AI Governance Taskforce autumn 2026 cohort

AI Governance Taskforce autumn 2026 cohort is sponsored by Arcadia Impact. This is a 12-week, remote (global), part-time (8 hrs/wk) research programme for experienced professionals working to reduce catastrophic risks from advanced AI through policy and governance research. It focuses on acute and systemic risks.

AI Governance Taskforce — Arcadia Impact We support experienced professionals transitioning into frontier AI governance careers focused on reducing catastrophic risks, acute and systemic, from transformative AI, by building credentials and connections through policy research remotely alongside their existing commitments. Applications for our Autumn 2026 cohort are now open.

Apply by 23:59 Anywhere on Earth on Monday 31 August 2026 (12:59 BST, Tuesday 1 September). The Autumn cohort runs from 12 October 2026 to 15 January 2027, with a two-week year-end break. The Summer 2026 cohort concludes on 21 August.

See our Research page for our forthcoming outputs from the summer cohort, and our prior publications. For questions about the Taskforce, please contact: Ben R Smith, Taskforce Lead: ben [at] arcadiaimpact [dot] org.

Our AI Governance Taskforce is a 12 week, remote (global), part-time (8 hrs/wk) research programme for experienced professionals working to reduce catastrophic risks, acute and systemic, from advanced AI through policy and governance research. The programme is designed to be compatible with existing commitments.

Research Associates work in multidisciplinary teams of 4, led by a Research Team Leader and working with an expert partner who provides research direction and oversight.

Expert partners are senior practitioners in AI safety and governance, contributing in a personal or organisational capacity, drawn from organisations such as the UK AI Security Institute (AISI), GovAI, the AI Incident Database (AIID), MIT’s AI Risk Initiative (AIRI), the Centre for Long-Term Resilience (CLTR), SaferAI, The Future Society (TFS), the Future of Life Institute (FLI) and academia.

The work produces policy-relevant research outputs and gives participants the credentials and connections to transition into work in the field. A core strength of the Taskforce is our in-house research leadership. Our Research Team Leaders work closely with our expert partners to co-design research questions, then lead and deliver research outputs with our Research Associate teams.

The research typically results in a long-form academic-style pre-print paper of policy brief, and short-form policy memos and articles for targeted audiences and stakeholders.

Alumni of the Taskforce have gone on to work at Google DeepMind, SaferAI, Pour Demain, ControlAI, collaborate with Yoshua Bengio, the Oxford Martin AI Governance Initiative, MIT AI Risk Initiative, Centre for Long Term Resilience (CLTR), SaferAI take up AI governance relevant academic positions, research management positions at other fellowships (ERA, Safe AI Germany, PRISM), and fellowships at GovAI, IAPS, Future Impact Group, placements via Talos, at The Future Society, pursue independent research, extend work building on their Taskforce projects, start their own initiatives, and progress into Research Team Lead and Research Management stints within the AI Governance Taskforce itself.

Our teams have output research that have been accepted at the International Conference on Machine Learning (ICML) Technical AI Governance workshop, including a best paper award, obtained co-authorship with some of our expert partners, and been featured in Time magazine, the Bulletin of the Atomic Scientists and the AI Policy Bulletin.

Applications close: 23:59 Anywhere on Earth, Monday 31 August 2026 (12:59 BST, Tuesday 1 September) When: 12 October to 18 December 2026, then 4 to 15 January 2027, with a two week year-end break in between Hours and duration: 8 hrs per week, 12 weeks of delivery in total Employment type and remuneration: voluntary, unpaid Location: remote-only, global Places: approximately 20 Research Associate places About the Taskforce Research Associate role We look for highly motivated candidates who are driven to transition from established careers into frontier AI governance focused on reducing catastrophic risks, acute and systemic, from advanced AI.

We are interested in candidates from a range of backgrounds, disciplines and experience levels, including law, public policy, economics, other social science and humanities, management, political science, international relations, engineering (of various kinds, e.g. ML, software, mechanical, systems, safety), philosophy and computer science that can be brought to bear with direct or indirect relevance to our projects.

If you only meet some of the criteria, we encourage you to apply. Individual projects may have additional requirements. Full project briefs will be shared with candidates invited to the project application stage.

To get a sense of the kind of projects we offer and work our teams produce, see our Research page. Participants are not paid; this is a voluntary opportunity designed to support participants' careers in frontier AI governance. If you only meet some of what follows, we encourage you to apply.

You do NOT need to have a deep technical understanding of frontier AI. • Committed to a career reducing catastrophic and systemic risks from frontier AI. This programme is not about general AI risk, corporate AI compliance & AI ethics.

• Able to commit at least 8 hours per week across the cohort dates, alongside your existing commitments • Willing to work in a small, structured, remote research team of four alongside a Research Team Leader and an expert partner • Readiness and drive: already moving towards this work, not deciding whether to • Situational awareness of the frontier AI landscape: capabilities, actors, risks, and the work already underway • A demonstrable competency & expertise from another field or discipline that transfers to frontier AI governance research • Strong reasoning and clear writing, research and policy skills are helpful • Judgement about what matters and what is tractable in frontier AI governance, and clear on the conditions that judgement is based on • Self-direction: pursuing goals without an existing mandate or structure Already moving towards the field.

Given the pace of frontier AI development, we prioritise candidates who have already started moving toward this work, in whatever way their circumstances allow: study, writing, volunteering, foundational programs (such as BlueDot courses), applications elsewhere. Self-direction. Useful work in this field may involves creating opportunities, evidence and structures that do not yet exist, often without a mandate.

Judgement. We are looking for clear reasoning about what matters, what is tractable, and openness to changing views in light of new arguments and evidence. For the Autumn 2026 cohort the process has three stages.

No writing sample is required at any stage. 1. General application .

Your CV and short written answers. We assess your motivation and readiness to move into the field, your situational awareness of frontier AI risk, your judgement about priorities and interventions, how your existing expertise transfers, your self-direction, and the clarity of your reasoning. 2.

Project application . Shortlisted candidates are given project briefs, and invited to apply to the projects that fit them best, completing a short written task related to their chosen project. 3.

Interview . A remote interview to get to know you, and expand on aspects of your application. 4.

Offers . Successful candidates are offered a place on a specific project team. We expect to offer approximately 20 Research Associate places, and we typically receive many more strong applications than we are able to accept.

If you are unsure whether your background fits, we would rather read your application and decide than have you rule yourself out. Anywhere on Earth (AoE) deadlines mean the deadline has not passed while it is still that date anywhere in the world. UK equivalents are given for the fixed deadlines.

Stage 1 general applications open: Wednesday 5 August 2026 Stage 1 general applications close: Monday 31 August 2026 AoE (12:59 BST, Tuesday 1 September) Stage 2 invitations issued by: Monday 7 September 2026 AoE (12:59 BST, Tuesday 8 September) Stage 2 project applications and written task due: Sunday 13 September 2026 AoE (12:59 BST, Monday 14 September) Stage 3 interview invitations issued by: end of day Monday 21 September 2026 (some earlier Interviews held: from mid-week commencing 21 September, into the week commencing 28 September 2026) Offers made: week commencing 5 October 2026 Our Winter 2026 Taskforce cohort ran 2 Feb - 24 April 2026 (12 weeks).

Our projects & experts for this cohort were: Sean McGregor (AI Incident Database) | Research Team Leader: Isaak Mengesha A pragmatic classification of AI incident trajectories Tommy Shaffer Shane (Centre for Long-Term Resilience) | Research Team Leader: Sarah Bollinger Leveraging open source intelligence (OSINT) for loss of control AI risk Henry Papadatos (SaferAI) | Research Team Leader: Steve Barrett Stress-testing a loss of control safety case Su Cizem (The Future Society) | Research Team Leader: Tara Thwing (formerly Peter Courtney ) Advancing Dangerous Capability Red Lines Through the AI Safety Institute Network Caio Machado (The Future Society) | Research Team Leader: Francesca Gomez Developing International AI incident tracking and response infrastructure: the Severity Threshold - technical triggers for ‘unacceptable risk’ Our Autumn 2025 Taskforce cohort ran from from mid-September to 19th December 2025 (12 weeks).

The research focus area for this cohort: frontier AI risk management and incident monitoring.

Our projects and expert partners for this cohort were: Aidan Homewood (Research Scholar at the Centre for the Governance of AI ) External assurance of frontier AI Richard Mallah (Executive Director of the Center for AI Risk Management & Alignment (CARMA) : Graduated loss of control characterisation Sean McGregor (Founder of the AI Incident Database and Responsible AI Collaborative ) A phase model for AI incident emergence In our pilot cohort, Autumn 2024, we produced two research projects on Safety Cases for frontier AI: Dynamic safety cases for frontier AI Assessing confidence in frontier AI safety cases (a summary of this was published on the AI Policy Bulletin) External expert guidance was provided by Marie Buhl ( GovAI , UK AISI ) together with feedback from AISI colleagues and partners.

The pilot attracted many very strong candidates with significant expertise and experience in other domains including senior industry and academic profiles, one of whom was invited to participate directly as a Research Team Leader.

Participants joined us having completed other education and training programmes ( BlueDot Impact , ARENA , LASR Labs ) and have gone on to pursue further training (Masters programmes, GovAI seasonal fellowships , Future Impact Group , Catalyze Impact , Talos ), obtain AI governance-related roles in organisations such as SaferAI , Deloitte , ERA , further work on their own independent or collaboration projects, found a new organisation , and discuss our research in other forums ( 1 , 2 , 3 ).

If you have questions about the Taskforce, please email: Ben R Smith, Taskforce Lead: ben [at] arcadiaimpact [dot] org.