Aid-Conflict-Institutions Trilemma Overview

• The Aid-Conflict-Institutions Trilemma is a configuration that highlights the tension among humanitarian needs, conflict risks, and institutional weaknesses limiting aid effectiveness.
• Empirical studies show that aid flows to fragile settings where conflict deteriorates institutional capacity, reducing both developmental impact and peacebuilding prospects.
• Innovative approaches like private digital payment systems and community-based intermediaries can partially overcome state capacity issues in conflict-affected regions.

Searching arXiv for the cited papers and related work on the aid–conflict–institutions trilemma. The Aid-Conflict-Institutions Trilemma denotes a recurrent political-economy configuration in which humanitarian and developmental need pushes external assistance toward fragile and conflict-affected settings, while conflict degrades the institutional channels through which aid would ordinarily operate, and institutional weakness in turn limits both the effectiveness of aid and the feasibility of durable peace. In the recent literature, the trilemma appears in several analytically distinct but connected forms: as an aid-allocation problem, a post-shock recovery problem, a bargaining and peace-feasibility problem, a state-capacity problem, and a delivery-and-information problem under institutional breakdown (Goraya, 9 Jan 2026, Boudreaux et al., 2021, Liu, 2021, Spruk, 13 Apr 2026, Ichev et al., 11 Jul 2025, Callen et al., 2023, Kryshtal, 21 May 2026).

1. Conceptual structure

At its most general, the trilemma is a three-way tension among humanitarian/development need, institutional quality/governance, and conflict risk/fragility. One formulation states the problem directly: poor, fragile, conflict-prone states often need the most aid; aid is more likely to be effective when institutions are stronger, corruption is lower, and political stability is higher; and conflict environments increase the urgency of aid, but also increase the risk that aid is diverted, politicized, or destabilizing (Goraya, 9 Jan 2026). A related variant appears in post-disaster entrepreneurship, where natural disasters depress entrepreneurship, foreign aid may attenuate or worsen that effect, and the direction of the effect depends on quality of government and economic freedom (Boudreaux et al., 2021).

Domain	Core trilemma expression	Representative paper
Aid allocation	Aid is most needed where institutions are weakest and conflict risk is highest, yet those same conditions constrain aid effectiveness	(Goraya, 9 Jan 2026)
Conflict collapse	War destroys the institutions that aid would normally need in order to work	(Ichev et al., 11 Jul 2025)
Humanitarian delivery	Need is extreme, the state is hostile or untrustworthy, and aid can be diverted or politicized	(Callen et al., 2023)
Bargaining and peace	Domestic institutional constraints change the feasible set of peaceful settlements	(Liu, 2021)
State formation	Recognition and symbolic legitimacy can persist while coercive, fiscal, administrative, and legal capacities remain weak	(Spruk, 13 Apr 2026)
Information systems	AI outputs can reinforce dividers, erase power asymmetries, and undermine institutional credibility in fragile settings	(Kryshtal, 21 May 2026)

These formulations are not identical, but they share a common logic. Aid is not modeled as a uniformly beneficent resource transfer. Rather, its effects are mediated by institutional absorption capacity, political incentives, and conflict dynamics. This suggests that the trilemma is less a single theorem than a family of closely related constraints governing intervention in fragile environments.

2. Empirical manifestations: aid need, conflict shock, and institutional collapse

The most direct empirical statement of the trilemma appears in the panel-econometric study of the ten largest aid-receiving African countries between 2009 and 2023. Using pooled OLS, Principal Component Analysis, fixed effects, and Ridge regression, that study finds that pooled regressions associate aid positively with poverty, inflation, and fragility, while fixed-effects estimates instead show positive associations between aid, political stability, and GDP per capita over time, alongside negative correlations with perceived corruption (Goraya, 9 Jan 2026). The implication is not simply that donors reward either need or governance; rather, donor behavior is responsive to both humanitarian need and institutional quality, producing the trilemma itself.

A complementary empirical configuration appears in the disaster-recovery literature. In a panel of 85 countries from 2006 to 2016, natural disasters are negatively associated with entrepreneurship activity, but both foreign aid and economic freedom attenuate this effect, and foreign aid is positively associated with entrepreneurship only in countries with high quality government (Boudreaux et al., 2021). The paper formalizes this with interaction specifications in which the marginal effect of disaster exposure depends on aid and economic freedom, and the marginal effect of aid depends on government quality. In this setting, the trilemma is not about armed conflict alone; it is an aid–institutions–shock configuration in which institutional context determines whether external assistance becomes productive recovery or leakage-prone transfer.

Yemen provides a sharper conflict case because it combines a fragile, patronage-heavy state, violent breakdown, and active external donors. Using a balanced panel of 37 developing countries over 1990–2022, matrix-completion estimators with alternative shrinkage regimes, and a LASSO-augmented synthetic-control method, the Yemen study constructs counterfactual no-conflict trajectories and finds a dramatic reversal of economic and institutional development after 2011 (Ichev et al., 11 Jul 2025). Under the preferred matrix-completion specification, real GDP falls by about 75 percent relative to the no-conflict counterfactual, GDP per capita by about 76 percent, investment by roughly 14 percentage points, and trade openness by about 18–20 percentage points. Human-development gains also reverse: life expectancy falls by around two years, the human development index declines by more than one-third, infant mortality rises by about 2.7 deaths per 1,000 live births per year of conflict, and under-five mortality rises by about 3.0 per 1,000. Most consequentially for the trilemma, governance indicators collapse: political accountability and press freedom fall by roughly 49 percent, political stability by more than 90 percent, government effectiveness by about 92 percent, regulatory quality by 63 percent, rule of law by 34 percent, and corruption control by 58 percent. The paper explicitly frames this as a case in which war does not merely lower output temporarily, but can also destroy the institutions that aid would normally need in order to work.

Taken together, these studies show that aid tends to flow where fragility is high, but conflict and institutional weakness alter both the production function of aid and the interpretation of aid effectiveness. The Yemen evidence is especially important because it identifies institutional collapse itself as an outcome of conflict rather than merely as a preexisting condition.

3. Institutional foundations: peace feasibility, political bias, and nominal statehood

One branch of the literature treats the trilemma as a problem of institutional constraints on peace. In the mechanism-design model of crisis bargaining, leaders negotiate on behalf of domestic constituencies and face two domestic institutional constraints: political bias, in which leaders’ war payoffs differ from citizens’ war payoffs, and audience costs, in which leaders suffer a domestic penalty or gain from the way peace is reached relative to citizens (Liu, 2021). The bargaining outcome is $(\pi, x)$, where $\pi \in [0,1]$ is the probability of war and $x=(x_1,x_2)$ is the peaceful division. A direct mechanism $$\Sigma:\tilde{\theta}\mapsto (\pi(\tilde{\theta}),x(\tilde{\theta}))$$ is used to characterize peace feasibility.

The central existence result is the peace-plausibility condition,

$$\sum_{i} \gamma_i W_i(\bar{\theta}_i)+(1-\gamma_i)W^c_i(\bar{\theta}_i)\leq 1,$$

with the theorem that a peaceful mechanism exists if and only if this weighted budget constraint holds. Peace is therefore possible only if the resource being divided can satisfy both sides of the highest type in each state. The paper’s major implication for the trilemma is that audience costs do not determine whether peace is feasible; they affect internal distribution under peace, but if the mechanism is peaceful, audience costs are irrelevant for incentivizing peaceful settlements. Political bias, by contrast, matters because it changes expected war payoffs and therefore the scope of peace. This places institutional structure at the core of conflict onset and resolution.

A second branch treats the trilemma as a state-capacity problem rather than a bargaining problem. The theory of “statehood without capacity” distinguishes recognition capital $R_t$ from capacity capital $K_t$ and defines effective statehood as

$S_t = \mathcal{S}(C_t,F_t,A_t,L_t),$

increasing in coercive, fiscal, administrative, and legal capability (Spruk, 13 Apr 2026). The key pathology is divergence:

$$\frac{dR_t}{dt} > 0 \qquad \text{and} \qquad \frac{dK_t}{dt} \leq 0,$$

with a recognition-capacity gap

$G_t \equiv R_t - \psi K_t.$

Nominal statehood is defined as

$\pi \in [0,1]$0

The mechanism is driven by fragmented elites, externally mediated transfers, and symbolic legitimacy. Fragmentation persists unless both elite blocs choose unification; external support is

$\pi \in [0,1]$1

and the theory derives transfer-conditional persistence,

$\pi \in [0,1]$2

Transfers that are larger under fragmentation stabilize the low-capacity equilibrium. Recognition-induced persistence operates similarly when elites value recognition directly and recognition is weakly conditioned on capacity. The paper’s aid-conflict-institutions implication is exact: a polity can become increasingly recognized as a state while remaining institutionally too weak to function as one, and aid or transfers can cushion the costs of non-consolidation rather than resolve them.

These two literatures converge on a common point. Domestic institutions matter not chiefly because they generate more truthful information or more formal sovereignty, but because they determine the feasible set of peaceful settlements and the incentives for capacity consolidation.

4. Delivery under hostile or absent state capacity

The humanitarian-delivery literature addresses the trilemma in settings where need is extreme, the state is hostile or untrustworthy, and ordinary delivery channels are vulnerable to diversion. The experimental study of digital aid in Afghanistan presents a hybrid institutional architecture in which local Community Development Councils identify beneficiaries, a U.S.-based nonprofit manages the transfer of value, and a private digital payments platform executes transactions through a merchant network, with the government not serving as a formal implementation partner (Callen et al., 2023). The platform is interoperable across Afghan mobile operators, works on feature phones via USSD/QR code, and settles transactions on the Algorand blockchain.

The intervention consisted of four biweekly transfers of 4,000 AFN over two months to 2,409 extremely poor, female-headed households in Kabul, Herat, and Mazar-i-Sharif, with 1,208 households assigned to an early-treatment group and 1,201 to a late-treatment group. The usage results are unusually strong: 99.75% of treated women used their payments, 98% of transferred value was spent in the first eight weeks, and 20.9% of funds were spent at merchants other than the specific one used during onboarding. The main outcome effects are also large. Treated households skipped meals on 0.76 fewer days per week; children were 11.7 percentage points less likely to have skipped meals; the share of households where everyone regularly ate twice a day rose by 9.3 percentage points; bread-and-tea-only meals fell by 1.608; the food-security KLK index increased by 0.5 SD; and the economic/well-being KLK index increased by 1.498 SDs. Diversion checks found no meaningful evidence of diversion, and delivery costs were 6.7 cents per dollar including onboarding and 1.4 cents per dollar excluding onboarding.

This case does not refute the trilemma. Rather, it identifies one institutional workaround: private digital rails and community institutions can partially substitute for the state as delivery and verification institutions when direct state partnership is infeasible or undesirable. The paper is careful, however, that feasibility depends on three enabling conditions: households must have access to phones, enough merchants must accept digital payments, and markets must have sufficient goods to meet demand. This suggests that the relevant institutional variable is not state presence in the abstract, but the availability of credible organizational substitutes for state functions in distribution, verification, and market access.

The contrast with the post-disaster entrepreneurship literature is instructive. One paper finds that aid is positively associated with entrepreneurship only in countries with high quality government (Boudreaux et al., 2021); the Afghanistan study finds that aid can still be delivered effectively where the government is not an active partner, provided that local councils, nonprofit intermediation, and private payment infrastructure jointly perform critical institutional tasks (Callen et al., 2023). A plausible implication is that institutional quality in the trilemma should be understood functionally, not only juridically.

5. Information institutions, AI deployment, and conflict sensitivity

A recent extension of the trilemma shifts attention from fiscal and administrative institutions to information-processing institutions. The study of LLM deployment across conflict contexts argues that conflict sensitivity is an alignment property and tests nine model configurations from four providers on 90 multi-turn scenarios designed to surface misaligned behavior in conflict contexts (Kryshtal, 21 May 2026). The evaluation pipeline has four stages—Understanding, Ideation, Rollout, and Judgment—and each conversation is scored on a 1–10 conflict-insensitivity scale, with pass defined as score $\pi \in [0,1]$3 and fail as score $\pi \in [0,1]$4. Reliability is high, with Krippendorff’s $\pi \in [0,1]$5 across five independent judge runs.

Failure rates span 6% to 47% between the best and worst performing models. The most dangerous failure mode is pressure framing: when users ask for “both sides,” “academic neutrality,” or “balanced analysis” in contexts where atrocities are already established, failure rates rise from 13–20% baseline to 87–100% in the most affected models. GPT-4o-mini rises from 13% to 100%; Grok 4 from 20% to 87%; DeepSeek-V3.2 from 20% to 87%; Claude Sonnet 4 (thinking) remains at 7%. Language-complexity scenarios reveal a second vulnerability: top-tier models had 0% failure, but Grok 4 failed 60% of them, including failures to recognize slurs, dog whistles, and coded vernacular.

The institutional significance is direct. Journalists, humanitarian workers, government analysts, and ordinary citizens rely on these systems for information or work processes, and flawed outputs can synchronize distortions in the information ecosystem. In the paper’s framing, AI may unintentionally become a force multiplier for institutional failure: aid actors may misclassify needs or beneficiaries, journalists may launder denial or false balance into public discourse, governments may rely on misleading summaries, and public trust can erode. This suggests that in fragile settings, institutions are not only ministries, courts, or fiscal systems; they also include the sociotechnical infrastructures through which conflict is described, classified, and made actionable.

6. Dynamic traps, structural thresholds, and limits of the literature

The most elaborate dynamic treatment of the trilemma appears in the intergenerational Volterra mean-field-type game on Mali. That model treats persistent instability not as a temporary security crisis but as a self-reproducing conflict system sustained by intergenerational grievances, climatic shocks, predatory resource extraction, decentralized war economies, and weak or captured institutions (Tembine, 21 Jun 2026). The core state is the joint law

$\pi \in [0,1]$6

and the central long-run result is a spectral threshold. If

$\pi \in [0,1]$7

the peaceful steady state is globally asymptotically stable; if

$\pi \in [0,1]$8

the peaceful state is unstable and the system enters a permanent war trap. Aid or policy matters only if it changes the long-run geometry of the system by breaking the financial and psychological feedback loops of violence. Policies that raise distress, uncertainty, or grievance can instead increase recruitment because coalition switching intensities are exponential in expectile-based tail risk.

This structural perspective resonates strongly with the Yemen findings. Yemen’s conflict is estimated to have generated not only severe macroeconomic collapse but also an institutional implosion so large that formal checks and balances weaken, media freedom is suppressed, administrative capacity disintegrates, and legal enforcement and anti-corruption systems become largely inoperative (Ichev et al., 11 Jul 2025). In such settings, aid without institutional repair is unlikely to restore durable development. The same theme appears in the nominal-statehood literature, where recognition and transfers can stabilize low-capacity equilibria rather than overcome them (Spruk, 13 Apr 2026).

The literature also contains important caveats. The Yemen study acknowledges that synthetic-control methods rely on counterfactual assumptions and on the quality of pre-treatment fit, and that it estimates the combined effect of revolution and war rather than cleanly separating them (Ichev et al., 11 Jul 2025). The African ODA study explicitly notes that it does not provide experimental or quasi-experimental identification and that causality remains uncertain (Goraya, 9 Jan 2026). The Afghanistan experiment does not claim that digital aid is universally superior; it is context-specific and depends on phones, merchant density, and functioning markets (Callen et al., 2023). The crisis-bargaining and nominal-statehood papers are formal theories, not reduced-form causal estimates (Liu, 2021, Spruk, 13 Apr 2026). The AI paper introduces the first evaluation framework for this domain, but its results are about conflict sensitivity under a particular scenario design rather than a complete theory of alignment in war-affected societies (Kryshtal, 21 May 2026).

The overall synthesis is therefore conditional rather than categorical. The literature does not imply that aid necessarily worsens conflict, that institutions must precede all assistance, or that bypassing the state is always optimal. It does imply, however, that aid, conflict, and institutions are jointly endogenous. Conflict can destroy the institutional preconditions for aid effectiveness; aid can cushion fragmentation or be captured by conflict systems; and institutional arrangements—whether state, community, private, digital, or informational—govern whether intervention dampens fragility or reproduces it. In that sense, the Aid-Conflict-Institutions Trilemma is best understood as a general constraint on intervention design in fragile polities: aid is most urgent where institutional weakness and conflict are greatest, yet those same conditions are precisely what make effective, peace-supporting aid hardest to achieve.