Build a robust foundation in one of AI safety’s most exciting and critical emerging fields.
8-week online structured programme.
Weekly exercises and group discussions facilitated by field practitioners.
Access to a dedicated forum and group chats.
Receive a Certificate of Completion.
Find out more and apply via this form! (Deadline: 12 June 11:59 AoE).If you have any questions, feel free to email my colleague Joss who will be coordinating the course.
NataliaFA
Karma: 7
- Hello! We (Cooperative AI Foundation) are running a facilitated 8-week online ‘Introduction to Cooperative AI’ course from 6 July − 28 August 2026.
Hi everyone! Make sure to join the Cooperative AI Foundation’s next seminar:
Title: Benchmarking Offline RL in Mixed-Motive Social SettingsDate: Thursday 21 May, 16:00 UTCSpeaker: Claude Formanek (University of Cape Town)About: Reinforcement learning is increasingly applied to social, data-limited environments where agents must interact with others without the luxury of online experimentation. However, standard offline RL benchmarks are non-social. This seminar introduces Molten Pot, a benchmark for offline mixed-motive social RL that spans five substrates from DeepMind’s Melting Pot environment and ~1TB of trajectory data. Testing agents across three levels of social complexity reveals a significant ‘social robustness gap’.Register for the seminar
Sharing the Cooperative AI Foundation’s next seminar in our ‘Updates in Cooperative AI’ series:
Title: Colosseum: Auditing Collusion in Cooperative Multi-Agent SystemsDate: Thursday 23 April, 16:00 UTCSpeaker: Sahar Abdelnabi (ELLIS Institute Tübingen, Max Planck Institute for Intelligent Systems)About: Sahar will present findings from applying Colosseum, a principled new framework to study collusion in multi-agent settings by jointly analysing communications and actions in rich yet verifiable environments.
The Cooperative AI Foundation is kicking off a ‘Fellows’ Spotlight’ seminar series featuring the most exciting research from our Cooperative AI PhD Fellows. Make sure to register for our first seminar:
Title: Safe Pareto Improvements: Cooperative Commitments without Compromise
Thursday 26 March, 17:00 UTC
Speaker: Nathaniel Sauerberg (University of Texas at Austin)
About: Advanced AI systems may introduce many new ways to credibly commit to and enforce agreements in strategic interactions. However, because players may disagree on what constitutes a mutually beneficial outcome, Safe Pareto Improvements address this by adjusting payoffs without altering the structure of the game—ensuring that outcomes improve for all while avoiding coordination problems.Register for the seminar: https://www.cooperativeai.com/seminars/safe-pareto-improvements-seminar
Update from the Cooperative AI Foundation: Summer School 2026
The Cooperative AI Summer School 2026 ‘Expression of interest’ applications are now open! Taking place in Canada (location TBA) from 3-7 August. Watch last year’s recap video for a flavour on what to expect.
This is an advanced programme for early-career professionals already working in or studying cooperative AI. Over five intensive days, participants will attend talks by leading researchers and work on collaborative projects tackling real cooperative AI challenges. Financial assistance is available, and we especially encourage people from underrepresented groups to apply.
Find out more and apply by 22 March 2026 (11.59pm AoE) on the Cooperative AI website.
Niklas Lauffer will talk at the Cooperative AI Foundation’s next ‘Fellows’ Spotlight’. He recently completed his PhD at UC Berkeley and will be joining Google DeepMind as a research scientist.
Title: Robust and Diverse Multi-Agent Learning via Rational Policy Gradient
Speaker: Niklas Lauffer, UC Berkeley
Date: Thursday 25 June 2026 (16.00–17.00 UTC)
About: This talk explores how to make adversarial training work in cooperative multi-agent settings by ensuring agents remain rational, thus preventing self-sabotage—a critical failure mode in these types of settings - while still discovering adversarial examples, improving robustness, and learning diverse policies.