Executive Summary

Artificial Intelligence is entering government through digitisation, procurement, and service delivery modernisation. But a new frontier is now emerging: agentic AI — systems capable not only of predicting or recommending, but planning, acting, and executing tasks with a degree of autonomy.

This evolution changes the governance question. It is no longer only about ethical use or responsible innovation. It is about a more foundational issue:

How does the state remain accountable under the rule of law when decision-making and administrative action are partially delegated to automated agents?

This paper argues that agentic AI introduces a diffusion of agency — where responsibility becomes blurred across institutions, vendors, data sources, and automated workflows. In such an environment, the rule of law cannot be assumed. It must be designed and enforced through institutional readiness.

1. The Governance Moment Has Shifted

Across Africa and globally, public sector AI discussions have matured quickly. Early attention focused on innovation, ethical principles, and high-level strategies. These discussions served a necessary purpose.

However, the field is entering a new phase — one defined less by intention and more by execution. AI increasingly arrives as administrative infrastructure: eligibility systems, registries, compliance tools, predictive allocation models, and vendor-managed “smart government” platforms.

Agentic AI accelerates this shift by introducing systems that interpret objectives, decide on actions, interact with digital environments, and execute multi-stage processes with reduced human instruction.

2. What Is Agentic AI — in Practical Governance Terms?

For public institutions, agentic AI is best understood through a rule-of-law lens. An agentic system is one that can take action, not merely provide information.

Rule of law stress arises at the point of action: when benefits are approved or denied, enforcement is triggered, records are updated, or services are withheld.

3. The Rule of Law Problem: Diffusion of Agency

Under the rule of law, public authority must be exercised through lawful mandates, accountable officials, and reviewable processes.

Agentic AI fragments this structure. Responsibility disperses across ministries, vendors, datasets, workflows, oversight units, and automated agents themselves.

When responsibility is diffused, accountability evaporates.

4. Rule of Law Tests for Agentic AI in Government

• Legality: What lawful authority permits the action?
• Accountability: Which named official is responsible?
• Reasons-Giving: Can the institution explain the decision?
• Contestability and Remedy: Can a citizen challenge it in practice?
• Auditability: Are records inspectable by oversight bodies?

If these tests cannot be satisfied, deployment is not merely risky. It is institutionally unsafe under a rule-of-law standard.

5. Why Ethical Principles Alone Will Not Protect Citizens

Ethical commitments matter, but values without operational controls remain aspirational. Institutional failure arises from missing enforcement mechanisms, audit readiness, procurement safeguards, and accountability chains.

6. Minimum Viable Rule-of-Law Governance for Agentic AI

• Named accountable official
• Human oversight protocols
• Mandatory audit trails
• Accessible contestation pathways
• Governance-aware procurement clauses
• Incident response and shutdown procedures

7. What This Means for Zimbabwe and Similar Contexts

National strategies alone are insufficient. Agentic AI enters unevenly, often through donor-funded or vendor-led projects embedded in sector ministries. Governance readiness must therefore exist at institutional level.

8. Conclusion: Modernising Without Losing the State’s Duty

Automation can increase capacity, but it can also weaken accountability if authority is delegated without controls.

The rule of law is not a ceremonial ideal. It is the operational backbone of legitimate governance.

---

AGCIH Closing Note
The Africa Governance & Civic Innovation Hub (AGCIH) supports governments and public institutions to move from digital and AI ambition to responsible implementation through governance readiness, institutional design, risk management, and public trust architecture. AGCIH does not advocate, regulate, or deploy technology. AGCIH operationalises governance.