Establishing the shared foundations for collective AI security (summary)

OECD AI Wonk - 21 May 2026

Authors: Hector de Rivoire - Director of Public Policy, Microsoft Office of Responsible AI; Constance de Leusse - Executive Director AI & Society, ENS-PSL AI & Society Institute; Elizabeth Seger - Senior Policy Advisor, AI Governance & Policy, Tony Blair Institute; Nicolas Butts - Director of AI and Cybersecurity Policy, Microsoft

Overview

This OECD AI Wonk article examines the growing international effort to establish common foundations for AI security governance as increasingly capable AI systems are deployed globally. The authors argue that AI security must be approached as a shared international challenge requiring interoperable governance structures, common terminology, coordinated risk management practices, and collaboration between governments, industry, academia, and civil society.

The piece frames AI security not solely as a technical cybersecurity issue, but as a broader governance challenge involving system robustness, misuse prevention, democratic resilience, and international coordination. It situates these discussions within ongoing OECD and Global Partnership on Artificial Intelligence (GPAI) initiatives aimed at developing globally compatible approaches to trustworthy AI governance.

The Need for Shared AI Security Foundations

The authors explain that AI development and deployment increasingly transcend national borders through shared cloud infrastructure, multinational companies, open-source ecosystems, and interconnected supply chains. As a result, fragmented national governance approaches risk creating regulatory gaps and inconsistent standards.

The article stresses that countries and institutions are beginning to converge around several core AI security concerns:

• model robustness and reliability,

• misuse prevention,

• safety evaluation,

• governance accountability,

• information integrity,

• and resilience against unintended system behavior.

The authors argue that without shared foundations, governments and organizations may adopt incompatible standards that complicate international cooperation and reduce the effectiveness of risk mitigation efforts.

OECD AI Principles as a Governance Baseline

A major focus of the article is the role of the OECD AI Principles as an international baseline for trustworthy AI governance. The authors describe the principles as an important common language that enables countries with differing political and legal systems to cooperate around shared governance objectives.

The article highlights several principles especially relevant to AI security:

• robustness, security, and safety,

• transparency and explainability,

• accountability,

• protection of human rights and democratic values,

• and international interoperability.

The authors emphasize that the principles were intentionally designed to remain adaptable despite rapid advances in AI capabilities and deployment models.

AI Security as a Collective-Action Problem

The article repeatedly frames AI security as a collective-action problem that no single country can solve independently. Since AI systems, compute resources, and datasets often span multiple jurisdictions, international coordination becomes essential.

The authors discuss the importance of:

• shared technical standards,

• interoperable governance frameworks,

• coordinated evaluation methodologies,

• incident reporting mechanisms,

• and multistakeholder cooperation.

The piece argues that international organizations can play an important convening role by helping states align governance practices even where regulatory approaches differ substantially.

GPAI and International Governance Coordination

The article discusses the growing role of GPAI and related OECD initiatives in bridging the gap between high-level principles and practical policy implementation. The authors explain that GPAI brings together:

• governments,

• industry representatives,

• technical experts,

• researchers,

• and civil society organizations.

This multistakeholder structure is presented as a mechanism for translating broad governance concepts into operational tools, policy guidance, and evidence-based practices.

The article also notes that the closer integration of GPAI into OECD AI governance work is intended to:

• reduce duplication,

• strengthen interoperability,

• improve policy coordination,

• and create stronger links between technical expertise and policymaking processes.

Risk Management and Trustworthy AI

Risk management is presented as central to collective AI security. The authors argue that governance frameworks should account for the varying levels of risk associated with different AI systems and deployment contexts.

Rather than promoting uniform regulation, the article supports proportional approaches based on:

• capability,

• deployment environment,

• societal impact,

• and technical feasibility.

The authors emphasize the importance of:

• safeguards against misuse,

• human oversight,

• incident monitoring,

• transparency mechanisms,

• and the ability to intervene or deactivate systems when necessary.

The article also links AI security directly to public trust, arguing that trustworthy AI systems must be resilient, explainable, and accountable in practice, not merely innovative.

Information Integrity and Democratic Resilience

Another major theme is the relationship between AI security and democratic stability.

The authors discuss concerns surrounding:

• misinformation and disinformation,

• synthetic media,

• manipulation of information ecosystems,

• and broader societal harms associated with increasingly capable generative AI systems.

The article argues that governance frameworks must balance security objectives with the protection of:

• freedom of expression,

• privacy,

• human rights,

• and democratic values.

The OECD governance framework is presented as an attempt to reconcile innovation and openness with safeguards designed to reduce harmful uses of AI technologies.

Building Practical Governance Infrastructure

The article concludes by emphasizing that effective collective AI security requires operational governance infrastructure in addition to high-level principles.

The authors identify the need for:

• common definitions,

• shared evaluation metrics,

• incident tracking systems,

• governance toolkits,

• and continuous international expert collaboration.

The OECD.AI platform and related initiatives are described as practical mechanisms intended to support evidence-based policymaking and improve international coordination around AI governance.

The overall argument is that the security challenges posed by advanced AI systems can only be addressed effectively through sustained international cooperation grounded in shared principles, interoperable governance mechanisms, and ongoing multistakeholder collaboration.

RESOURCES

• Establishing the Shared Foundations for Collective AI Security - the source OECD.AI Wonk article

• OECD AI Principles - first intergovernmental standard, the governance baseline discussed throughout

• Global Partnership on Artificial Intelligence (GPAI) - integrated partnership bridging principles and implementation

• OECD.AI Policy Observatory - the practical platform for evidence-based AI policymaking

• GPAI-OECD integration announcement - New Delhi Summit, July 2024

• Microsoft Office of Responsible AI - affiliation of co-authors Hector de Rivoire and Nicolas Butts

• AI & Society Institute, ENS-PSL - where co-author Constance de Leusse is Executive Director

• Tony Blair Institute - where co-author Elizabeth Seger advises on AI governance