This note builds on the previously published white paper, "How Agentic AI can rebuild the Lebanese State" which outlined the transformative potential of autonomous digital agents in governance, service delivery, and legal processing. As a follow-up, it focuses on the legal voids and necessary reforms that must accompany the adoption of such systems within Lebanon’s civil law tradition.

Lebanon’s current legal framework, structured around human agency, codified norms, and institutional accountability, does not yet recognize or regulate non-human actors capable of autonomous reasoning and action. To responsibly integrate Agentic AI into public administration, legal reform must address foundational questions of liability, authority, oversight, and legitimacy.

The ten proposals below aim to establish a structured legal approach to Agentic AI in Lebanon. They draw on European legal models, particularly from France, Germany, and the EU AI Act, while remaining grounded in the principles of Lebanon’s Romano-Germanic legal tradition. Four use cases follow related to public procurement, tax administration, urban planning, and border control.

Ten proposals for responsibly integrating Agentic AI in Lebanon’s public administration

1. Legal Personality: Lebanese law does not recognize AI as a legal person. Proposal: Create a new classification, "Electronic Legal Agents", granting limited legal capacity for predefined tasks under strict human oversight.

2. Civil Liability: The law does not cover AI-caused harm. Proposal: Adopt a joint liability attributing responsibility to developers, operators, and deploying authorities depending on the root causes of the harm and request mandatory insurance.

3. Contracting Capacity: AI cannot contract under Lebanese law due to lack of intent/capacity. Proposal: Recognize AI as “electronic mandataries” acting under mandates from legally responsible entities.

4. Criminal Liability: AI lacks criminal intent; cannot be held criminally liable. Proposal: Expand Article 210 of the Penal Code to assign indirect liability to negligent humans or institutions using AI.

5. Administrative Authority: AI cannot issue binding administrative decisions. Proposal: Limit use to preparatory tasks (e.g., document verification), with mandatory human validation of final acts.

6. Source of Legal Reasoning: Codified norms are the primary source in Civil law despite the presence of other non-binding sources. AI should rely only on laws and official regulations, excluding jurisprudence and doctrine, as primary inputs to insure objective reasoning.

7. Human Oversight: Ensure human-in-the-loop validation for all legally relevant AI outputs. Amend Law 81/2018 to codify this safeguard for AI-assisted interpretation or action.

8. Accountability Mechanisms: Require public registries for Agentic AI and assign traceable human data responsibility for AI actions across development, deployment, and decision-making.

9. Risk Mitigation: Implement auditability, transparency, and reversibility requirements for AI actions, especially in public services, to protect against misuse and error.

10. Comparative Harmonization: Use frameworks from jurisdictions that have advanced in AI regulation including the EU (e.g. AI Act, Liability Directive) and civil law doctrines in countries like France & Germany as reference models for legal reform in Lebanon.

Download the white paper