FATHOM - A Physics-Bounded Framework for Human Control of Armed Uncrewed Underwater Vehicles under Intermittent Communications

FATHOM: A Physics-Bounded Framework for Human Control of Armed Uncrewed Underwater Vehicles under Intermittent Communications Framework for Authority, Time and Human Oversight in Maritime autonomy Uncrewed underwater vehicles (UUVs) are beginning to carry lethal payloads. In September 2026 a US Mk 48 heavyweight torpedo was launched from the Royal Navy's XV Excalibur. Existing work establishes that meaningful human control over autonomous weapons requires a practical ability to intervene, and that underwater communications make such control difficult. It does not provide a way to calculate, for a specific armed UUV mission, whether a human can intervene in time, or how long a pre-delegated authority remains valid once contact is lost. FATHOM is a design-science framework that fills this gap with four contributions: Loop Closure Ratio with two loops. The framework compares the engagement window with the latency of a full-evidence decision loop (human in the loop) and of a compact-alert veto loop (human on the loop). Both are evaluated probabilistically under intermittent acoustic links. Three-way separation of validity. Scene Observation Freshness is empirical. Authority validity is a governance construct combining continuous decay with Class I and Class II invalidating events and a human-set threshold. Assessment validity is a legal judgement that stays outside the framework. Routed control zones. Four zones: Z1 Direct, Z2 Supervised, Z3 Delegated and Z4 Hold. Consistency floors can only tighten control, and lost-link rules state that reconnection is not re-authorisation. Evidence-tiered, reproducible model and research tool. Every parameter is labelled vendor, field or assumed. Key findings (model results under stated assumptions; Monte Carlo with 10,000 draws per evidence band) Under field-anchored link conditions, assured human-in-the-loop control within five minutes is essentially unavailable beyond 1 km. Veto supervision within five minutes remains assured in 54% of draws at 10 km. Link bandwidth, evidence size and packet loss dominate loop latency; human decision time ranks fourth. Reducing the decision-evidence package from 400 to 25 kbit raises assured in-the-loop control at 10 km within 15 minutes from 3% to 86% of field-anchored draws. Relays extend reach but, under a constant rate-range product, do not reduce transmission time. Vendor specifications materially overstate controllability. A review of primary national texts from the United Kingdom, Australia, India and the United States found no public rule for weapon behaviour under degraded or lost communications. FATHOM proposes one. Relation to CALIBRE. FATHOM extends the author's CALIBRE governance framework (doi:10.5281/zenodo.22707378). CALIBRE determines that an AI weapon function requires human oversight. FATHOM determines whether that oversight can physically be exercised at sea, in which form, and when delegated authority must end. Its output feeds back to CALIBRE as a proposed human-intervention-time sub-score. Three worked cases show the full lifecycle: an open-ocean anti-submarine barrier; seabed-cable protection in a congested littoral; an allied chokepoint operated remotely. Research tool. The framework is implemented in an open web tool that reproduces every table and figure in the paper: https://calibre-fathom.onrender.com/fathom. The free hosting tier may take up to a minute to start. It is verified against the reference model by 577 automated tests.