Research Fellow (111504-0426)

Location: University of Warwick Campus, Coventry

Duration: Fixed-term contract until 31 December 2028.

About the Role  

Informal Queries

• For informal queries, please contact Professor Thomas Hills at T.T.Hills@warwick.ac.uk.

Flexible Working

• We will consider applications for employment on a part-time or other flexible working basis (e.g. job share), despite the position being advertised as full-time.

Project Title

• “KnoWare: Assessing Risks and Emergent Capabilities in General-Purpose AI”

What scientific question will you investigate? Outline of project aims:

• Recent advances in foundation models (e.g. large language models and multimodal AI systems) have led to rapidly evolving capabilities, often exhibiting emergent behaviours that are difficult to predict, interpret, and control. Current evaluation approaches largely focus on observable performance, leaving significant gaps in understanding how capabilities arise and how they relate to systemic risks.

This project investigates a central research challenge: how can we proactively identify, test, and evaluate emergent capabilities and risks in General-Purpose AI (GPAI) systems before they fully manifest?

The University of Warwick’s contribution to the KnoWare project focuses on developing adversarial prompting, robustness testing, and evaluation methodologies to probe AI systems for emergent capacities, which may be either positive or negative. The goal is to create systematic approaches for identifying emergent behaviors that are not revealed by established benchmarks, but may be revealed through principled prompt design informed by behavioural science and decision making research.

To address this, the research will:

• Design and implement adversarial and stress-testing protocols for foundation models
• Analyse model behaviour across diverse inputs and contexts, including edge cases and high-risk scenarios
• Contribute to the development of capability–risk assessment frameworks and benchmarking methodologies
• Support the integration of technical evaluation methods with broader interdisciplinary approaches, including regulatory and societal considerations

The outcomes will contribute to a new generation of proactive, interpretable, and policy-relevant AI evaluation tools, aligned with emerging frameworks such as the EU AI Act.

The post allows for hybrid working.

About You  

We are looking for a candidate with expertise in human-AI interaction and AI model evaluation. Experience working with foundation models (e.g. large language models) and their evaluation, testing, or analysis is highly desirable.

The successful candidate should have:

• Strong programming skills (e.g. Python) and experience with modern AI systems
• Experience in designing and analysing experiments or evaluation pipelines for AI systems
• Knowledge of model robustness, bias, interpretability, or AI safety methods
• Strong quantitative and data analysis skills

An interest in responsible and trustworthy AI, including fairness, transparency, and regulatory alignment, is essential.

For details on the experience and skills required, please refer to the job description as a PDF below.

JD - Research Fellow (111504).pdf – 105KB

PhD Status

If you are near submission of your PhD, or have not yet had it conferred, any offers of employment will be made at Research Assistant level, at the highest spinal point of pay grade 5 (£34,610 per annum).

Upon receipt of evidence confirming the successful award of your PhD, you will be promoted to Research Fellow, at the lowest spinal point of grade 6 (£35,608 per annum).

CLOSING DEADLINE: Monday 4 May 2026 at 11:55pm (UK Time)

Full details of the duties and selection criteria for this role can be found in the vacancy advert on the University of Warwick's jobs pages. You will be routed to this when you click on the 'Apply' button.

Advert information