人工智能治理问责制的国际协调：第六届雅典人工智能与法治圆桌会议要点

InternationalCoordination for Key Takeaways from the Sixth Edition ofthe Athens Roundtable on Artificial George Gor,Frank Ryan,Tereza Zoumpalova,Niki Iliadis Table of Contents About the Sixth Editionof the Athens Roundtable on Artificial Intelligence and the Rule of Law4 GovernmentsIntergovernmental Organizations The Sixth Edition In Numbers Opening RemarksSession 1: Mapping the Landscape of AI Governance and Accountability AcknowledgementsSummary of Key Insights and Considerations The Sixth Edition of the Athens OECD Headquarters, Paris, FranceDecember 9, 2024 Under the patronage of H.E. the President of the Hellenic ORGANIZED BY: IN PARTNERSHIP WITH SUPPORTFROM: About The Sixth Edition establish binding global AI standards andaccountability mechanisms. Rigorous testing,independent evaluations, and third-party audits,supported by AI Safety Institutes, were recognizedas vital for building safe and accountable AI. Whileadvancing scientific consensus on AI risksremains essential, our discussions highlighted that The Sixth Edition of the Athens Roundtable on AIand the Rule of Law, hosted at the OECD in Parison December 9, provided a valuable platform forengaging with one of the defining challenges ofour time: establishing robust accountabilitymechanisms to ensure AI is developed andgoverned in alignment with the rule of law.Convening 135 in-person participants—including As we reflect on the outcomes of this eventthroughout this report, one thing is clear: buildingeffective governance for AI is not an endpoint butan ongoing endeavor. The stakes are high:without effective accountability mechanisms, AIwill exacerbate inequality, erode human rights,and destabilize our democratic institutions. Ashighlighted in our opening remarks: “if AIdevelopers hold themselves accountable to We were honored to host this event under thepatronage of Her Excellency the President of theHellenic Republic, Ms. Katerina Sakellaropoulou,and in partnership with the OECD, UNESCO, theAI & Society Institute, CNNum, and Make.org, withsupport from Arnold & Porter and Fathom. Discussions revolved around four themes:assessing progress in AI governance, identifyinggaps in existing governance regimes, craftingsolutions, and defining actionable priorities for the With appreciation and hope, Nick MoësExecutive Director Niki IliadisDirector, Global AI Governance Recommendations for Key Who holds the responsibility for ensuring that thedevelopment and deployment of AI serve tobenefit humanity? Our view is that this must be ashared effort, requiring proactive engagementfrom governments, intergovernmental GOVERNMENTS Harmonize International Standards Engage in bilateral and multilateral efforts to alignAI governance frameworks through establishedinternational mechanisms such as the OECD AIPrinciples, ensuring interoperability acrossjurisdictions. Secure high-level political Each actor brings unique perspectives, resources, and expertise tothe governance process. Together, we can tackle the significant we identified key recommendations for whatdifferent actors can do to advance AI governance.To derive these recommendations, we analyzedkey insights from each session and categorizedthem based on the actors who could most benefit.This approach synthesized what was voiced Establish Enforceable Oversight Move beyond reliance on voluntary industrycommitments to mandatory requirements bycreating binding legal mandates. Adoptcomprehensive oversight mechanisms and clear Require Rigorous Testing, Independent Mandate frequent and standardized independent INTERGOVERNMENTAL mechanisms and incident reporting requirementsacross all stages of AI development and Develop Global Standards Collaborate with theInternational Network of AISafety Institutesand the OECD to establish andcontinuously update interoperable standards forAI governance, leveraging the momentum of the Commit to Lifecycle Accountability Take full responsibility for AI throughout theirlifecycle, including proactively monitoring theirimpacts post-deployment. Address unintended Create Binding International Frameworks CIVIL SOCIETY Support agreements that require companies andgovernments to adopt specific and clear oversightprotocols and measurable compliancerequirements, including with clear incentives and Enable Inclusive Participation in AI GovernanceBuild coalitions across civil society organizations to ensure underrepresented voices are activelyincluded in decision-making forums. Fosterequitable representation and create avenues for Monitor and Assess AI Impacts Foster Global Cooperation Establish independent, well-funded initiatives tosystematically track, analyze, and publicly reporton the societal impacts of AI. Use theseassessments to hold actors (primarily Institutionalize joint oversight mechanisms—especially through a permanent structure for theInternational Network of AI Safety Institutes—toaddress cross-border risks, support a holistic INDUSTRY