Climate Change AI Innovation Grants for Machine Learning Applied to Climate Action

CIFAR and the Canadian AI Safety Institute fund Catalyst Project proposals addressing sociotechnical considerations in AI safety. The program supports interdisciplinary research in machine learning applications to science and society, with recent funded projects spanning misinformation combat, trustworthy language models, democratic alignment of AI systems, Indigenous AI governance, and real-world safety in autonomous systems. Designed to catalyze new research areas and collaborations at the intersection of social sciences, humanities, and AI safety.

The AI Safety Policy Entrepreneurship Fellowship is FAS's route for turning technical and domain expertise into frontier AI safety policy that actually moves. It is a part-time hybrid programme running 30 September 2026 to 28 February 2027, with applications due 7 September 2026, and it pays a USD 5,000 stipend plus up to USD 1,000 as a merit award. Fellows commit roughly five hours a week to developing a policy memo on a specific AI safety challenge, attend training sessions, join an in-person retreat in California from 4 to 7 November 2026, and present at a capstone event in Washington DC during the week of 22 February 2027. The design target is explicit and unusual: it recruits early- to mid-career professionals who have limited direct public policy experience but deep expertise elsewhere - technical AI research, academia, think tanks, civil society, industry, law, cybersecurity and national security - and teaches them the mechanics of getting an idea adopted. Selection weighs clear understanding of AI governance challenges, concrete implementation-oriented solutions rather than broad principles, awareness of which stakeholders must be moved, and a credible commitment to translating expertise into policy outcomes. Because the stipend is modest and the time commitment part-time, this is designed to sit alongside an existing job rather than replace one.

The VESRI Climate Modeling Challenge is a Schmidt Sciences call, run through the Virtual Earth System Research Institute, that funds research teams to make coupled climate models faster to improve, more reproducible and more accurate. The challenge will fund up to five teams with up to 2 million US dollars each over 24 months to implement and test new methods, explicitly including machine learning, improved representation of physical processes, and advanced calibration workflows. Expressions of intent were invited through 11 September 2026, with a full proposal stage to follow for shortlisted teams. The framing matters: VESRI is targeting the engineering bottleneck in climate modelling rather than climate science questions as such. Coupled model development cycles are slow because calibration is expensive and model updates are hard to reproduce, and the challenge asks teams to demonstrate methods that shorten that loop. Proposals that treat machine learning as an end in itself, rather than as a means to faster and more reproducible model iteration, are mismatched to the brief. VESRI already coordinates hundreds of climate and data scientists across nine projects, 17 countries and 65 research institutions, and Schmidt Sciences has granted 26 million dollars to researchers working on translating climate models into climate action, so this challenge extends an established portfolio. Applicants should expect to compete against teams with existing coupled-model infrastructure, and a proposal without access to a working coupled model to improve is at a structural disadvantage.