Global Governance of AI

Global challenges require global coordination.

Current efforts, gaps and building blocks for an international AI governance framework.

Why do we need global governance for AI?

We need international coordination to ensure the benefits of AI are distributed across the globe. Poor coordination is likely to result in concentration of benefits and power in a small number of hands.

Risks and harms from AI are transboundary–they are not contained to a single jurisdiction and will affect all of humanity as they continue to materialize.

Benefits

Scientific progress
Industry and productivity
Health
Education
Governance
Reduced inequality and poverty
Advancing other SDGs

Risks

Bias
Manipulation and disinformation
Cybersecurity
Labor displacement
Enabling construction of weapons of mass destruction
Great power conflict
Power concentration and exacerbated inequality
Social instability
Loss of control

Building blocks for an international governance regime

To reduce the complexity of international AI governance, it is useful to divide it into components or 'building blocks' that can be analyzed in more depth.

Based on the findings of recent comparative analyses of analogous international governance regimes (most notably, Villalobos, Maas and Winter, forthcoming; Maas and Villalobos, 2023; and Cass-Beggs, Clare and others, 2024), we consider the following building blocks for international AI governance:

Red lines, safety standards, technical cooperation, emergency response, benefit sharing, compliance mechanisms and institutional arrangements.

Current efforts and comparative analysis

Filter by topic:
All initiatives
Red lines
Safety standards
Technical cooperation
Emergency response
Benefit sharing
Institutional arrangements
Compliance mechanisms
Currently showing:
Institutional arrangements

Ensuring that we all benefit from AI while its risks are sufficiently mitigated will require one or more international institutions that implement and enforce international rules on AI. Dozens of institutional designs for AI governance have been recommended by experts, States and international organizations. With the potential exception of the forthcoming UN Independent International Scientific Panel on AI and the Global Dialogue on AI Governance, these institutions have not been created yet.

Table key:
Nonexistent
Insufficient for criteria
Sufficient for criteria
Hints:
Tap / hover over the icon in any yellow or green cell to view the justification for the classification.
Scroll the table horizontally for full content

Research questions 

  1. What is the role of current international organizations in the governance of AI? What key functions are they not fulfilling?

  2. What should be the mandate of novel international organizations created to govern AI?

  3. What lessons of institutional design can we apply to the creation of international organizations for AI governance?

  4. What makes an international organization effective?

 

 Recommended reading 

Governing AI for humanity (High-level Advisory Body on Artificial Intelligence, 2024)

International AI institutions: a literature review of models, examples, and proposals (Maas and Villalobos, 2023)

International institutions for Advanced AI (Ho and others, 2024)

International Governance of Civilian AI: A Jurisdictional Certification Approach (Trager and others, 2023)

What should be Internationalised in AI governance? (Dennis and others, 2024)

Currently showing:
Compliance mechanisms

The most effective international agreements contain robust verification and enforcement mechanisms. States, whether adversarial or not, may be reluctant to enter an international agreement on AI without reliable, secure and privacy-preserving methods to produce timely information on compliance. Hardware-enabled mechanisms have emerged as promising policy solutions for verification and enforcement in international AI governance.

There are no existing governance initiatives that cover this topic.
Hints:

Research questions  

  1. Which parts of the AI lifecycle should verification and enforcement mechanisms target?

  2. What verification mechanisms are currently available to States? What are their main vulnerabilities?

  3. Which hardware-enabled verification and enforcement mechanisms show most promise? What is required to implement them and in what timeline?

  4. What is the role of the chip industry in facilitating the establishment of effective verification and enforcement mechanisms for AI?

 

 Recommended reading 

Mechanisms to Verify International Agreements about AI development (Scher and Thiergart, 2024)

Verification mechanisms for international AI agreements (Wasil and others, 2024)

Secure, Governable Chips  : Using On-Chip Mechanisms to Manage National Security Risks from AI & Advanced Computing (Aarne, Fist and Withers, 2024)

Hardware-enabled Governance Mechanisms (Kulp and others, 2024)

Governing Through the Cloud  : The Intermediary Role of Compute Providers in AI Regulation   (Heim and others, 2024)

Currently showing:
Benefit sharing

While AI poses serious risks, it also promises to be one of the most positively transformative technologies in the history of humanity. However, it is almost certain that AI benefits are not going to be distributed evenly or widely accessible by default. Conscious of this reality, and of not being left behind, many countries that are not at the frontier of AI development are likely to condition their participation in an international AI governance regime on the distribution of its benefits.

Table key:
Nonexistent
Insufficient for criteria
Sufficient for criteria
Hints:
Tap / hover over the icon in any yellow or green cell to view the justification for the classification.
Scroll the table horizontally for full content

Research questions 

  1. What commitments should States make to share the benefits of advanced AI?

  2. Which mechanisms are required to distribute the benefits of AI?

  3. What is the relationship between benefit sharing and the reduction of power concentration?

  4. To what extent can benefit sharing mitigate global labor displacement caused by the adoption of AI?

 

 Recommended reading: 

Assessing potential future artificial intelligence risks, benefits and policy imperatives   (OECD, 2024)

The Windfall Clause: Distributing the Benefits of AI for the Common Good (O'Keefe and others, 2020)

Options and Motivations for International AI Benefit Sharing (Clare and others, 2025)

Voice and Access in AI: Global AI Majority Participation in Artificial Intelligence Development and Governance (Nur Adan and others, 2024)  

Why AI Is WEIRD and Should Not Be This Way: Towards AI For Everyone, With Everyone, By Everyone   (Mihalcea and others, 2024)

Currently showing:
Emergency response

Advanced AI systems pose a plausible threat to international security given that they can facilitate cyberattacks on critical national infrastructure or aid malicious actors in the creation of weapons of mass destruction. These and other scenarios could lead to global emergencies, with transboundary consequences that require a coordinated international response.

There are no existing governance initiatives that cover this topic.
Hints:

Research questions 

  1. What types of possible emergencies relating to global risks from AI should States be prepared for?

  2. What mechanisms should States put in place in order to handle such situations effectively?

  3. What existing emergency response systems could be activated for AI-related emergencies?

 

 Recommended reading 

Sendai Framework for Disaster Risk Reduction 2015-2030 (UNDRR, 2015)

Draft articles on Prevention of Transboundary Harm from Hazardous Activities, with commentaries (International Law Commission, 2001)

Currently showing:
Technical cooperation

As an emerging and fast-paced technology, AI presents significant challenges for the identification and evaluation of risks, as well as for technological sovereignty. International technical cooperation among States and other actors may be required to build capacity, reach scientific consensus on the risks and benefits of AI, conduct joint safety and security research, develop technical mechanisms for verification and enforcement and AI for good, among other areas.

Table key:
Nonexistent
Insufficient for criteria
Sufficient for criteria
Hints:
Tap / hover over the icon in any yellow or green cell to view the justification for the classification.
Scroll the table horizontally for full content

Research questions 

  1. Which types of international technical cooperation on AI should be prioritized?

  2. Which mechanisms would best enable such cooperation (e.g., coordination among affiliated research entities, joint research projects)?

  3. Should States collaborate on the development of advanced AI? If so, what should be the main conditions for this type of joint enterprise?

 

 Recommended reading 

International Scientific Report on the Safety of Advanced AI (Bengio and others, 2025)

Examining AI Safety as a Global Public Good: Implications, Challenges, and Research Priorities (Blomquist and others, 2025)

Key questions for the International Network of AI Safety Institutes (Nur Adan, Guest and Araujo, 2025)

AISIs' Roles in Domestic and International Governance (Ziosi and others, 2024)

We need a science of evals (Apollo Research, 2024)

Superintelligent Agents Pose Catastrophic Risks: Can Scientist AI Offer a Safer Path?   (Bengio and others, 2025)

Currently showing:
All initiatives

Click one of the topics above the table to filter it by topic and find related research questions and literature.

Table key:
Nonexistent
Insufficient for criteria
Sufficient for criteria
Hints:
Tap / hover over the icon in any yellow or green cell to view the justification for the classification.
Scroll the table horizontally for full content

Click one of the topics above the table to filter it by topic and find related research questions and literature.

Currently showing:
Red lines

Many leading AI experts and governments agree that certain practices relating to AI should be internationally prohibited to reduce the likelihood of global scale harm. Top scientists from the United States and China agree that red lines for AI development should include autonomous replication or improvement, power seeking, assistance with weapon development, cyberattacks and deception.

Table key:
Nonexistent
Insufficient for criteria
Sufficient for criteria
Hints:
Tap / hover over the icon in any yellow or green cell to view the justification for the classification.
Scroll the table horizontally for full content

Research questions 

  1. Which types of AI systems or practices, if any, should be prohibited to limit the risk of global scale harm?

  2. Under what conditions might such prohibition(s) be lifted?

  3. What mechanisms should States put in place to ensure that such prohibitions are respected?

 

 Recommended reading 

IDAIS-Beijing 2024 statement

Regulating advanced artificial intelligence agents (Cohen and others, 2024)

Risk thresholds for frontier AI (Koessler, Schuett and Anderljung, 2024)

A Narrow Path (Miotti and others, 2024)

EU AI Act, article 5

Currently showing:
Safety standards

Many international efforts on AI governance focus on the development of principles and best practices to mitigate risks from AI. These can serve as the basis for international technical standards that establish requirements for advanced AI developers and enhance coherence, harmonization, and regulatory effectiveness across jurisdictions.

Table key:
Nonexistent
Insufficient for criteria
Sufficient for criteria
Hints:
Tap / hover over the icon in any yellow or green cell to view the justification for the classification.
Scroll the table horizontally for full content

Research questions  

  1. Which existing international standards can be applied to advanced AI developers? What are their main gaps?

  2. Can international principles and codes of conduct, like the OECD Principles and the Hiroshima Process International Code of Conduct for Advanced AI Systems, act as blueprints for international safety standards?

  3.   What should be the scope of international AI safety standards?

Further Reading:

A Frontier AI Risk Management Framework: Bridging the Gap Between Current AI Practices and Established Risk Management   (Campos and others, 2025)

Safety Standards delivering Controllable and Beneficial AI Tools (Ben Cumming, 2025)

Model evaluation for extreme risks (Shevlane and others, 2023)

Frontier AI Regulation: Managing Emerging Risks to Public Safety (Anderljung and others, 2023)

Responsible scaling policy (Anthropic, 2024)

Hiroshima Process International Code of Conduct for Advanced AI Systems (2023)

Join the email list

(Watch this space)

Recommended reading

Contact Us

José Jaime Villalobos:
Multilateral Governance Lead
Future of Life Institute
[email protected]
Ima Bello:
Multilateral Engagement Lead
Future of Life Institute
[email protected]
Global Governance of AI
Global challenges require global coordination.