Stronger together.
Independent by design.
Our primary goal is to secure cooperation and lessen the chance of war between nations. Mutual AI Resilience explores how countries can protect essential civilian systems while keeping control of their own decisions.
Peace is power. Read the founding essay ↗
Safe Nations exists to help nations find practical reasons to cooperate. Mutual AI Resilience connects that goal to the systems that keep people alive, informed, and able to choose their own future.
“A living population connected to resilient AI, energy, food, trade, and space infrastructure is power.”
Jesse Shelley · Peace Is Power
Explore Jesse’s related projectsPast disruptions reveal concrete coordination problems. Explore the documented events, the people a MAR-style process could help, and the evidence needed to judge that benefit.
Explore all real-world examplesA cyberattack halted a major U.S. fuel pipeline carrying gasoline, diesel, heating oil, and jet fuel. U.S. Energy Information Administration · 11 May 2021
Connect verified supply updates with the people planning essential fuel deliveries.
Attacks on commercial shipping disrupted Suez routes, prompting diversions and adding time and cost. UN Trade and Development (UNCTAD) · 22 February 2024
Help ports, carriers, importers, and public authorities plan from compatible, verified updates.
IEA members coordinated emergency oil stock releases after Russia’s invasion of Ukraine. International Energy Agency · 14 April 2022
Test whether shared, verified analysis improves the fit between available supply and a civilian need.
Historical events are sourced below. The MAR applications and benefits are Safe Nations’ proposals; these events do not establish a tested MAR effect.
Imagine conflicting AI-assisted alerts, disputed evidence, and interruptions to an essential supply route affecting two countries. Civilian response teams need to establish what is happening, preserve essential services, and clarify the crisis before uncertain claims drive consequential decisions.
Compare approved reports, provenance, and uncertainty across independent sources. Keep disputed claims visibly unresolved until corroborated.
Evidence checks and evaluationBring civilian operators, evidence reviewers, and authorized diplomatic contacts into a limited clarification process. Separate the disruption from claims about responsibility.
Roles and decision authorityCompare essential-service needs with available support and constraints. Agree on practical assistance without transferring operational control.
How to structure an exerciseAuthorize each local action, retain a tested fallback, and communicate consistent corrections and service updates.
Continuity and safeguardsReview consultation time, verification errors, service continuity, and privacy or dependency failures against the agreed baseline.
How results are interpretedChecks reports against domestic service conditions, identifies authorized assistance needs, and retains responsibility for local decisions.
Independently checks the reports, shares approved evidence and available support, and raises constraints before agreeing to joint work.
Coordinates traceable checks, records disagreement and corrections, and supports an authorized diplomatic clarification channel.
Could a verified civilian cooperation channel improve clarification and continuity during a compound AI-related crisis?
Time between verified warning and first cross-border consultation.
False confidence if warning data are politicized.
Illustrative multi-domain exercise design. The linked hypotheses remain separate and untested; no combined outcome is calculated.
Showing the compound AI crisis cooperation example.
These are editorial illustrations and proposed exercises. Each underlying hypothesis remains untested, and no combined effect is inferred.
Essential services can be affected by disruptions that cross borders. A shared warning may help—but only if it can be checked, understood, and used responsibly.
Mutual AI Resilience asks whether limited cooperation can improve civilian continuity without creating a new source of dependence. AI could help organize reports and identify inconsistencies. Authorized people and institutions would decide what to share and how to respond.
The proposal brings technical design together with diplomacy, privacy, procurement, local fallback, and independent oversight. Each is part of the same public-security question.
Understand the full MAR frameworkA developed policy proposal should explain what institutions would commit to, who would be responsible, and what evidence would change the decision.
Read every commitment and trade-offCooperation should strengthen a country’s capacity to act and preserve a practical ability to withdraw.
Build limited channels for verified civilian warnings and clarification, including when wider political relations are difficult.
A civilian continuity program needs an explicit boundary around the information it may collect and use.
Prioritize the electricity, water, food, health supplies, and communications that people depend on.
Every material decision should have an identifiable authority, a reviewable record, and a route for challenge.
Fund clear questions, controlled comparisons, and independent evaluation before making broad performance claims.
Treat integration, verification, oversight, maintenance, and exit as part of the full cost of a proposal.
Explain the public purpose, publish meaningful results, and protect the independence of oversight.
A narrow use case makes it possible to define the authority, comparison, outcomes, and failure conditions. These are proposed areas for evaluation.
Can coordinated, verified information reduce avoidable essential-cargo delays?
Can electricity or essential fuel continuity improve while participants retain independent control?
Can participants establish a shared observation without assuming agreement on rights?
Does a verified alert improve continuity compared with existing planning?
Can preparedness improve without creating disproportionate privacy or protection risks?
Can a recipient check a claim and receive a correction without relying on a single authority?
Do the procedures remain useful and legitimate when trust is low?
An illustrative first 100 days would establish the question, responsibilities, safeguards, and evaluation plan. Further work would depend on reviewed evidence and a separate decision.
Review the implementation roadmapName the service, purpose, authority, baseline, and alternatives.
Examine limited sharing, human authorization, disconnection, and withdrawal.
Report outcomes, harms, costs, uncertainty, and reasons to stop or continue.
The founding matrix links MAR with peace and resilience strategies. Every record preserves its proposed mechanism, downside, observable test, and untested status.
Inspect the complete research libraryLimited exchanges between independently run systems could support cooperation while preserving each participant’s authority.
Untested hypothesisVerified warnings about electricity disruptions could help participants coordinate restoration.
Untested hypothesisIndependent checks of where evidence came from could help identify fabricated crisis information.
Untested hypothesisAll 20 source connections remain untested hypotheses. The platform and implementation plans are independent proposals. The available material establishes no validated MAR effect, combined outcome, or probability of success. Review the evidence standard.
Mutual AI Resilience (MAR) is a proposed framework for countries to protect essential civilian services through verified warnings, AI-supported analysis, and voluntary cooperation, while retaining sovereign decision-making, civil liberties, and the ability to operate independently.
No. Every connection in the current Safe Nations library is an untested hypothesis. The material does not establish a causal effect on peace, assign success probabilities, or predict outcomes for particular countries.
Choose a disruption to see an illustrative exchange of needs, available support, and agreed work between participants. Each example links to an original untested hypothesis and shows its observable measure and risk. The examples are editorial illustrations, not actual incidents or validated scenarios.