and industry standards such as IEEE Ethically Aligned Design and ISO/IEC 42001, have demonstrated structural

inadequacy in preventing ethical failures arising from AI-driven business decision-making. Responding to these

documented deficiencies, this paper proposes and evaluates an Integrated Ethical AI Governance Framework

(IEAGF) comprising four novel, complementary mechanisms: (1) Pre-Deployment AI Impact Assessments

(AIIA), which mandate bias auditing, fairness evaluation, and stakeholder impact mapping before system

deployment; (2) Explainable AI with Algorithmic Contracts (XAI-AC), which legally bind AI systems to defined

behavioural parameters and transparency obligations; (3) Ethics-by-Design (EbD) Frameworks, which embed

ethical principles, fairness constraints, and stakeholder inclusivity into AI development lifecycles; and (4)

Algorithmic Sustainability Indices (ASI), which introduce standardised metrics for quantifying the energy

consumption, socioeconomic impact, and renewable infrastructure usage of AI deployments. The IEAGF is

GDPR, ESG, ethical AI, algorithmic contracts, responsible AI deployment

The proliferation of AI-driven decision-making systems in commercial, financial, and public-sector contexts has

exposed a critical governance gap: the absence of preventive ethical infrastructure operating at the design and

pre-deployment stages of AI development. A companion analysis [1] establishes, through rigorous case-study

examination of Clearview AI, Facebook’s advertising algorithm, and Uber’s management system, that

documented harms from these systems were structurally enabled by reactive governance mechanisms that

activate only after deployment and documented harm have occurred.

This reactive-compliance paradigm—embodying the principle of “deploy first, regulate later”—is insufficient

for AI systems that can generate harm at scale before any human review is triggered. The GDPR’s right to

The IEAGF is structured as a lifecycle-spanning governance architecture operating across four functional stages:

pre-development, development, pre-deployment, and postdeployment monitoring. Fig. 1 presents the complete

framework architecture.

Fig. 1. Integrated Ethical AI Governance Framework (IEAGF) — four-component lifecycle architecture.

Environmental Impact Assessments (EIAs) have been a prerequisite for major infrastructure projects in most

jurisdictions for decades. The AI Impact Assessment (AIIA) applies this proven preventive governance logic to

AI systems, mandating structured evaluation of potential harms before deployment authorisation [2].

governance requirements. Feasibility challenges for small and medium enterprises (SMEs) include technical

capacity constraints and audit costs. The paper recommends government-subsidised AIIA support programmes

for SMEs operating in high-risk AI categories, analogous to existing regulatory compliance support schemes [3].

The AIIA directly addresses the EU AI Act’s limitation of not mandating pre-deployment bias audits for high-

risk systems. By formalising AIIA as a regulatory prerequisite, it converts the Act’s risk-classification framework

from a descriptive taxonomy to an actionable governance instrument [12].

GDPR Article 22 establishes the right of individuals to contest automated decisions but provides no mechanism

for enforcing meaningful explanation. Black-box models—particularly deep neural networks operating in

rights for regulatory bodies and affected individuals [5].

The XAI-AC framework requires that AI systems deployed in contexts with legal or financial consequences

prioritise interpretable model architectures (decision trees, generalised additive models, logistic regression)

where feasible, or provide post-hoc explanations (SHAP, LIME, ANCHOR) where deep learning is operationally

necessary. Model cards—structured documentation of model performance, limitations, and intended use—

become legally required rather than optional disclosures [6].

The prevailing compliance-first paradigm treats ethical requirements as constraints added to AI systems after

development, typically in response to regulatory requirements or public controversy [7]. Ethics-by-Design (EbD)

inverts this sequence, embedding ethical principles, fairness constraints, and stakeholder considerations into the

operationalises preventive governance as a structural requirement rather than an aspirational principle.

This approach is analogous to the shift in pharmaceutical regulation from post-market surveillance to pre-market

clinical trial requirements. The pharmaceutical analogy is instructive: AI systems that make consequential

decisions about human life outcomes (credit allocation, medical diagnosis, employment screening) warrant a

comparable standard of predeployment evidence.

The IEAGF faces three implementation challenges. First, its technical requirements demand expertise that many

regulatory bodies and SMEs currently lack. Second, AIIA certification processes introduce development timeline

extensions that may disadvantage regulated organisations relative to entities operating in laxer jurisdictions.

Third, algorithmic contracts require legal precedents that do not yet exist in most jurisdictions, necessitating

This paper has proposed, detailed, and evaluated the Integrated Ethical AI Governance Framework (IEAGF) as

a response to the structural inadequacies of existing AI governance identified in the companion analysis. The

four components—AIIA, XAI-AC, EbD, and ASI—collectively address regulatory fragmentation, algorithmic

bias, transparency deficits, and sustainability failures through preventive mechanisms that operate across the AI

development lifecycle.

Feasibility analysis confirms that each component is implementable within existing legal, technical, and

organisational infrastructures, with regulatory and professional standards support already in place for EbD and

XAI-AC. The IEAGF provides governments, regulatory bodies, and AI-deploying organisations with a practical,

evidence-based blueprint for governing AI systems in a manner that is technically rigorous, ethically