Direct answers.
Clear boundaries.
A guide to the concept, the independent policy proposal, and the questions the available evidence cannot yet answer.
What is Mutual AI Resilience?
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.
How would Mutual AI Resilience work?
The proposal is to share limited, verified warnings, let authorized people assess them locally, and coordinate help where participants agree. Essential functions would need to keep working if the shared connection failed. These arrangements have not been validated by this project.
Does MAR mean a single AI system controls several countries?
The proposed design keeps systems under separate local authority. Information sharing would need to be limited and revocable, with a tested ability to operate independently. Whether those safeguards work is itself a research question.
Is Mutual AI Resilience proven to prevent conflict?
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.
What does “mutual” mean here?
It means participants would seek a shared benefit from helping one another maintain essential civilian services. Benefits, responsibilities, and risks may be unequal; reciprocity has to be assessed rather than assumed.
How is MAR different from deterrence?
MAR centers on continuity of civilian services and the possible benefits of cooperation. Deterrence centers on influencing behavior through expected consequences. The library includes their possible interaction as an untested question, with explicit attention to escalation risk.
What can I do in the Resilience Lab?
Choose one to three strategies alongside MAR and inspect their source-backed pairwise hypotheses. You can compare assumptions and export your selection. The lab does not calculate a combined outcome or generate a prediction for an unsupported combination.
Where does the research come from?
The initial library comes from the supplied MAR link matrix: 20 proposed mechanisms, downsides, and observable tests. The source was undated; it was added to this site on 5 November 2026. Plain-language explanations and proposed test plans are editorial additions.
What would make a hypothesis more credible?
Independent review, reproducible tests, comparisons against a baseline, and documented results could strengthen or weaken a hypothesis. Historical validation and calibration would be needed before making forecasting claims. A plausible explanation alone is insufficient.
Does Safe Nations use live crisis data?
No. This edition has no live-event feed, country scoring, or automated crisis advice. A future event-based feature would need separately sourced, dated snapshots, documented revisions, and independent verification.
How does the event cooperation map work?
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.
What would a policymaker actually commit to?
The platform proposes a limited civilian mandate, verified and revocable information sharing, local operational control, privacy safeguards, independent review, and staged evaluation before expansion. Proposed commitments are policy choices; the expected benefits still require evidence.
Would MAR require a new international authority?
The proposed approach starts with separately accountable participants and limited agreements. It does not assume that a new supranational body controls national infrastructure. Any coordinating role would need a clearly bounded mandate, oversight, and an exit process.
Why use AI rather than existing warning channels?
That is a question for evaluation. A pilot should compare the proposed AI-supported process with existing channels and a simpler non-AI alternative. If a simpler process provides the same or better result at lower cost and risk, the proposal should be revised accordingly.
What would a first 100 days accomplish?
The illustrative agenda focuses on defining a civilian use case, mapping authority and dependencies, setting safeguards, and publishing an evaluation plan. It does not promise an operational international system, a treaty, or demonstrated peace benefits within 100 days.
How much would the proposal cost?
No credible budget or savings estimate is available yet. A scoped proposal would need to identify staffing, integration, verification, privacy, security, independent evaluation, maintenance, and exit costs before seeking a specific funding decision.
What would stop a pilot?
Examples include unauthorized data sharing, inability to operate independently, repeated unverifiable warnings, unacceptable civilian harm, or failed oversight. The pilot plan should define who can pause it and what evidence is required before restarting.
How could the public hold the work accountable?
The proposed governance model includes a public purpose statement, published evaluation criteria, documented decisions, independent review, accessible summaries of results, and a way to raise concerns. Sensitive information would need a protected review path rather than automatic public disclosure.