of looming Chinese military supremacy, they also take the headlines. The more immediate challenge for US security, and the one closer to hand, however, is the mass proliferation of systems assembled from commercially available components.
The good news is that we do not have to wait for great power strategic competition to enter the realm of cost-imposition. The bad news is that it is already here, and it is here to stay. China has near-absolute dominance in commercial drones and on dual-use components, which gives Beijing a great structural advantage, enabling partners and proxies to impose costs on US forces while China maintains strategic distance.
The New Kalashnikov
In many ways, what is happening is the 21st-century analog of the AK-47, which permanently altered 20th-century warfare by democratizing infantry combat. It did so because it is cheap, rugged, requires almost no training to operate, and, importantly, because of the Soviet Union’s willingness to turn it into a 7.62×39 geopolitical instrument. Commercial drones are now doing comparable work for precision strike. For policy, the difference is that the AK-47 arrived as a weapon while a drone becomes one only after conversion. Forward-deployed forces and the ability to maintain freedom of navigation are central to US alliances and deterrence in the Indo-Pacific and the Middle East. Until now, challenging this power projection has been an incredibly costly exercise. The spread of low-cost autonomous and semi-autonomous drones, however, is changing the equation by dramatically lowering the cost of contesting this presence.
Indeed, the uncomfortable reality that has emerged over the past several years is that the cost and technical barriers to fielding lethal capability have collapsed rather dramatically. Non-state actors do not need a world-beating defense industrial base and eye-watering budgets to field increasingly capable drone swarms.
Non-state actors require civilian drone frames, 3D-printed munitions, object recognition, tracking, and navigation software that are all widely and publicly available on platforms such as GitHub, often adapted or combined with commercial flight controllers. The future truly is a Google search away.
Policymakers, who can hardly be blamed for not possessing a deep understanding in all things artificial intelligence – or technology in general – are trapped in an outdated paradigm that treats AI as a monolith rather than a spectrum of capabilities.
Ukraine Already Wrote the Manual
We do not have to hypothesize the nature of the strategic threat. The Russo-Ukrainian war has already made it plain at a terrible cost. Moscow has brought human suffering at an extraordinary scale in 21st-century Europe, yet the words of one of Ukraine’s foremost writers, Lesya Ukrainka, written more than a century ago, remain strikingly apt: those “who have not lived amidst the storm do not know the value of strength”. Ukraine has shown strength in industrial form, as the country has spent the past years demonstrating what cheap commercial frames, 3D-printed warheads, and increasingly autonomous terminal guidance are capable of.
The method is mass-produced and cheap. Mass production, low costs, and a last-mile brain mean that a drone costing less than an entry-level iPhone can destroy a tank worth more than an oceanfront penthouse in Fort Lauderdale. Meanwhile, the operator no longer needs to be a particularly gifted flyer. Reports of drones selecting and striking with little or no real-time piloting are no longer speculative. With the lines blurring, the lessons are not limited to Ukraine, as state and non-state forces may now field remarkably similar stacks.
The recipes in the Big are comparatively simple: pair cheap commercial airframes with open-source or adapted autonomous navigation and terminal guidance software, then season to taste with fiber-optic links. Serve with gusto. Cooks can significantly reduce vulnerability to traditional electronic warfare jamming and can threaten forward-deployed US forces at a fraction of the traditional cost of building an air force.
Closing the Governance Gap
Digesting the threat is further complicated by global diplomatic -and administrative- frameworks being fundamentally mismatched to the pace and character of AI-enabled warfare. The Group of Governmental Experts (GGE) is the right forum for developing rules and principles governing state responsibilities, such as how a system is characterized, whether it can meet distinction and proportionality, and who remains responsible when it engages. Traditional arms control mechanisms are built for identifiable weapons categories, but they work far less well in a dual-use commercial ecosystem in which the weapon appears only at conversion.
The GGE, however, works by consensus, and consensus moves slowly. Russia has frequently used consensus procedures to slow binding outcomes while China has taken a relatively clear position in Geneva, but both continue to enable and encourage the spread of lower-tier drone technology and components to proxies and partners. Consequently, when it comes to governing agile, dual-use commercial technologies, existing multilateral institutions and mechanisms are struggling to keep pace.
What makes this rapidly emerging threat different is accessibility. High-end autonomous systems still require industrial depth and political backing. A drone assembled from commercial parts in a garage does not. Such parts can now be turned into a usable strike capability at a price that non-state actors can readily absorb. It is easy for policymakers to focus on the most visible and expensive systems, yet by doing so they will keep missing the cheaper ones that spread far more easily.
The United States cannot realistically ban the civilian staples of such systems, as many, like smartphone processors, commercial optical sensors, and agricultural drone motors, are ubiquitous in legitimate commercial markets. Unlike international agreements, which usually take years to negotiate and formalize, the hardware development cycle is much shorter.
Denying the Stack, Not the Algorithm
The United States should therefore pivot to a pragmatic supply denial strategy. Yet while the strategy is sound, putting it into practice is not. Dual-use controls face a deep structural problem. The same components that underpin legitimate commercial applications can, with comparatively little modification, support terminal guidance and autonomous navigation for attack drones.
Consequently, any effective supply denial regime must differentiate between legitimate end uses and transactions that present a meaningful risk of diversion without effectively shutting down civilian markets and inadvertently facilitating large-scale economic disruption. This is an intrinsically dynamic and fraught task, and one of the central problems of autonomous systems governance. Not only does this problem involve discerning the commercial technologies most critical to national security, but it also entails imposing effective constraints that are workable even as commercial technology races ahead of regulatory cycles.
Skeptics will contend that because open-source computer-vision models — from general-purpose detectors to specialized weights trained to recognize military drones — are already mirrored across thousands of public repositories, the software genie is out of the bottle. And the sceptics will be right. They will, however, be missing the point and the physical reality of fielding autonomous capability at scale. Even when the software is freely available, a host of practical limitations stand in the way of turning it into reliable, long-range attack systems. There are specialized low-power edge chips that can run inference without rapidly draining batteries or overheating. Energy-dense batteries that determine endurance are a necessity. Airframes and motors that set payload and range limits are a must. So are munitions integration that turns a commercial platform into a weapon. Finally, much operational learning is required to reliably deploy these systems under electronic-warfare conditions.
There is no way of denying that chips are a very important choke point, but they are not the only one. Rather than targeting a single component in isolation, dual-use export controls must target the combination of enabling hardware. By tightening controls across physical and software layers on both domestic and foreign suppliers, the United States can raise the costs and lower the reliability of proliferated autonomous systems, reducing strategic uncertainty before international consensus on governing autonomous weapons is reached.
Building a Minilateral Denial Coalition
It is paramount that Washington takes an interagency approach to the problem. The US should build a minilateral technology denial coalition with key component-producing allies, specifically Taiwan, Japan, South Korea, and relevant European States. By harmonizing export restrictions on dual-use microcontrollers, optical sensors, and drone motors across manufacturing hubs, the coalition can bottleneck the supply chain for non-state actors. Battlefield exploitation of recovered airframes should feed dual-use control lists at commercial speed so the items that actually appear on last-mile drones are the ones that get scrutinized.
Instead of treating export controls and multilateral norm setting as separate tracks, Washington should treat supply denial as one practical instrument within a broader autonomous systems governance approach. In the short term, the most pressing security concern is uncontrolled proliferation of the technologies that allow non-state actors to field autonomous systems. A focused multilateral export control arrangement, drawing on the Missile Technology Control Regime (MTCR) but aimed at the critical dual-use components of small autonomous drones, can raise the threshold for proliferation while diplomatic processes continue.
While missiles and drones are very different beasts, the MTCR provides a useful model because it targeted tangible delivery systems and related technologies rather than abstract concepts. It would be quite an overstatement to call the MTCR faultless, as its record is quite mixed. Nonetheless, a similarly focused regime targeting critical physical components of the autonomous drone supply chain could prove more agile than broad treaty efforts, and the United States has reinterpreted aspects of the MTCR for drone exports.
The first step is exploiting hardware and dual-use chokepoints through targeted export thresholds. One may attempt a rather futile blanket ban on basic microcontrollers, but there is a far better way, as controls should focus on performance and characteristics, and power thresholds. These enable real-time onboard autonomy in small airframes, low-power edge accelerators, high-end optical sensors, and related components that significantly improve autonomous targeting and navigation. Restricting the most capable of these items and applying tighter scrutiny to diversion pathways could push proliferators toward less capable, more jammable, and/or shorter-range configurations.
Making Manufacturers Own the Conversion Risk
The second most important step is to impose enforceable strict due diligence and security obligations on major manufacturers. Drones may seem to appear out of thin air, but they must be manufactured somewhere. There are a myriad of Chinese drone producers, but the market is dominated by a handful of giants, with the most notable of them being DJI, whose platforms are routinely modified for military use by proxies and non-state actors. In mid-August, the company won a partial victory in its long legal battle with the US government as an appeals court ordered a lower court to reconsider designating the firm as a “Chinese military company.” The case illustrates why designation needs to be paired with specific obligations to the manufacturer and the penalties for their non-fulfilment.
No manufacturer of any equipment is. However, when firms knowingly profit from platforms that are regularly diverted to armed conflict while refusing to adopt reasonable measures to prevent this from happening, governments should be prepared to impose market access restrictions and sanctions. DJI has largely shifted its US GEO system from automatic flight restrictions toward advisory warnings. In a similar vein, any additional limitations imposed by manufacturers, firmware modifications, or other restrictions that make militarization more difficult should be encouraged or outright required as a condition of access to the Western markets.
Enforcing this liability, however, requires using existing regulatory architecture rather than creating yet another framework. Where manufacturers’ platforms do not include cryptographically signed, secure-boot firmware that makes unauthorized flight-controller overrides more difficult, Washington should condition access to specified US software and technology on compliance with those security requirements, and use export control authorities including, where applicable, the Foreign Direct Product Rule (FDPR).
Furthermore, Treasury’s Office of Foreign Assets Control should consider targeted sanctions against foreign distribution and modification networks that facilitate the transformation of civilian airframes into weapons. If a manufacturer is unwilling to make its civilian technology resistant to cheap conversion to military use, it should be completely cut off from Western markets.
Lastly, Washington should require major platforms and repositories hosting dual-use target-recognition AI to adopt end-user screening and threat-reporting protocols for high-risk models. The goal is to add friction to the otherwise easy-to-download terminal-guidance software.
Acting at Market Speed
To state the glaringly obvious, the democratization of automated lethality is here, and it is going to stay. We cannot un-invent open-source algorithms, nor can we recall the millions of civilian quadcopters already in circulation. Slow as it may be, the UN establishes vital long-term benchmarks for meaningful human control and legal accountability. These are of great importance. But protecting forces and commercial shipping today also requires agile supply-side tools that raise the cost of turning commercial drones into autonomous or semi-autonomous weapons.
Since China can effectively leverage dual-use commercial technologies, inaction risks handing the strategic advantage to Beijing. The US can act swiftly and move almost as fast as the market does to adopt export controls, corporate oversight, and screening measures on dual-use technologies. Until long-term rules catch up, these will serve as effective policy levers in regulating autonomous systems.
Two clocks are running. In November, the High Contracting Parties decide what to do with GGE elements such as characterization, international humanitarian law, human judgment, and responsibility that stays with people. The other clock is the conversion cycle on commercial frames. Export thresholds and conversion-resistant firmware are crucial for governing the second clock. Washington cannot treat denial as a substitute for international rules and expect to have much to say about control when it gets to Geneva. But the reverse is equally problematic, as Geneva will not stop a commercial drone from being turned into an autonomous weapon if the components are already in the hands of those who want such a weapon. By then, the last mile will already have been fielded.
