OECD.AI

The 5th GPAI Plenary brought some of the sharpest minds in AI governance to the OECD last week. The conversations that happened here about embodied AI, agentic systems, investor accountability and transparency will shape what comes next. A sincere thank you to everyone who joined us in Paris for the workshops and expert group meetings on the margins of the 5th Plenary of the Global Partnership on Artificial Intelligence. A special word of gratitude to our GPAI Co-Chairs, Denise Wong of Singapore and Hayun Kang of the Republic of Korea, for their leadership in steering this community and keeping our collective work grounded in both ambition and rigour. Over two days, we heard from an exceptional group of experts: On embodied AI and autonomous systems: Leslie Kaelbling (MIT), Tom Erez and Barath Harithas (Google DeepMind), and Gregorio Ameyugo (CEA) helped us think through what it really means for AI to act in the physical world, and where technical advances are fueling progress and where challenges to deployment still lie. On trustworthy AI investment: Amir Banifatemi and Francesca Rossi (GPAI TAI Expert Group Co-Chairs), Laurie Fitzjohn-Sykes (Omidyar Network / Investor AI Resource Hub), Sophie Walker (EQT Partners), and Paul Fehlinger (Project Liberty) pushed us on the hard question of how financial incentives and trustworthy AI outcomes can actually align. On agentic AI, data governance, and privacy: presentations from the OECD Secretariat and Tokyo Centre of the GPAI Expert Community, the Infocomm Media Development Authority, Covington & Burling LLP, and the Spanish Data Protection Authority mapped out key points for governance. And throughout, the Hiroshima AI Process Reporting Framework 2.0 remained a throughline, a reminder that voluntary transparency, done well, can build the cross-border trust that AI governance so urgently needs. Special thanks to Ashish Tewari, Hector de Rivoire, Pauline Charazac, James G. and Yukio Teramura for their participation. These meetings matter because the people in them do. Thank you for bringing your expertise, your candour, and your commitment to the table. 🔔 Follow OECD.AI on LinkedIn to stay up to date with the latest on AI policy, governance, and the work of the GPAI community. #AIPolicy #GPAI #TrustworthyAI #ArtificialIntelligence #OECDai

A big thanks to all the experts who took the time to speak at two timely workshops exploring the governance and technological frontiers of AI at this week's GPAI Plenary. The first workshop focused on version 2.0 of the Hiroshima AI Process (HAIP) Reporting Framework 2.0, bringing together policymakers, companies and civil society to discuss transparency, accountability and international trust in advanced AI systems. Speakers included: 🔹  Sara Rendtorff-Smith, Head of Division, OECD 🔹 Yukio Teramura, Assistant Vice Minister for International Affairs, Japan Ministry of Internal Affairs and Communications 🔹 Pauline Charazac, Head of Policy Engagement, Centre pour la Sécurité de l’IA (CeSIA) 🔹 Hector de Rivoire, Director, Responsible AI Public Policy, Microsoft 🔹 Ashish Tewari, Head of Responsible AI Office, Infosys 🔹 James G., Standardisation Lead, SaferAI The second workshop explored recent advances in embodied AI and autonomous systems operating in the physical world, from robotics to industrial automation. Experts included: 🔹 Leslie Kaelbling, Panasonic Professor of Computer Science and Engineering, MIT 🔹 Tom Erez, Research Scientist, Google DeepMind 🔹 Barath Harithas, Frontier Policy Manager, Google DeepMind 🔹 Gregorio Ameyugo, Head of AI & Interactive Systems Division, CEA As AI capabilities continue to evolve, conversations that connect technical advances with practical governance approaches are becoming increasingly important. Learn more about the Hiroshima AI Process Reporting Framework in the link in the comments. #AIGovernance #GPAI #TrustworthyAI #OECDAI #HAIP #EmbodiedAI

How do we feed a growing population when climate shocks, labour shortages and fragile supply chains are putting global food systems under increasing strain?   Artificial intelligence is emerging as a powerful tool to help build more resilient, inclusive and sustainable agri-food systems.   In a new AI Wonk article, Mr Marten Van den Berg, Director General Agriculture of the Dutch Ministry of Agriculture, Fisheries, Food Security and Nature, and Sara Rendtorff-Smith, Head of Division at the #OECD, explore how AI can help:   🔹  Improve crop yields while reducing water, fertiliser and pesticide use 🔹  Make agri-food supply chains more efficient and secure 🔹  Give farmers earlier warnings of droughts, pests and crop diseases 🔹  Make agricultural advice more accessible through local-language AI tools   The article also highlights three critical challenges for AI to deliver on its promise:   🔹 Improving data interoperability and data-sharing 🔹 Designing AI solutions to fit the local needs of farmers, especially smallholders 🔹 Successfully scaling AI agri-food pilots in the field   As countries look to strengthen agri-food resilience, international co-operation will be essential to ensure the benefits of AI are widely shared.   𝗥𝗲𝗮𝗱 𝘁𝗵𝗲 𝗳𝘂𝗹𝗹 𝗮𝗿𝘁𝗶𝗰𝗹𝗲 𝗮𝘁 𝘁𝗵𝗲 𝗹𝗶𝗻𝗸 𝗶𝗻 𝘁𝗵𝗲 𝗰𝗼𝗺𝗺𝗲𝗻𝘁𝘀 𝗯𝗲𝗹𝗼𝘄.👇   #Netherlands #GPAI #ArtificialIntelligence #Agriculture #AgriFood

🛠️🧰⚙️🔧 Today is the last day for submissions to the OECD.AI Catalogue of Tools & Metrics for Trustworthy AI to qualify for LinkedIn promotion. Anyone can submit a tool at any time, but tools submitted during the campaign period will qualify for promotion on this LinkedIn account. The Catalogue is a global reference for policymakers, regulators, enterprises, and researchers seeking tools to make AI systems and the environments they operate in more trustworthy. Inclusion makes your work findable to the audiences most likely to apply it. 👉 Submit now: https://lnkd.in/eXwtBkAU #TrustworthyAI #OECDAI

Governments do not struggle to agree on AI principles. They struggle to put them into practice. That is the challenge the new OECD AI Policy Toolkit addresses, launched yesterday on the sidelines of the OECD Ministerial and under the OECD's Global Partnership on AI (GPAI). In a new AI Wonk article, H.E. Minister Paula Bogantes Zamora, Minister of Science, Innovation, Technology and Telecommunications of Costa Rica, explains why policymakers need practical guidance to design, implement and improve AI policies. Key takeaways: 🔹 The Toolkit is designed to help translate the OECD AI Principles into concrete policy action. 🔹 AI policy must reflect and be designed to fit national contexts, capacities and priorities. 🔹 Real-world examples and proven approaches can be of tremendous help to Policymakers designing AI policies. 🔹 International co-operation is essential for addressing shared AI challenges. The Toolkit was developed by and for policymakers, through a year-long co-creation process involving policymakers, experts and stakeholders across regions. 𝗥𝗲𝗮𝗱 𝘁𝗵𝗲 𝗳𝘂𝗹𝗹 𝗮𝗿𝘁𝗶𝗰𝗹𝗲 𝘃𝗶𝗮 𝘁𝗵𝗲 𝗹𝗶𝗻𝗸 𝗶𝗻 𝘁𝗵𝗲 𝗰𝗼𝗺𝗺𝗲𝗻𝘁𝘀. 👇 Lucia Russo Guillermo H.  Eunseo Dana Choi  Isabella DeClue #AIGovernance #TrustworthyAI #PublicPolicy #OECDAI

Governments and businesses are turning AI principles into practice while building trust.   Join the OECD-Global Partnership on Artificial Intelligence (GPAI) side event at #OECDministerial to learn how governments and companies are working together to support trustworthy AI innovation and wider adoption.   The session will showcase the #OECDAI Policy Toolkit as well as the Hiroshima AI Process Reporting Framework (HAIP) 2.0, a resource for organisations to identify and disseminate good practice by sharing information about how they manage AI risks, safety and transparency.    2 June | 15:00-17:00 CEST For more information about speakers and other details, visit the event's webpage: https://lnkd.in/eukQQqXa

What does practical AI governance look like?   𝗝𝗼𝗶𝗻 𝘂𝘀 𝗼𝗻𝗹𝗶𝗻𝗲 𝘁𝗼𝗱𝗮𝘆 for the #OECD/GPAI side event at #OECDministerial to learn how the AI Policy Toolkit and updated HAIP Reporting Framework can support transparency, implementation and international co-operation.    2 June | 15:00-17:00 CEST   🔗 https://lnkd.in/eFwMz94T   #OECDAI

How can governments and businesses move from AI principles to implementation while building trust?   Join us online for the OECD/GPAI side event at #OECDministerial, Advancing International Cooperation on Artificial Intelligence: The AI Policy Toolkit and the Hiroshima AI Process Reporting Framework.   ➡️The OECD/GPAI AI Policy Toolkit ➡️ The updated Hiroshima AI Process Transparency Reporting Framework for organisations developing and working with advanced AI    𝟮 𝗝𝘂𝗻𝗲 | 𝟭𝟱:𝟬𝟬-𝟭𝟳:𝟬𝟬 𝗖𝗘𝗦𝗧 🔗 https://lnkd.in/eukQQqXa   #OECDAI

What does it take to evaluate an AI system responsibly? There are many tools to help make AI systems more trustworthy. Some are open source, built by communities who understand that scrutiny is a public good. The OECD.AI Catalogue of Tools & Metrics for Trustworthy AI currently hosts almost one thousand tools. It is consulted by national AI safety institutes, regulators, enterprises, and academic teams. 𝗦𝘂𝗯𝗺𝗶𝘁 𝗯𝘆 𝟱 𝗝𝘂𝗻𝗲, 𝗮𝗻𝗱 𝘆𝗼𝘂𝗿 𝘁𝗼𝗼𝗹 𝘄𝗶𝗹𝗹 𝗾𝘂𝗮𝗹𝗶𝗳𝘆 𝗳𝗼𝗿 𝗮 𝗰𝗵𝗮𝗻𝗰𝗲 𝘁𝗼 𝗯𝗲 𝗽𝗿𝗼𝗺𝗼𝘁𝗲𝗱 𝗼𝗻 𝘁𝗵𝗲 𝗢𝗘𝗖𝗗.𝗔𝗜 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 𝗮𝗰𝗰𝗼𝘂𝗻𝘁. 👉 https://lnkd.in/eXwtBkAU 𝗦𝗵𝗮𝗿𝗲 𝗮𝗻𝗱 𝘀𝗽𝗿𝗲𝗮𝗱 𝘁𝗵𝗲 𝘄𝗼𝗿𝗱! #TrustworthyAI #OECDAI #OpenSource

AI is becoming part of critical systems. But are we building the security foundations fast enough? As AI systems become more capable, connected and agentic, security risks are evolving beyond traditional cybersecurity. Shared approaches to evaluation, testing and governance will be essential to support safe and trusted adoption. New on the AI Wonk, Hector de Rivoire, Nicholas Butts, Constance de Leusse and Elizabeth Seger explore three priorities for collective AI security: 🔹 Defending against scalable and transferable prompt injection attacks 🔹 Securing AI agents connected to tools, memory and external services 🔹 Detecting and mitigating model poisoning across the AI supply chain The authors argue that stronger international collaboration on benchmarks, information-sharing and secure design practices in forums like the #G7 and the #OECD - #GPAI will be critical as AI becomes more embedded across economies. 𝗥𝗲𝗮𝗱 𝘁𝗵𝗲 𝗯𝗹𝗼𝗴 𝗽𝗼𝘀𝘁 𝗮𝘁 𝘁𝗵𝗲 𝗹𝗶𝗻𝗸 𝗯𝗲𝗹𝗼𝘄.👇 #AISecurity #TrustworthyAI #PromptInjection #AgenticAI #AIPolicy #AIWonk