Research Associate in Detecting Unverbalized Dangerous Capabilities in Frontier Models

Overview

Imperial College London is offering a fully funded, fixed-term position for a Postdoctoral Research Associate within the Department of Computing. The role takes place in the Safety and Alignment lab, led by Dr. Oana-Maria Camburu, and focuses on the project detecting unverbalized dangerous capabilities in frontier models. The appointee will work on identifying dangerous capabilities that large language models do not verbalize in their Chain-of-Thoughts, investigating techniques to monitor and defend against model obfuscation.

What you get

The position offers a salary ranging from £50,733 to £59,484 per annum, alongside a remuneration package that includes 43 days off per year and pension schemes. Candidates who have not yet officially been awarded their PhD will be appointed as a Research Assistant with a salary range of £45,399 to £48,876 per annum. Employees gain access to career support for researchers, promotion pathways, and staff resources within a collaborative work culture.

Requirements

Applicants must hold, or expect to obtain shortly, a PhD in Computer Science, Mathematics, Physics, or a closely related discipline, or possess equivalent research or industrial experience. Essential qualifications include practical research environment experience, publications in refereed conferences and journals, and familiarity with model checking, monitoring, and Chain-of-Thoughts. Candidates must also demonstrate research experience with frontier models and coding proficiency using deep learning libraries such as PyTorch.

How to apply

Applicants should visit the employer jobs portal and search for vacancy number ENG04085. Submissions must include a full curriculum vitae featuring a publication list, as well as a research statement outlining relevant expertise and interesting research issues connected to the post.

Key dates

The application closing date is 22nd November 2026 at midnight. The appointment is for a fixed term of 12 months with a start date as soon as possible and before 8th April 2027.