Claude Integrated into US Military's Project Maven: AI-Assisted Target Selection, Humans Retain Final Decision-Making Power

According to the latest reports, the AI model Claude, developed by Anthropic, has been integrated into the US Department of Defense's Project Maven system to assist in the identification and prioritization of military targets. The system can integrate multi-source intelligence, generate lists of potential targets, and accelerate analysis processes in time-sensitive scenarios. In addition, it can also be used for post-war assessments, quickly comparing actual strike effects with expected targets, providing decision-making references for commanders.

However, all decisions involving lethal force are strictly reserved for human commanders. AI only provides advisory suggestions, and final authorization must be made by personnel with legal authority in accordance with established procedures. This mechanism embodies the core principle of "human-in-the-loop": humans not only need to review AI output, but also need to have the ability to suspend or modify suggestions at any time, and bear clear responsibility for the results.

This architecture aims to improve operational efficiency while strengthening accountability mechanisms and ethical constraints. While generative AI can process massive amounts of data and discover patterns that humans easily overlook, it also carries the risk of misjudgment or overfitting. Therefore, the entire system operation needs to record the basis for decisions, ensuring that every action is traceable and auditable, in order to minimize the risk of civilian casualties. Currently, there are different claims regarding the specific geographical scope of the system's deployment and the number of targets – some reports mention Iran, while others mention Iraq, but related details have not been independently verified. There are also policy differences between the US Department of Defense and Anthropic regarding the boundaries of AI use: the former advocates for "full coverage of legal uses," while the latter insists on stricter ethical restrictions. Despite the questionable details, the core principle of "human-led decision-making" remains consistent in all descriptions, reflecting the current mainstream consensus on military AI applications: technology is a tool, and responsibility is never outsourced.