Questions for Congress: AI in Targeting and Civilian Protection

The discussion centers on the use of Artificial Intelligence in Department of Defense targeting operations, specifically concerning autonomous weapons systems (AWS) and AI Decision Support Systems (AI-DSS). The text posits that AI is driving a philosophy of speed and scale in warfare, raising fundamental legal and ethical questions regarding civilian harm. A key focus is on the division between AWS, which can select and apply force without human control, and AI-DSS, which aids decision-makers through data analysis and recommendations, exemplified by the Maven Smart System (MSS). While MSS integrates vast data to facilitate rapid targeting, the use of these systems creates a "black box" regarding human roles, legal constraints, and accountability.
The inquiry prompts detailed questions for Congress regarding the process of targeting decisions, including how human judgment is integrated into the design and deployment of these systems. Specific concerns arise about the time allotted for human review of AI-nominated targets, the quality and accessibility of information provided to reviewers (such as underlying data and model confidence), and mechanisms for independent verification. Furthermore, there are specific concerns raised by incidents like the Iran war regarding flawed intelligence leading to civilian harm and the necessity of establishing clear lines of responsibility when AI systems cause errors. The text also addresses the role of private commercial partners in these systems and explores international safeguards for autonomous weapons.

* Cameron Stanley, Chief Digital and Artificial Intelligence Officer of the Department of Defense, asked how to achieve better decisions faster than adversaries at AIPCon 2026.
* AI capabilities in U.S. military operations are divided into Autonomous Weapons Systems (AWS) and AI Decision Support Systems (AI-DSS).
* AWS can select and apply force without human control, including functions like navigation and intelligence.
* AI-DSS systems use AI to analyze data and provide actionable recommendations to decision-makers; Maven Smart System (MSS) is the flagship system for the DoD.
* MSS integrates various platforms, data sources, and tools to recommend targets and select munitions, reducing the time from detection to strike.
* DoD employs both AWS and AI-DSS in targeting operations.
* A loophole exists where expert analysis suggests AI-DSS may fall outside the scope of DoD Directive 3000.09 regarding autonomous weapons testing and human judgment.
* Questions address the time spent by human reviewers on target approval in conflicts, the information available to reviewers (sources, model confidence), and the rate at which human reviewers reject or modify AI nominations.
* Concerns were raised regarding errors in the Iran War targeting, specifically a strike on an elementary school in Minab resulting from flawed intelligence in MSS.
* Questions address responsibility when AI-assisted strikes cause civilian harm, including the role of command, system, and operators.
* The text questions the extent to which lessons from after-action reviews of AI-assisted strikes are fed back to developers and units.
* Private partners like Palantir and Anduril are involved in developing these systems using third-party components.
* Questions are posed regarding vendor access to operational data and imposing safety limits on other AI companies.
* The U.S. position on requiring human review of AI-generated targets was debated during CCW deliberations.

The tension presented in the material lies between the demonstrable operational utility of AI—speed and scale—and the established legal, moral, and accountability frameworks designed to govern lethal force. The structure of the questions reveals a systemic gap: the technology is deployed rapidly (evidenced by the rise in target generation rates) before robust governance structures can be fully formalized or adapted.
The focus on the MSS system highlights that AI-DSS, rather than just autonomous weapons, represents a more pervasive shift in decision-making authority. The queries about embedded human judgment and constraints throughout the targeting cycle suggest an attempt to locate responsibility within the complex layers of AI integration, moving beyond a simple "human-in-the-loop" designation toward scrutinizing the entire chain of algorithmic influence.
The analysis of the Iran incident—where flawed intelligence in MSS led to civilian casualties—serves as a critical demonstration of how systemic failures in data integrity and verification can be amplified by the mandate for speed inherent in AI systems, leading to outcomes that defy traditional notions of accountability. This moves the discussion from *if* AI can cause harm to *who* is responsible when opaque systems generate errors, especially when intelligence feeds are implicated.
The concerns about cognitive offloading and deskilling point toward a potential long-term erosion of human expertise; if decision pathways become heavily reliant on automated recommendations, the capacity for independent, context-sensitive judgment may atrophy. The challenges posed by private vendor involvement introduce a layer of complexity regarding external oversight, suggesting that accountability must extend beyond purely governmental structures into the supply chain and external regulatory settings.
The pattern suggests an effort to impose retrospective legal and procedural requirements onto technology that is inherently designed for rapid, high-velocity operation, forcing a confrontation between technological velocity and deliberative necessity. The lack of clear protocols for tracking AI-derived data and outcomes indicates a struggle to apply existing accountability structures in novel, algorithmically mediated conflicts.
Bridge Questions:
If the current framework cannot accommodate the speed of AI targeting, what new temporal or procedural standards must be established to ensure that human oversight remains meaningful rather than merely ceremonial? How can international norms evolve to address the potential for systemic failures arising from autonomous data processing across different operational contexts? What mechanisms are necessary to ensure that accountability for AI-induced errors is traceable through multi-layered systems, including those involving private contractors and fused intelligence databases?