The Materiality Gate: AI Sovereignty Risk Under Geopolitical Shocks

Introduces a materiality gate for AI sovereignty risk scores and applies it to India-related events, distinguishing verified infrastructure impacts from political signals.

Endorsed by

Cibeles

Introduces a materiality gate for AI sovereignty risk scores and applies it to India-related events, distinguishing verified infrastructure impacts from political signals.

Endorsed by

Cibeles

Project Details

The Materiality Gate paper argues that AI sovereignty risk scores should only change when a geopolitical event passes three tests: it must be verified through credible sources, it must exercise real authority over a specific infrastructure dependency rather than just general political influence, and it must have a documented material pathway showing how it actually changes access to accelerators, cloud regions, models, or safety monitoring. Events that fail this gate stay on a watchlist with a review trigger instead of moving the score, while events that pass enter one of four states, WATCH, PENDING, ACTIVE, or WITHDRAWN, preserving the gap between announcement and operational reality. Applying this to India between 2025 and 2026, the authors find that of seven tested events, only the 2025 US AI Diffusion Rule (later WITHDRAWN after rescission) and the announced 20,000 additional GPUs for the IndiaAI Mission (still PENDING) reach a material state, while the disputed Trump mediation claim, his meeting with Pakistan's army chief, the Islamabad Memorandum of Understanding, and the Pacific Command renaming all remain in WATCH since none of them documents a real change to an AI dependency. The framework's key strength is its symmetry, holding favorable domestic announcements to the same evidentiary standard as adverse geopolitical signals, and its treatment of safety continuity as its own exposure dimension, since equal capacity losses can have very different effects depending on whether safety evaluation and incident response keep running, all while acknowledging that public evidence is incomplete, the scale is for comparison rather than precise probability, and the method explicitly avoids judging whether US policy favors either country since favor itself is not an infrastructure variable.

Hackathon paper: The Materiality Gate: Dynamic Updating of AI Sovereignty Risk under Geopolitical Shocks

Theory of Impact

AI sovereignty matters for AI safety when control over infrastructure determines whether institutions can evaluate advanced models, monitor deployments, preserve audit evidence, and respond to incidents. The project targets public compute programs and AI evaluation organizations making procurement and continuity decisions under geopolitical uncertainty.

Two errors can leave safety work without a usable fallback: treating a possible restriction as an existing cutoff, or treating announced replacement capacity as already operational. The gate is designed to distinguish present conditions from forecasts while allowing preparation before a disruption materializes.

The intended causal pathway is traceable evidence, better-supported preparedness and substitution decisions, and fewer avoidable interruptions to safety oversight. For example, an evaluation program should not retire its backup capacity merely because a domestic cluster has been announced.

The research will test whether the protocol adds practical value over a simpler checklist. Adoption and improved safety continuity remain outcomes to establish, not assumed consequences of publication. The objective is dependable control and usable alternatives for safety work, not unrestricted capability expansion or circumvention of legitimate controls.

People

Subramanyam Sahoo

Team Member

Track Record

Global South AIS Hackathon (Apart Research), 2026

Grants Received– no grants recorded

Add grant

Funding Asks

The Materiality Gate: Dynamic Updating of AI Sovereignty Risk under Geopolitical Shocks

Open

Minimum

$30,000

Ideal

$60,000

What this ask is

I would test when geopolitical developments change AI sovereignty and what preparation they justify. The Materiality Gate traces an actor’s action through a named infrastructure dependency to changes in access, control, substitution options, and safety continuity. It applies the same evidence standard to foreign restrictions and domestic capacity promises.

With $30,000, I would conduct a six-month study remotely from India, targeting 60 documented event histories across India and Brazil. I would first audit existing cases and publish sampling and evaluation rules. Cases would cover export controls, provider restrictions, and compute commitments progressing to commissioning, delay, or cancellation. Each history would follow one action or commitment over time; linked country impacts would not count as independent shocks.

Twenty histories would support protocol development; 40 would evaluate the frozen protocol. Related events would remain in the same split. I would reconstruct dated evidence packets and record subsequent outcomes separately, retaining unresolved cases where evidence is insufficient.

I would compare the gate against a simpler evidence checklist on unsupported updates, missed documented changes, detection delay, and evidence requirements. Reference findings would describe observable legal or operational facts rather than reproduce the gate’s labels. Single-analyst judgment and historical hindsight would remain explicit limitations.

The minimum grant would deliver an event database, focused five-layer rubric, reproducible scoring code, an empirical technical AI governance manuscript, a policy brief, and a public evidence website.

The $60,000 total budget would extend the program to twelve months, add Vietnam, and target 120 histories overall. Subject to the initial findings, I would add prospective tracking and 12 separately labeled hypothetical shock sequences testing delayed restrictions, incomplete substitutes, and reversals with residual effects.

I would perform all research, verification, analysis, coding, website development, and writing myself. No contractors or international travel are budgeted.

How the money will be spent

Amounts below show the six-month minimum / twelve-month ideal. The ideal request is $60,000 total, including the initial six-month project. I will perform all research and implementation personally; no assistants, contractors, paid independent reviewers, or international travel are included.

Principal researcher support: $18,000 / $36,000. Six or twelve months at $3,000 per month for methodology development, geopolitical research, documentary verification, dataset construction, evaluation, rubric and scoring implementation, website development, and preparation of the paper and policy brief. All my labor is included here and will not be charged again under other categories.

Data and archival access: up to $3,000 / $6,000. Historical policy documents, archived provider terms, procurement records, and relevant infrastructure datasets where public sources do not resolve an identified evidence gap. Before purchasing access, I will document the source, price, research question it supports, and why a free alternative is insufficient. Public releases will respect licensing restrictions.

Compute, APIs, and analysis software: up to $3,000 / $6,000. Document extraction, searchable research indexes, assisted retrieval, data processing, analysis software, and reproducibility runs. I will estimate processing volume and service costs before committing expenditure and track actual usage. Extracted claims will be checked against source documents; model outputs will not serve as independent evidence. No large model-training run is planned.

Website, storage, and reproducibility infrastructure: up to $1,500 / $3,000. Domain registration, hosting, database services, storage, backups, and deployment. The website will provide source-linked event records, assessment histories, uncertainty notes, and downloadable data and code. I will use a lightweight implementation and document the selected service configuration and recurring costs. Engineering labor is already covered by researcher support.

Publication, accessibility, and dissemination: up to $2,000 / $4,000. Applicable publication charges, remote conference participation, accessibility tools, and direct expenses associated with releasing the research. Free repositories and publication routes will be prioritized. Publication acceptance and paid publication charges are not assumed. Writing, figure preparation, and documentation labor are already covered by researcher support.

Administration: up to $500 / $1,000. Eligible payment-processing charges, recordkeeping tools, and other documented grant-administration expenses. Administrative labor performed by me is included in researcher support.

Contingency: up to $2,000 / $4,000. Necessary, unforeseen project expenses, such as additional archival access or unexpected processing requirements. This reserve is not additional compensation or a spending target. Its use and any reallocation will follow the agreed grant terms.

Total: $30,000 for six months / $60,000 for twelve months.

Non-labor amounts are planning allowances rather than confirmed supplier costs. Before the award is finalized, I will provide itemized estimates and adjust the operating request where those estimates do not support the proposed allocation.

I will provide quarterly progress and spending reports linking expenditure to completed evidence records, evaluation results, code, and public releases. The second phase will be reviewed after six months against evidence availability, performance relative to the simpler checklist, and the value of further expansion. Unspent funds will be handled under the grant agreement rather than spent to exhaust the budget.

Discussion

Private comments in this thread are only shown to approved funders and grant reviewers. Email hi@grantmaking.ai or to request access.

Private UserPrivate15d?

AAnonymous judgeJudge · Global South AIS Hackathon15d

I love how sharp and ready-to-use this project is. The matching rule where favorable announcements go through the same gate as adverse geopolitical signals is a great design choice. A potential next step is a trustworthiness test: have a second reviewer track the same India events on their own to see if the tool yields the same results. Adding one or two more nations, like Taiwan or South Korea, would test if this five-layer setup actually works everywhere.

Cibeles Garcia BurtJudge · Global South AIS HackathonEndorsement15d

I greatly admire what you built because it is rigorous and ready to operate: a three-test gate, a four-state machine, layer-by-layer rubrics with concrete anchors, and a coding protocol. The symmetry property is the standout idea, applying the identical skeptical test to a favorable GPU announcement and an adverse diplomatic event directly targets the analyst bias that inflates risk on threats and deflates it on press releases, and the positive-control design (deliberately including the Diffusion Rule as an event that should pass) pre-empts the obvious objection that this is just a machine for dismissing geopolitics.
My main reservation is that the proposal is validated on a single case coded by a single analyst, so reproducibility is specified but not demonstrated. For example, the gate's judgment calls are exactly where two analysts might diverge, and with one coder and one country you can't see that variation; a second coder on the same events, or the same gate on a second country, would turn the framework into a validated one. Something to consider when it comes to framing since it changes how you'd pitch this to a government: sovereignty scores inside a real foreign ministry are political artifacts shaped by who's in the room, not dispassionate measurements which means the gate's deepest value may be less "more accurate scoring" and more "a discipline that holds the line against political pressure to move a number." Leaning into that bureaucratic function is both more honest and more sellable.