Multi-Agent Risks from Advanced AI
This technical report by the Cooperative AI Foundation offers the comprehensive early attempt to map the risks that emerge when multiple advanced AI agents interact, adapt, and evolve together.
AI Governance Library: all resources
Every reviewed AI governance resource in one place — reports, standards, guidance and regulation.
Lean meets Data & Generative AI
Lean meets Data & Generative AILean meets Data & Generative AI.pdf797 KBdownload-circle What’s Covered? The paper is built around a simple but often neglected insight: technology only
Putting Explainable AI to the Test
This resource takes a close look at one of the most cited — and least consistently defined — goals in responsible AI: explainability.
Temporal Quality Degradation in AI Models
This resource breaks new ground by tackling a blind spot in model lifecycle management: the phenomenon of “AI aging.” The authors propose that temporal degradation is distinct from known issues like concept drift.
Introduction to AI Safety, Ethics, and Society
This is the book you’d hand to someone serious about understanding AI risk but unsure where to start. With clarity and precision, it lays out how AI could cause major harm—through misalignment, misuse, or sheer scale—and what we can do about it.
NIST Guidelines for Evaluating Differential Privacy Guarantees
Aimed at helping technical and policy audiences evaluate privacy guarantees in practice, NIST SP 800-226 offers tools to reason about parameters, algorithms, and trust assumptions in differentially private systems.
US Open-Source AI Governance
This report lays out a practical framework for evaluating US open-source AI policy through both ideological and geopolitical lenses. It avoids hype and polarization.
Human Oversight under Article 14 of the EU Al Act
This research offers a crisp, nuanced breakdown of what Article 14 AI Act really demands from human oversight—moving beyond vague calls for “humans in the loop.” It highlights the challenges of effectiveness, the shared roles of providers and deployers, and why human oversight is no silver bullet.
The Model Trust Score: The Framework for Strategic Enterprise AI Model Selection
A practical framework by Credo AI that helps enterprises filter and evaluate foundation models using context-specific trust scores. It introduces “Model Trust Scores” to guide business-informed decisions about AI adoption across capabilities, safety, cost, and latency.
Looking Ahead: Synergies Between the EU AI Office and UK AISI
This policy brief outlines a structured approach for collaboration between the EU AI Office and the UK AI Safety Institute (AISI), proposing a practical framework based on four types of institutional engagement: collaboration, coordination, communication, and separation.
A Competency Framework for AI Integration in India
India’s new AI Competency Framework equips public sector leaders with the behavioural, functional, and domain-specific skills to responsibly integrate AI in governance. Anchored in the IndiaAI Mission, it marks a major step toward building ethical, capable leadership for AI-driven transformation.
THE IMPACT OF HUMAN-AI INTERACTION ON DISCRIMINATION
This 2025 report from the European Commission’s Joint Research Centre shows that human oversight isn’t a silver bullet against discrimination in AI-aided decisions. In hiring and lending experiments, human biases often reinforced, revealing serious gaps in current oversight assumptions.
AI Explainability in Practice
“AI Explainability in Practice” by The Alan Turing Institute offers a practical, activity-based approach to embedding explainability into public sector AI projects. It’s part of a broader ethics workbook series supporting responsible, transparent, and accountable use of AI in government settings.
Final Report of the Expert Group on B2B data sharing and cloud computing contracts
A practical blueprint for fairer, more balanced B2B data-sharing and cloud contracts in the EU. The Expert Group distills key concerns into actionable guidance aimed at leveling the playing field between powerful cloud providers and smaller business users.
AI Auditing Checklist for AI Auditing
This checklist is a practical tool for conducting end-to-end socio-technical audits of AI systems. Commissioned by the EDPB, it provides a hands-on methodology for identifying risk, bias, and compliance gaps across the AI lifecycle—from data handling to deployment.
Privacy-Enhancing and Privacy Preserving Technologies in AI
This report from CIPL explores how privacy-enhancing technologies (PETs) can support responsible AI development. It walks through real-world use cases, technical trade-offs, and recommendations for deployers, policymakers, and regulators aiming to balance data utility with individual privacy.
NIST: Adversarial Machine Learning A Taxonomy and Terminology of Attacks and Mitigations
Adversarial ML is no longer just a research topic—it’s a real-world problem. This NIST report gives public and private sector orgs a shared vocabulary to talk about attacks and defenses. Perfect for aligning security, AI, and policy teams around common ground.
European AI Standards Technical Standardization and Implementation Challenges under the EU AI Act
This research paper breaks down how European AI standardization efforts interact with the upcoming EU AI Act. It digs into the role of harmonized standards, challenges in their development, and how different stakeholders—regulators, industry, and technical bodies—can align.
AI Fairness in Practice: The Alan Turing Institute
AI Fairness in Practice is a workbook published by The Alan Turing Institute as part of its AI Ethics and Governance in Practice programme. It offers a practical, public sector-focused guide to identifying, mitigating, and managing bias across the AI development lifecycle.
Third Draft of the General-Purpose AI Code of Practice
This third draft of the General-Purpose AI Code of Practice sets out voluntary commitments to help general-purpose AI model providers—especially those facing systemic risk—meet their legal duties under the EU AI Act.
AI Risk-Management Standards Profile for General-Purpose AI (GPAI) and Foundation Models
This resource is a versioned framework published by UC Berkeley’s Center for Long-Term Cybersecurity. It provides targeted risk-management guidance for developers of general-purpose AI and foundation models and aligns with the NIST AI Risk Management Framework.
Draft: Critical AI Security Guidelines v1.1 (2024)
Critical AI Security Guidelines v1.1 is a practitioner-driven, point-in-time guide developed by SANS-affiliated experts and contributors from leading cybersecurity and AI organizations. It presents hands-on security recommendations for teams developing generative and LLM-based models.
Foundational artificial intelligence risk assessment framework
The Foundational Artificial Intelligence Risk Assessment Framework (FAIRA) is a public-sector risk tool published by the Queensland Government in 2024. It’s designed to help government teams assess AI systems against legal, ethical, technical, and societal risks across their lifecycle.
