Consensus Statement on AI Safety as a Global Public Good

• Authority: Other multinational
• Jurisdiction: Multinational
• Collections: Multinational;Editors' Picks
• Most recent activity: Enacted
• Most recent activity date: 2024-09-08

Risk subdomain coverage

Good coverage — Comprehensive governance measures or mitigations with clear procedures relating to the risk subdomain are explicitly described.

• Competitive dynamics
• Governance failure
• AI pursuing its own goals
• AI possessing dangerous capabilities
• Lack of capability or robustness
• Lack of transparency or interpretability
• Multi-agent risks

Minimal coverage — Brief mention but no specific mitigations, recommendations or governance measures relating to the risk subdomain are described.

• Compromise of privacy
• AI system security vulnerabilities
• Disinformation, surveillance, and influence at scale
• Cyberattacks, weapon development or use
• Loss of human agency and autonomy

All other risk subdomains had no coverage in this document.

Risk Coverage Summary

The document has good coverage of approximately 10-12 subdomains, with strong focus on AI system safety failures (7.1, 7.2, 7.3), governance failure (6.5), competitive dynamics (6.4), malicious actors (4.1, 4.2), AI system security (2.2), and multi-agent risks (7.6). Coverage is concentrated in system safety, governance, and misuse prevention domains.

Sector coverage

Good coverage — Comprehensive governance measures or mitigations with clear procedures relating to the sector are explicitly described.

• Scientific Research & Development Services
• Public Administration (excl. National Security)

Minimal coverage — Brief mention but no specific mitigations, recommendations or governance measures relating to the sector are described.

• National Security

All other sectors had no coverage in this document.

Sectors Summary

This is a cross-sectoral governance framework that does not target specific economic sectors. Instead, it establishes general principles and mechanisms for AI safety governance applicable to all sectors where advanced AI systems may be developed or deployed. The document focuses on frontier AI developers and state governance mechanisms rather than sector-specific applications.

Legislative Status

Soft Law

Legislative Status Summary

This is a consensus statement that uses voluntary and recommendatory language throughout, calling on states and developers to take actions without establishing binding legal obligations or enforcement mechanisms.

Actors

Proposer

• Entity Types: AI_Governance_Actor
• Summary: The document is titled 'Consensus Statement' and uses collective language ('we call on', 'Collectively, we must') suggesting it represents agreement among multiple parties in the AI safety community rather than a single proposing entity.

Target

• Entity Types: AI_Governance_Actor, AI_Developer
• Summary: The document explicitly calls on states to take governance actions and frontier AI developers to demonstrate safety. It also addresses AI Safety Institutes, philanthropists, corporations and experts to support research and verification efforts.

Enforcer

• Entity Types: AI_Governance_Actor
• Summary: The document calls for states to develop domestic authorities and an international body to enforce safety measures. It also describes roles for independent auditors and third parties in verification and enforcement.

Monitor

• Entity Types: AI_Governance_Actor
• Summary: The document describes extensive monitoring roles including post-deployment monitoring, incident tracking, independent audits, and verification of safety claims by domestic authorities and international bodies.

AI Lifecycle Stages

Lifecycle stages covered by the documents, as defined in the OECD Framework for the Classification of AI Systems

• Plan and Design
• Build and Use Model
• Verify and Validate
• Deploy
• Operate and Monitor

AI Lifecycle Summary

The document covers multiple lifecycle stages with primary focus on Verify and Validate, Deploy, and Operate and Monitor stages. It emphasizes pre-deployment testing, safety cases, independent audits, and post-deployment monitoring. There is also coverage of Build and Use Model through discussion of training thresholds and model development practices.

Technical Scope

Terms relating to technical scope that are mentioned or implicitly referenced in the document

• AI Models
• Frontier AI
• AI Systems
• Compute Threshold

Technical Scope Summary

The document explicitly mentions AI systems and models throughout. It focuses heavily on frontier AI and advanced AI systems, with references to capability thresholds and early-warning thresholds. It does not explicitly define or distinguish between general purpose AI, task-specific AI, foundation models, generative AI, or predictive AI. There are no specific compute thresholds mentioned in FLOP terms, though capability thresholds are discussed. Open-weight models are not explicitly mentioned.