ERA:AI Fellowship Winter 2027

About the role

The Cambridge ERA:AI Fellowship is a 10-week research and entrepreneurship stay in Cambridge, UK, from 18 January 2027 to 26 March 2027. Fellows work on understanding and reducing catastrophic risk from frontier AI. Each person runs an individual project with close mentorship and career support, then uses ERA's network afterward. Winter 2027 takes people on three tracks only: AI Governance, Technical AI Governance, and Technical AI Safety. A separate AIxBiosecurity fellowship is planned for the same dates and is not this form.

Key facts

What you'll do

• Lead one project for the full 10 weeks, aimed at research, founding work, or policy, not a set of class assignments.
• Pick AI Governance when the work is about rules, institutions, and how labs and governments should be held to account.
• Pick Technical AI Governance when the work is about making those rules checkable: measurement, compute, audits, and verification.
• Pick Technical AI Safety when the work is about how models behave, how they fail, and how they stay under human control.
• Read ERA's Past Research and AI Risk pages before you apply, so the project matches the kind of work they fund.
• Use weekly mentorship and career support during the stay, then the wider ERA network once you leave, with the stated aim of later work as a researcher, founder, or policymaker.
• Sit a Q&A session if you have process questions. Registration is on the official pages, not invented here.

Requirements

• Write the application yourself. Short essays, about two hours. Argue the substance; polish is secondary.
• Do not ask another person to draft or edit the essays. Do not use AI to generate or refine the text. ERA says any outside help can get the file dropped.
• Apply even if the CV does not look like a perfect match. The list of welcome backgrounds is open, not a closed degree filter.
• Submit before 13 September 2026, 11:59pm Anywhere on Earth. Files are read as they come in. People who pass the written stage get two interviews.
• Be ready to be in Cambridge for the 10 weeks unless a later offer talk opens a remote exception.
• If a visa is needed: one blank passport page for the vignette, and validity through at least 26 September 2027. Start a renewal now if that is not true. ERA can sponsor; the passport constraint is on the applicant.
• Be 18 or older.

Nice to have

• Early-career research, a current PhD, or professional work in law, engineering, computer science, or a neighbouring field.
• A project idea that sits on catastrophic risk from frontier systems, including work that mixes technical and policy questions.
• Time spent on ERA's applicant guidance from the last cycle.

Skills & tools

• AI Governance: regulatory design, lab accountability, coordination across governments.
• Technical AI Governance: compute, auditing, monitoring, hardware-enabled checks, standards a regulator could use.
• Technical AI Safety: evaluations, interpretability, robustness, alignment, control.

Practical notes

Housing is not supplied or guaranteed. Part of the £10,000 stipend is meant for rent, and ERA helps people look for rooms together in Cambridge. Travel and visa costs are supported; working-hour meals are covered on the official site. Mentors are matched after offers. Confirm dates, the Q&A signup, and the live form on ERA's apply page.

Good to know

A 10-week research stay is a sprint with one artifact at the end, not a taught degree. Governance tracks usually turn a risk into a rule someone could follow. Technical tracks usually measure or constrain a model. Founder-shaped projects still need a written case another person can stress-test. Application essays in this field are scored on how you reason, including what you do not know. Integrity rules that ban AI-written answers exist because the first filter is whether the thinking is yours. Visa pages and passport expiry are mechanical gates: missing them late is a self-inflicted reject. Reading prior work from a lab before you apply is cheaper than guessing the topic they fund.