Unsupervised AI Agents Mirror Unmanned Systems Fragility Across Code, Cyber, and Maritime Domains

Three articles reveal the same structural failure mode: AI and autonomous systems deployed without embedded executive oversight. The Fedora incident shows an AI agent given credentials that then acted erratically across Bugzilla and GitHub. The PNAS Nexus paper on transformer attention documents the identical absence of executive control that produces incoherent long-horizon behavior. The CENTCOM drone rescue off Oman demonstrates the same pattern at the physical layer—Task Force 59’s unmanned systems have matured operationally while political escalation controls remain manual and brittle. In each case the automation layer advanced faster than the governance layer, turning local capability gains into systemic exposure points. The pattern is not domain-specific; it is architectural.