When May AI Decide?

Why This Research Matters

Artificial intelligence is no longer confined to back-office analysis.
It is beginning to shape decisions themselves. Large language models
can retrieve and synthesise rules, classify applications, draft opinions
and generate recommendations. Predictive systems can rank cases, flag
anomalies and forecast outcomes, while automated workflows can initiate
actions at a scale and speed that conventional administration cannot match.

For Islamic social finance institutions, these capabilities may improve
information access, operational efficiency, fraud detection, beneficiary
service and institutional learning.

Yet they expose a more fundamental governance question:
the ability to generate an answer does not, by itself,
confer the authority to decide.

Research Question

Contribution to the Literature

This paper contributes to the literature by developing a
Shariah–Maqasid decision-authority framework for determining when
artificial intelligence may assist, recommend, execute, or should be
restricted from consequential decisions in waqf, zakat and Islamic
social finance institutions.

It distinguishes technical capability from legitimate decision authority
and brings Shariah legitimacy, Maqasid reasoning, amanah, potential harm,
reversibility and meaningful human oversight into the governance of AI
delegation.

Rather than treating Islamic AI ethics primarily as a general collection
of principles, the paper translates these considerations into an
institutional decision logic that can inform governance practice and
provide a foundation for subsequent expert validation and empirical research.

Decision-Authority Logic

Before an institution determines how much authority may be delegated
to an AI-enabled system, five questions need to be considered together.
These factors progressively determine the appropriate level of AI
involvement in the decision.

The Five Decision-Authority Tests

1. Legitimacy of Purpose

The first question concerns why AI is being used. The institutional
purpose must be legitimate, consistent with Shariah and aligned with
the institution’s mandate. Technical efficiency cannot legitimise an
otherwise impermissible or institutionally inappropriate objective.

2. Nature of the Decision

Institutions should distinguish administrative, analytical, advisory,
fiduciary, distributive and explicitly religious decisions. The type
of decision affects the degree of authority that may legitimately be
assigned to an AI-supported process.

3. Degree of Normative Judgement

Some decisions depend substantially on factual processing; others
require contextual interpretation, ethical judgement, Shariah reasoning
or the balancing of competing interests. The greater the normative
content of a decision, the stronger the case for retaining qualified
human authority.

4. Harm, Scale and Reversibility

Potential harm must be considered alongside the number and vulnerability
of people affected and whether an erroneous decision can realistically
be reversed.

A reversible administrative recommendation does not carry the same
governance implications as an automated decision that excludes a
vulnerable beneficiary or irreversibly allocates entrusted assets.

5. Meaningful Human Authority

Human oversight cannot be reduced to a person formally approving an
AI-generated recommendation.

Meaningful authority requires an authorised and competent human actor
who has sufficient information, time and institutional power to question,
reject, override, escalate or stop the AI-supported process.

Shariah–Maqasid Governance of AI Decision Authority

The broader conceptual architecture begins with the Shariah–Maqasid
normative foundation and connects institutional context to the
decision-authority tests, permissible AI involvement, governance
safeguards, responsible outcomes and continuous institutional learning.

Application to Waqf

AI can support waqf administration in areas such as document retrieval,
asset information management, anomaly detection, forecasting, drafting
and operational analysis.

However, the governance implications change when a decision concerns
the interpretation of waqif conditions, alteration of endowed purposes,
beneficiary rights, preservation or substitution of assets, or other
matters requiring substantive Shariah, fiduciary or contextual judgement.

In such cases, computational assistance should not be confused with
the authority entrusted to trustees, boards, Shariah authorities and
other legitimate institutional actors.

Application to Zakat

Zakat institutions may use AI to improve administrative processes,
identify anomalies, analyse applications, support information retrieval,
forecast demand and assist authorised officers.

Greater caution is required where AI-supported systems influence
eligibility, beneficiary exclusion, prioritisation among vulnerable
applicants or determinations involving contested or context-sensitive
Shariah considerations.

Efficiency in processing cannot replace the institution’s responsibility
for the religious legitimacy, fairness and consequences of its decisions.

Application to Islamic Social Finance and Impact Decisions

The same decision-authority logic extends beyond waqf and zakat.
Islamic social finance institutions increasingly operate across
investment, development finance, philanthropy, impact assessment,
beneficiary services and blended social-finance structures.

AI may strengthen analysis, forecasting, monitoring and institutional
learning in these areas. Yet decisions affecting entrusted capital,
beneficiary interests, social priorities or Shariah-sensitive investment
choices continue to require clearly assigned human and institutional
accountability.

Meaningful Human Oversight

Responsibility remains with the people and institutions to whom it was
entrusted: trustees, boards, Shariah authorities, zakat administrators
and other authorised actors. Adding AI to the decision chain does not
dilute that answerability.

A nominal approval at the end of an automated process is therefore
insufficient.

Meaningful oversight exists only when a qualified and authorised person
has enough information, time, competence and institutional authority
to understand what the system is doing, challenge its output, intervene,
override it and, when necessary, stop the process altogether.

Decision Significance and Governance Intensity

Not every AI-supported activity requires the same level of governance.
Administrative and informational functions may permit comparatively
greater automation where consequences are limited and reversible.

As decisions become more consequential, more normatively complex,
more difficult to reverse, or more significant to vulnerable
beneficiaries, governance intensity should increase accordingly.

The central question is therefore not simply whether AI can perform
a function, but whether the institution can legitimately delegate
that degree of authority while preserving accountability.

Relationship to GF-AIRISF

This paper does not replace or modify the
Governance-First Artificial Intelligence Readiness for Islamic
Social Finance (GF-AIRISF) framework introduced in Research
Paper No. 01.

The two papers address different but complementary institutional
questions.

Together, the two perspectives connect institutional readiness with
the governance of actual AI decision authority.

Research Propositions

Implications for Institutional Leaders

The practical implication is not that Islamic social finance institutions
should avoid AI. It is that the authority granted to AI should be
proportionate to the nature and consequences of the decision.

Boards, trustees, Shariah committees and executive leaders should
therefore determine in advance which functions may be automated,
which require human approval, which require specialist or Shariah
review, and which decisions must remain substantively human.

Institutions also require documentation, escalation, appeal, audit,
monitoring and periodic reassessment so that human authority remains
meaningful throughout the AI lifecycle.

Future Research

The framework presented in this paper is conceptual. Future research
should therefore test and refine the decision-authority architecture
through expert validation, comparative institutional studies and
empirical application across different waqf, zakat and Islamic social
finance environments.

Further work may examine how decision classes vary across jurisdictions,
how Shariah governance bodies interpret acceptable levels of AI
delegation, how beneficiaries perceive AI-supported decisions, and
how meaningful human oversight can be assessed in practice.

Conclusion

AI can retrieve, compare, predict, classify and recommend, and in doing
so it can materially improve the information available to responsible
decision-makers.

What it cannot do is inherit the Amanah, Shariah authority or fiduciary
accountability entrusted to human beings and institutions.

The implication is straightforward: as the significance of a decision,
the need for normative judgement, beneficiary vulnerability, potential
harm and irreversibility increase, so too must the strength of human
authority and governance.

Suggested Citation

About the Research Series

The Global WaqfTech Executive AI Academy Research & Thought
Leadership Series advances applied scholarship at the
intersection of artificial intelligence, institutional governance,
Shariah, Maqasid, waqf, zakat and Islamic social finance.

The series supports the development of responsible institutional
approaches to emerging technology while creating a research foundation
for executive learning, policy dialogue and future empirical inquiry.