Chick Edmond
6 min
The increasing number of artificial intelligence (AI) elements within military systems has introduced new forms of security dilemmas related to speed, level of secrecy, and transfer of responsibility from humans to machines. This article addresses the question of whether or how AI enabled autonomous weapons can lead to greater levels of strategic instability. Three causal mechanisms were determined by this study to potentially create destabilizing effects due to the introduction of autonomy; the first mechanism is a reduction in time available for decision making. The second mechanism involves the creation of multiple pathways of escalation. The third mechanism is the loss of human control over decisions made using autonomous weapons. While the authors do not assert that any of the previously described mechanisms have been proven empirically, they instead make the narrower assertion that current doctrine development and operational trends in both the U.S., China, and Russia suggest that an AI based security dilemma is developing. A qualitative content analysis was conducted of official military doctrine documents, national AI strategy documents, and policy documents (2016–2024). Findings indicate that all three countries are currently developing capabilities that include AI elements that are creating potential risks for compressed decision-making times, fragmented escalation pathways, and a diminished role of human oversight. Furthermore, findings indicate there is considerable variability among the countries’ doctrines for achieving strategic objectives, their tolerance for risk, and their ability to be transparent about the use of autonomous weapons. In conclusion, absent effective governance frameworks that establish limits on the use of autonomous weapons, the increasing use of autonomous weapons will likely reduce overall strategic stability; however, this conclusion should be viewed as a likely risk rather than a certain fact.
This paper investigates whether and how the adoption of AI-enabled autonomous weapons by major powers—specifically the United States, China, and Russia—contributes to strategic instability. The author develops the concept of an "AI Security Dilemma," extending Robert Jervis’s classical security dilemma to account for the unique technological properties of AI. Using a qualitative content analysis of official military doctrines, national AI strategies, and policy documents from 2016 to 2024, the study evaluates these nations against four criteria: doctrinal approach to autonomy, governance of human control, integration with escalation frameworks, and transparency.
The author identifies three causal mechanisms that, when combined with AI, create a self-perpetuating cycle of insecurity. First, the speed of AI-enabled systems compresses decision-making timelines, forcing states to delegate authority to algorithms and creating a "use it or lose it" dynamic. Second, the decentralized nature of autonomous systems creates fragmented, incremental escalation pathways that are difficult to track or attribute. Third, the "black box" nature of deep learning and the speed of operations lead to an erosion of meaningful human control, making it difficult for states to signal intentions or maintain trust, thereby forcing adversaries to assume the worst-case scenario.
The analysis reveals that while the U.S., China, and Russia exhibit different strategic cultures—ranging from the U.S. focus on "decision superiority" to China’s "intelligentized warfare" and Russia’s use of automation as a force multiplier—all three are developing capabilities that align with the identified risks. A significant "transparency gap" exists across all three nations, where strategic ambiguity and a lack of disclosure regarding operational AI practices exacerbate mutual distrust. The author concludes that without effective international governance frameworks, the pursuit of AI-driven military advantage will likely reduce overall strategic stability.
Sam: [slower and more deliberate] Precisely. And the paper's survey of military doctrine in the U.S., China, and Russia suggests all three are moving toward this model—not because any of them wants instability, but because the competitive logic pushes in that direction regardless. That said, this is where the methodological constraint bites hard. The analysis rests entirely on public doctrine. We're not seeing classified operational realities, and public doctrine can diverge significantly from what systems are actually designed to do. So what Edmond is mapping is a structural tendency, not a confirmed trajectory.
Alex: [beat] That's an important distinction. It's a framework for understanding a risk, not an empirical record of how close we've come.
Sam: [affirming] Exactly. Think of it as a map of the danger zone rather than a record of a near-miss. The value is analytical—it gives you a vocabulary for the failure modes before they occur. [[RP_SECTION:governance-and-verification|Governance and Verification]]
Alex: Which brings us to the governance side. The paper identifies what it calls a "governance gap." What's the actual content of that argument?
Sam: [steady] The crux is a verification problem. Major powers disagree on what counts as "meaningful human control," so arms control negotiations stall at the definitional level. And even if you got past that, the dual-use nature of the underlying technology makes verification essentially unworkable. You can't audit an adversary's AI software without revealing your own intelligence methods. So you're stuck in a cycle of self-help, where each side prioritizes tactical advantage over shared norms, which deepens the uncertainty for everyone.
Alex: So prohibition is off the table. What does Edmond actually propose? [[RP_SECTION:attribution-stability-proposal|Attribution Stability Proposal]]
Sam: [quiet confidence] He pivots to what he calls "attribution stability"—a narrower, more tractable goal than arms control. The idea is to mandate that autonomous systems generate audit trails or digital watermarks for their actions. You're not trying to control what the weapon does; you're trying to ensure that if it acts, you can trace the decision back to its source. The logic is that even partial transparency about intent could interrupt the "flash war" dynamic—if you can verify that an action was a system error rather than a deliberate strike, you have a basis for standing down rather than retaliating.
Alex: [deliberate] That's a meaningful reframe. It shifts the goal from preventing autonomous action to making autonomous action legible after the fact.
Sam: [measured] And it's politically more realistic than prohibition, which is presumably why Edmond lands there. Though it does raise a question the paper doesn't fully answer: whether adversaries would actually trust audit trails generated by the other side's systems, or whether that just introduces a new layer of verification problems.
Alex: [reflective] So the paper is most useful as a diagnostic tool—it gives you a precise account of the structural mechanisms driving instability, and a direction for governance that's at least tractable, even if the implementation questions remain open.
Sam: That's a fair read. The theoretical framework is the load-bearing contribution. The governance proposals are directionally interesting but underdeveloped. For a researcher working on AI policy or international security, the value is in the failure-mode taxonomy—time compression, fragmented escalation, attribution breakdown—and in the argument that these risks are structural rather than contingent on any particular actor behaving badly.
Alex: Which makes it harder to solve, but also more important to understand clearly. Thanks for walking through this one, Sam. And thanks to everyone listening to ResearchPod.