Sławomir Krakowski · Zenodo (CERN European Organization for Nuclear Research) 2026 · 2026
DOI: 10.5281/zenodo.23104328
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
This article proposes a new way to think about human control over autonomous weapon systems. Instead of asking only whether a human is formally “in” or “on” the decision loop, it asks a more demanding question: when new evidence reveals that an autonomous system may be wrong, does a safe and practically executable path of correction still exist? The paper develops the concept of correction reachability and formalizes it through reachable safe sets, latency budgets, reversibility, partial observability, communication loss and decision pressure. It introduces several related concepts, including uncertainty-conditioned reversibility, reachable-future closure, second-order human control, correction memory and communication-loss inversion. The argument is tested through three computational pilot studies comprising more than 80,000 simulated episodes. The results show that systems with similar classification accuracy can differ substantially in their ability to preserve corrective options; that nominal human supervision can remain unchanged while effective correction collapses under degraded sensing, communication and time pressure; and that multi-agent coupling does not automatically produce escalation, but reciprocal reaction chains can sharply contract the set of safe reachable futures. The article connects control theory, human factors, AI safety, international humanitarian law and the governance of autonomous weapon systems. Its central claim is simple but consequential: formal authority to intervene is not the same as the practical ability to change a system’s trajectory. The paper is intended as a falsifiable, engineering-oriented contribution to the debate on meaningful human control, with implications for system design, testing, procurement, weapons review and governance.
No comments yet — start the discussion below.