Intelligence notes

◉Dashboard↗

Paul Scharre

8 min read

Paul Scharre

Profile

Paul Scharre is a Senior Fellow and Director of the Technology and National Security Program at the Center for a New American Security (CNAS). He is the foremost US policy analyst on lethal autonomous weapons systems (LAWS) and artificial intelligence in warfare — the author of the two most analytically rigorous English-language works on these subjects. His analysis is grounded in operational experience: Scharre served in the 75th Ranger Regiment as an Army Ranger and participated directly in drafting the 2012 Department of Defense Directive 3000.09, the foundational US policy document on autonomous weapons.

This practitioner-to-analyst trajectory distinguishes Scharre from most academic contributors to the LAWS debate. His assessments are grounded in how military systems are actually designed, acquired, and operated — not merely in the ethical and legal principles that ought to govern them.

Key Works

Army of None: Autonomous Weapons and the Future of War (W.W. Norton, 2018)

The definitive analytical treatment of lethal autonomous weapons systems available in any language. Scharre approaches LAWS through four interlocking lenses: definitional, taxonomic, ethical, and strategic.

Definitional precision: Scharre opens with a fundamental problem the arms control and ethics communities have consistently avoided — defining what “autonomous” actually means in a weapons context. He distinguishes between systems that automate the search for targets, systems that automate the selection of targets from a human-presented set, and systems that automate the engagement decision without human involvement. Each level of autonomy creates different accountability structures and different escalation risks. Policy debates that treat “autonomous weapons” as a unitary category fail before they begin.

The human-control taxonomy: Scharre introduces the organizing framework that has become standard in the field:

  • Human in the loop — a human makes each individual targeting and engagement decision
  • Human on the loop — a human can override an automated system but is not required to approve each action
  • Human out of the loop — the system selects and engages targets without human intervention

This taxonomy clarifies why many existing weapons systems already cross the “autonomous” threshold in some operational modes — and why the policy debate over a future “killer robot” often obscures current deployments.

The accountability gap: Scharre’s most operationally significant contribution. When an autonomous system kills unlawfully — targeting a protected person, a civilian object, a combatant who has surrendered — no individual has committed a war crime, because no individual made the targeting decision. The commander who authorized deployment did not select the specific target. The programmer who wrote the algorithm did not know this specific engagement would occur. The officer who approved the targeting parameters did not pull the trigger. International humanitarian law (IHL) requires human accountability; autonomous systems dissolve the actor required to bear it.

This accountability gap is directly applicable to the architecture of the IDF’s Gospel and Lavender AI targeting systems, where targeting recommendations are generated algorithmically and approved at high throughput by officers who cannot meaningfully review each individual recommendation. See The IDF’s Kill Machine.

The flash war risk: Scharre’s most analytically original strategic contribution. If both sides in a conflict deploy autonomous systems that engage at machine speed — milliseconds to seconds — a localized tactical engagement could escalate to general war within minutes, faster than any human decision chain could intervene to de-escalate. Scharre draws the analogy to the 2010 Flash Crash, in which algorithmic trading systems interacting with each other caused the Dow Jones to drop 1,000 points in 45 minutes before partial recovery — with no human able to intervene in real time. Applied to kinetic military systems operating against each other in a contested airspace or maritime zone, the result is not a market disruption but a war no political leadership intended to start.

This risk is qualitatively different from miscalculation risks in conventional conflicts. Human decision-makers can pause, communicate, signal — autonomous systems operating at machine speed have no such mechanism.

Four Battlegrounds: Power in the Age of Artificial Intelligence (W.W. Norton, 2023)

Scharre’s follow-on work extends the analysis from LAWS specifically to AI competition generally, arguing that the geopolitical contest over artificial intelligence is being decided across four domains: data, compute, talent, and institutions.

The book’s central analytical claim cuts against the dominant framing of US-China AI competition as a technology race: the decisive variable is not which country develops the most capable AI systems first, but which country can build sufficient institutional trust in AI systems to deploy them at scale — and which country can build the legal-regulatory frameworks that make such trust legitimate. An AI system that cannot be trusted by the soldiers, commanders, or legal structures that must use it cannot be deployed regardless of its technical capability.

Scharre is more optimistic about US institutional capacity than many analysts — a significant limitation. His comparison of US and Chinese institutional frameworks underweights the Chinese state’s demonstrated capacity to mandate adoption and override institutional resistance, and overweights the US regulatory framework’s capacity to build trust in accelerated AI deployment timelines.

Analytical Contributions

Meaningful Human Control

In international LAWS governance debates at the UN Convention on Certain Conventional Weapons (CCW), Scharre has been the leading US voice arguing that “meaningful human control” requires humans to understand and predict system behavior — not merely to be “in the loop” in a nominal or formal sense.

The distinction is consequential: a targeting officer who approves packages generated by an opaque machine-learning system — under time pressure, at high throughput, without visibility into the system’s decision logic — does not exercise meaningful control even if he formally authorizes each engagement. The human is performing a ritual function, not a deliberative one. Scharre’s standard demands genuine understanding of system behavior, not mere procedural presence in the targeting chain.

This argument has direct application to any algorithmic targeting architecture where human review is nominally maintained but operationally compressed to the point of meaninglessness.

Flash War as a Strategic Risk Category

Scharre has elevated the flash war risk from an obscure systems-safety concern to a recognized strategic risk category. The argument has influenced both the academic literature on AI risk and the policy debate over autonomous systems in great-power competition. It remains underweighted in most assessments of US-China or NATO-Russia conventional deterrence architecture.

Positioning Against Other Analysts

vs. Michael C. Horowitz

The two are complementary rather than competing. Horowitz’s adoption-capacity theory explains the political economy of why AI military capabilities diffuse unevenly across states — which states can absorb and deploy new technologies depends on organizational capacity, not just resources. Scharre explains the operational and ethical implications of what is being diffused: what autonomous systems actually are, what they do to accountability structures, and what strategic risks they create. Horowitz explains the diffusion pattern; Scharre explains what is being diffused and why it matters.

vs. P.W. Singer

Singer (Wired for War, 2009) wrote the popular introduction to military robotics and unmanned systems. Scharre’s work is analytically deeper and more current — Army of None supersedes Wired for War as the field’s standard reference. Singer’s strength is accessibility and breadth; Scharre’s is analytical rigor and policy specificity.

Critical Limitations

Liberal-Western normative framework: Scharre’s accountability gap argument and meaningful human control standard presuppose adversary willingness to be bound by IHL. The framework is designed for a world in which states share a commitment to the laws of armed conflict — a world that does not describe Russia, China, or non-state actors in their actual operational behavior. His prescriptions have limited traction against adversaries who do not accept the normative premises.

Institutional optimism in Four Battlegrounds: Scharre’s assessment of US institutional capacity to govern AI deployment is more optimistic than the evidence warrants. The US regulatory ecosystem for AI in national security contexts has demonstrated limited capacity for timely, coherent action. His comparison with China underweights the CCP’s capacity to mandate adoption at scale.

Deployment pace vs. governance pace: Across both books, Scharre’s governance-focused analysis consistently lags the pace of actual autonomous systems deployment. Systems he treats as hypothetical or emerging have been deployed — often by adversaries operating outside his normative framework — before governance debates have concluded.

Key Works in Vault

Sources

Graph View

Backlinks