Building the ethical AI framework of the future: from philosophy to practice
New paper proposes enforceable 'metric, governance, and eco gates' to operationalize AI ethics from data to deployment.
A new research paper by Jasper Kyle Catapang tackles the critical gap between high-level AI ethics principles and enforceable, operational controls. The work, titled 'Building the ethical AI framework of the future: from philosophy to practice,' argues that current governance instruments like the EU AI Act and NIST AI Risk Management Framework provide necessary guidance but lack the specific, end-to-end mechanisms needed to prevent ethical risks from concentrating in AI pipelines, especially with the rise of multimodal and agentic systems.
The core innovation is an 'ethics-by-design' architecture that implements a 'triple-gate' structure at each stage of the AI lifecycle, from data collection to post-deployment monitoring. Each gate serves a distinct purpose: Metric gates enforce quantitative performance and safety thresholds; Governance gates ensure legal, rights, and procedural compliance; and Eco gates impose carbon, water, and sustainability budgets. The framework specifies measurable trigger conditions, escalation paths, and audit artifacts, enabling direct integration with existing MLOps and CI/CD pipelines.
To demonstrate its practicality, the paper includes illustrative examples from large language model (LLM) pipelines, showing how these gates can surface and constrain technical, social, and environmental risks before release and during runtime. Crucially, the framework is accompanied by a preregistered evaluation protocol that defines ex ante success criteria, making the effectiveness of the gates falsifiable and testable. By providing a concrete method to translate normative commitments into enforceable controls, the framework offers organizations a scalable, auditable basis for operational AI governance across different jurisdictions and deployment scales.
- Proposes a 'triple-gate' control architecture (Metric, Governance, Eco) for each AI lifecycle stage, moving beyond vague guidelines to enforceable checks.
- Designed to integrate with existing MLOps/CI/CD pipelines and maps controls to major regulations like the EU AI Act and the NIST RMF.
- Includes a preregistered evaluation protocol to make the framework's effectiveness testable and falsifiable, a key step for credible AI governance.
Why It Matters
Provides organizations with a concrete, auditable blueprint to operationalize AI ethics, turning philosophical principles into enforceable pipeline controls.