The Controllability Trap: A Governance Framework for Military AI Agents

Subramanyam Sahoo · 2026

Read The Controllability Trap: A Governance Framework for Military AI Agents by Subramanyam Sahoo (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-03515.txt.

Description

Subramanyam Sahoo builds a control plan for military AI agents. Agentic systems interpret goals, model the world, plan, use tools, and coordinate over long horizons. Each capability breeds control failures that older safety frameworks miss. His 2026 paper names six such control failures and shows how they erode real human control in military settings, the field's stated goal.

The proposed approach organizes defenses in three pillars: preventive, detective, and corrective control. Its core instrument is a Control Quality Score, a combined real-time metric that puts a number on how much control humans actually hold. As the score weakens, graduated responses kick in. Every failure type gets concrete mechanisms, assigned duties across five bodies, and formal scoring metrics. A worked operational scenario shows the machinery in action.

Catalog record

CREATORSubramanyam Sahoo
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54364
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.03515

Browse the full library · About the Tower of Babel