MAR + Preventive Diplomacy
Verified warnings could give diplomats a shared starting point for clarifying a developing crisis.
Untested hypothesisMAR +
Anti-Corruption & Rule of Law
Make responsibility visible. Transparent oversight could make it easier to establish who authorized an AI-supported alert or decision.
What is the proposal?
Transparent oversight could make it easier to establish who authorized an AI-supported alert or decision.
This is a possible mechanism to investigate. The available source does not contain a completed evaluation, an effect estimate, or evidence that the proposed benefit occurs in practice.
What decision does it raise?
Can oversight discover and remedy an unauthorized decision?
Map the authority for material alerts, the evidence retained, the reviewer’s access, and the route for concerns. Include an exercise in which a rule is bypassed and must be challenged.
Policy interpretation and scope are editorial proposals, not findings in the source matrix.
What could go wrong?
Original source downside
Formal rules can be bypassed without enforcement.
A pilot should account for this risk in its scope, comparison, and stop conditions. The absence of an observed failure in a limited exercise would not establish that the risk has been eliminated.
What safeguards would need testing?
Require enforceable review, documented authority, and a process for challenging decisions.
The responsible institutions would need to define who can authorize sharing, issue or correct a warning, approve a response, and suspend the activity. Safeguards should be assessed through observable behavior and an independent review process.
Proposed editorial safeguards. See the governance register for responsibilities and acceptance tests.
What would we measure?
Share of material alerts with documented authorization and audit.
A proposed evaluation
Measure the share of material alerts with documented authorization and audit, and record cases where rules were bypassed.
Before testing, define the setting, baseline, comparison condition, measurement period, appropriate data access, and interpretation criteria. Document the contribution of additional staff or better procedures separately from any claimed AI contribution.
How to avoid overstating the measure
A completed audit field is insufficient if the reviewer lacks independence or remedies are ineffective. Assess whether the process can correct conduct, not merely record it.
What would challenge this idea?
The proposal would be weakened if formal records exist but oversight cannot prevent or remedy unauthorized action.
An evaluation should preserve unfavorable, null, and inconclusive results. A revised claim should explain what changed and why; it should not quietly replace the original question after seeing the data.
Challenge conditions are proposed editorial criteria, not results from completed tests.
Original source record
The fields below preserve the wording of the supplied, undated matrix. Imported 5 November 2026. Original review status is unknown.
- Record ID
- MAR-LINK-019
- Combined with MAR
- Anti-Corruption & Rule of Law
- Predicted mechanism
- Transparent oversight may make AI alerts and decisions more accountable.
- Downside
- Formal rules can be bypassed without enforcement.
- Observable test
- Share of material alerts with documented authorization and audit.
- Source status
untested_hypothesis
Evidence status
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.
Related research
Continue the review.
MAR + Peace Agreements
Information about whether everyday services are working could help parties assess parts of a peace agreement.
Untested hypothesis