A policy framework
for Mutual AI Resilience.
Eight proposed commitments connect national capability with voluntary cooperation. The objective is to protect essential civilian services while preserving sovereign authority, civil liberties, and accountable public decisions.
Keep essential civilian services available when disruption crosses borders.
Every participant retains authority, independent operation, and a practical exit.
Evaluate benefits, costs, rights impacts, and failure before expansion.
01. Preserve sovereign control
Cooperation should strengthen a country’s capacity to act and preserve a practical ability to withdraw.
A resilience arrangement fails its purpose if a participant becomes dependent on permissions, providers, or decisions it cannot control. Sovereignty has to be demonstrated through operating arrangements, contracts, and continuity exercises.
Proposed commitments
- Keep authority for material decisions with the legally responsible domestic institutions and service operators. A shared warning would inform a decision; it would not authorize a foreign actor to operate essential systems.
- Make participation specific, voluntary, and revocable. Agreements should define the information exchanged, permitted uses, recipients, review process, and conditions for suspension.
- Require a tested local fallback and a documented exit plan, including data handling, replacement services, staff responsibilities, and the cost of withdrawal.
How to assess the commitment
Demonstrate that essential functions remain available during a planned disconnection and that a participant can revoke access without losing its core operating capability.
Trade-off to address
Independent operation can require spare capacity, additional staff, and less convenient integration. Those costs should be visible before a cooperation arrangement is approved.
02. Use practical diplomacy to reduce misunderstanding
Build limited channels for verified civilian warnings and clarification, including when wider political relations are difficult.
A shared problem does not erase competing interests. A narrow, clearly authorized channel may still allow participants to clarify an emergency or coordinate a limited response. Whether it actually reduces confusion is a testable question.
Proposed commitments
- Define approved civilian purposes and named points of contact before an emergency. Connect new arrangements to existing diplomatic and emergency-management channels.
- Separate an observed disruption from an allegation about who caused it. Preserve uncertainty, dissenting assessments, and the ability to request clarification.
- Exercise delayed messages, disputed reports, translation problems, and withdrawal. A process should remain understandable when participants disagree.
How to assess the commitment
Compare verification time, consultation time, disputed notices, and misunderstandings against the current process under the same exercise conditions.
Trade-off to address
Additional verification can slow communication. Sharing information with a potential adversary can create risks even when the stated purpose is civilian protection.
03. Protect civil liberties and private information
A civilian continuity program needs an explicit boundary around the information it may collect and use.
A technically useful system can still be unacceptable if it expands access to personal information, weakens private communications, or makes decisions people cannot challenge. Privacy and civil liberties belong in the initial design and funding decision.
Proposed commitments
- Use the least sensitive information that can answer the agreed civilian question. Record the purpose, access rules, retention period, and authority for each information category.
- Exclude political profiling and routine monitoring of lawful public debate from this proposed civilian program. Do not treat disagreement with a warning as proof of a security threat.
- Preserve secure communications and provide a documented route for complaints, correction, and independent review. Proposed protections must work in practice during an emergency.
How to assess the commitment
Verify that approved purposes, access restrictions, deletion rules, and complaint procedures are followed. Record violations and remedies, alongside any operational benefit.
Trade-off to address
Reducing the amount of information shared may reduce analytical detail. The program must show why each additional data category is necessary instead of treating more access as inherently better.
04. Make civilian service continuity the objective
Prioritize the electricity, water, food, health supplies, and communications that people depend on.
The public purpose should be expressed in concrete service outcomes. A faster model, larger dataset, or more connected network is useful only if it contributes to an agreed civilian need without creating unacceptable harm.
Proposed commitments
- Begin with one service, a defined disruption, and an existing operating baseline. Include the people who operate the service and the communities who rely on it.
- Evaluate people’s ability to receive understandable warnings and access essential services. Record where technical coverage or response capacity is uneven.
- Keep cross-sector coordination within an approved scope. Add sectors only after reviewing dependencies, authority, and the results of the earlier evaluation.
How to assess the commitment
Use service availability, recovery time, warning acknowledgment, and documented civilian harms as relevant to the specific use case. Set interpretation criteria before testing.
Trade-off to address
Focusing on a single service limits the initial scope. Expanding too quickly can hide the source of failures and create new interdependencies that are difficult to reverse.
05. Keep people accountable for consequential decisions
Every material decision should have an identifiable authority, a reviewable record, and a route for challenge.
AI output does not supply public authority. Participants need to know who may issue a warning, accept a risk, share information, or pause the arrangement. Responsibility cannot disappear between a model vendor, an operator, and an institution.
Proposed commitments
- Publish a decision-rights framework that distinguishes technical analysis, operational approval, legal review, public communication, and independent oversight.
- Record material decisions and their stated basis, including uncertainty, alternatives considered, and any override of the normal process.
- Limit emergency exceptions by purpose and duration. Require later review and an explicit decision before an exception becomes an ongoing practice.
How to assess the commitment
Check that consequential actions have documented authorization and that reviewers can reconstruct how an alert led to a decision. Test the challenge and correction process.
Trade-off to address
Records and review take staff time. A practical design needs to preserve timely action while ensuring urgent decisions remain accountable afterward.
06. Require evidence before expansion
Fund clear questions, controlled comparisons, and independent evaluation before making broad performance claims.
A plausible mechanism is an appropriate starting point for research. It is insufficient grounds for claiming that a system improves national security or prevents conflict. Each stage should answer a defined question and have a decision point.
Proposed commitments
- Predefine the baseline, outcomes, comparison conditions, harms to measure, and circumstances that would challenge the hypothesis.
- Compare the AI-supported approach with current practice and a simpler non-AI alternative where feasible. Keep the effect of governance changes distinguishable from the effect of software.
- Publish methods, limitations, and unfavorable findings at an appropriate level of detail. Require independent review before moving beyond the approved scope.
How to assess the commitment
Determine whether the agreed civilian outcome improves in a reproducible evaluation without exceeding the preset limits for cost, rights impacts, or operational risk.
Trade-off to address
A disciplined evidence standard can delay expansion and may conclude that no new AI system is needed. The program should be designed to accept that result.
07. Make costs and dependencies visible
Treat integration, verification, oversight, maintenance, and exit as part of the full cost of a proposal.
An inexpensive demonstration can lead to a costly operating commitment. Public decisions need a realistic account of recurring work, vendor dependence, training, independent review, and alternatives.
Proposed commitments
- Publish a cost model with assumptions and uncertainty before proposing a specific budget. Compare the full operating costs of feasible alternatives.
- Prefer interfaces and contract terms that permit migration, inspection, and replacement. Review concentration risk where multiple participants depend on the same outside provider.
- Release resources in bounded stages with explicit deliverables and renewal decisions. Tie further funding to reviewed results rather than the size or visibility of the demonstration.
How to assess the commitment
Track full lifecycle costs, time required to replace an outside service, portability test results, and the reasons for each funding decision.
Trade-off to address
Portability and independent review may increase initial expense. Low headline cost should not conceal continuing obligations or make cancellation impractical.
08. Earn public trust through reviewable work
Explain the public purpose, publish meaningful results, and protect the independence of oversight.
Trust should rest on procedures that people can inspect and challenge. Public reporting needs enough substance to evaluate the program while protecting personal information and genuinely sensitive service details.
Proposed commitments
- Maintain a public register of approved purposes, responsible institutions, evaluation stages, and significant changes. Clearly label proposals, observations, and findings.
- Give an independent reviewer access to the information needed to assess compliance and results through appropriate protected channels. Disclose conflicts of interest.
- Publish a plain-language account of material failures, corrective action, and renewal decisions. Provide accessible ways for affected people to raise concerns.
How to assess the commitment
Check whether reviewers can independently verify the evidence, whether significant findings receive a documented response, and whether public explanations match actual practice.
Trade-off to address
Complete public disclosure can expose people or essential services. The program needs a defensible boundary between public reporting and protected scrutiny, with oversight of that boundary.
All 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.