Research in public.

Digital Twin Ethics and Governance Framework (DTEGF)

Abdulmotaleb El Saddik · Roberto Alejandro Martinez Velazquez

Digital twins are evolving from models of physical assets into intelligent representations of identifiable people that can analyze, predict, communicate, and act on a person’s behalf in digital and physical environments. This shift raises governance questions that general AI ethics instruments do not fully operationalize: who may authorize a twin to act, how faithfully it must represent the person, how it should be governed as it drifts from the person over time, and what should happen on incapacity or death. This document proposes the Digital Twin Ethics and Governance Framework (DTEGF). DTEGF rests on nine ethical pillars and a three-tier governance architecture in which safeguards become harder to override as the consequences of overriding them increase, with non-overridable protections backed by technical or organizational controls the twin cannot remove. It classifies twins along two largely independent dimensions: four risk levels, from private representational twins to high-impact ones, and four embodiment classes, from software-only twins to those capable of hazardous physical action. The pillars are translated into 30 governance requirements and six embodiment requirements, linked through an applicability matrix and pillar traceability. Its distinctive contributions are person-twin separation, authority minimization, representation by explicit delegation, safeguards beyond the model, and lifecycle governance from creation to post-mortem. DTEGF is designed to be interoperable with the UNESCO Recommendation on the Ethics of Artificial Intelligence, the OECD AI Principles, the EU AI Act, and Saudi Arabia’s SDAIA AI Ethics Principles and Personal Data Protection Law.

Risk-Based AI Assurance in the Saudi Public Sector

Aligning International Assurance Practices with the National AI Risk Management Framework

Nawaf Alsahmah, Ph.D.

Saudi Arabia has established a national foundation for managing artificial intelligence (AI) risk through the National AI Risk Management Framework, issued by the Saudi Data and Artificial Intelligence Authority (SDAIA) in 2026. The framework structures risk management around context and scope, risk identification, risk assessment, risk treatment, and monitoring and review. The policy challenge addressed in this paper is therefore not the absence of AI risk governance, but how outputs from this established process can be connected to a distinct assurance process that supports justified confidence that applicable requirements have been satisfied.

Protecting Artistic and Artisanal Creativity from Generative AI in Developing Countries

A Legal Readiness Framework Based on the Moroccan Case and the "Digital Maâlem" Model

Pr. Hanane Nahid

Generative models trained on global corpora reproduce the decorative and musical vocabulary of developing countries without consent, attribution or payment, and the artisans who authored those traditions bear the commercial consequence. This paper develops legal readiness as a qualitative conceptual and policy framework for assessing whether a national legal system can respond. The framework structures analysis across four policy axes: the protection of human authorship, training data governance, the safeguarding of folklore and traditional cultural expressions, and judicial capacity. It assigns no indicators, weights or scores to any jurisdiction. The framework extends UNESCO's Readiness Assessment Methodology (RAM). Morocco has completed a RAM, but its published legal findings concern personal data, the regulation of cyberspace, liability and public procurement, not cultural intellectual property. No readiness instrument currently asks who may license a traditional cultural expression for machine training, what provenance obligations attach to heritage material in a training corpus, or whether a court can establish that a protected work entered a model. That is the gap this framework addresses. Morocco is examined as an illustrative national case rather than a statistically representative sample of developing countries. Applying the four axes to Moroccan law returns a mixed result. Axis 3 is strong: Article 7 of Law No. 02-00 protects expressions of folklore against reproduction, communication to the public, adaptation and fixation for commercial purposes, vests the authorisation right in the Moroccan Copyright Office, requires that the community or locality of origin be named, and earmarks the sums collected for professional purposes and cultural development. Axis 4 is partial: Law No. 013.26 of 2026 empowers authorised officials to inspect the information systems of users of protected works, which is an evidentiary tool of direct relevance, though its reach to a training corpus held abroad is untested. Axis 2 is indirect. Axis 1 shows no readiness on the express question, since no Moroccan instrument distinguishes AI assisted from AI generated work, and the human authorship requirement that Law No. 013.26 preserves resolves the matter only by implication. The paper then proposes the Digital Maâlem model, under which AI systems deployed in the artisanal and creative sectors operate under the responsibility of a certified master artisan, paired with a professional digital twin through which an artisan registers technique and draws a statutory remuneration when it is exploited commercially. Both components are anchored on institutions that already exist: the authorisation right of the Copyright Office under Article 7, and the artisan registers maintained by the regional chambers of crafts. Questions of consent, inheritance, attribution of diffuse communal style and cross border enforcement are set out as the conditions the model must satisfy before implementation, and the objections of state custodianship and cultural access are addressed directly. The conclusions are offered as foundational hypotheses and policy pathways for broader comparative research.

A Dignity Safety Layer for Autonomous AI

Dr. Mazhar Ali Bari

AI systems increasingly build detailed working models of the people they interact with. Some of what these models hold is straightforward: declared preferences, practical needs. But they can also carry inferred vulnerability, financial distress, likely reactions, susceptibility to persuasion, sensitive characteristics the person never chose to disclose. That creates a safety problem you cannot see by checking the final action alone. How the system got from knowledge to action matters too. This brief presents the Dignity Safety Layer (DSL), a proposed runtime governance architecture built around a simple principle: just because a system can know or predict something about a person does not mean it is permitted to use it. The DSL does not attempt to compute human dignity or replace human judgment. Instead, it asks whether particular uses of person-specific information are authorized for the relevant purpose, under the relevant relationship and over the history of interaction. In practice, it checks how person-specific information is being used, for whose purpose and under what authority, within what relationship and history, how confident the system really is, and whether the person can contest what it thinks it knows. Each check ends in one of four outcomes: ALLOW, REASSESS, ESCALATE or BLOCK. Synthetic experiments illustrate how the architecture behaves. They test the mechanism: they do not show that dignity has been measured or guaranteed.

The Calendar Gap

Mechanism-Level Diagnosis of Arabic Agentic Failures

Abdullah Almohammedi

Aggregate benchmarks report that LLM agents perform worse in Arabic than English; they do not say why. We introduce a mechanism-level diagnostic built on matched task variants: instances byte-identical in tools, semantics and phrasing but for a single toggled property. The pilot isolates Hijri calendar reasoning, entity transliteration and serialization discipline across the language boundary, with an Eastern-numerals control. Five arms are tested: four open-weight (20B to 397B), including a hybrid pair holding identical weights with reasoning toggled, plus one closed lite-tier model. Gregorian dates in Arabic pass at 0.90–1.00 on every arm, while the matched Hijri variants collapse to 0.00–0.20. Across arms, 45 of 50 matched set pairs flip against Hijri and none the other way; the isolated Hijri delta is 0.80–1.00. Two arms survive Holm correction in the pre-specified 25-test primary family (adjusted p = 0.0488 each). A deterministic heuristic finds an attempted conversion in every failure but one, where qwen3.5-397b asks for clarification. Mean absolute error among failed dated records ranges from 145.1 to 20.6 days, never reaching exactness. Transliteration failures show a different, contract-like signature. Scoring is deterministic, calendar golds derive mechanically from the Umm al-Qura calendar, and hypotheses and thresholds were frozen before any model call. The scorer audit remains below its own 0.95 gate (0.9070), with the author as one annotator. The primary Hijri dates occupy 1448 AH, a near-future window; a separately frozen control repeats the same ten M2 sets in 1440 AH, well before the arms' likely training cutoffs. The matched penalty persists on all five arms at 0.50–1.00. The control protocol did not pre-specify a separate five-test Holm family, so its multiplicity analysis is post-hoc sensitivity only, not confirmatory significance. This weakens a recency-only or cutoff-only account, while heterogeneous movement across arms leaves temporal exposure a possible moderator rather than a settled mechanism.

From Awareness to Responsible Application

A Community AI Readiness Framework Linking Assessment to Needs-Based Capacity Building

Dr. Reem K. Alshammari · Abdullah G. Alanazi

This paper proposes an AI readiness framework that assesses residents of governorates and local communities and links the results directly to needs-based learning, training and access interventions. Individuals are assessed across three progressive stages: Awareness & Exposure, Understand, and Apply, with responsible AI treated as a cross-cutting layer present at every stage rather than a terminal level, and with contextual enablers and barriers reported separately from the readiness stages.