Postdoctoral Research Associate/Fellow (technical AI safety)

About us

The Department of Informatics at King’s College London is looking to appoint a Postdoctoral research associate or fellow to work in the area of technical AI safety. 

About the role

The position is funded by the Open Philanthropy grant “Verifiably Robust Conformal Probes”. The project’s goal is to develop methods for latent probing (aka activation monitoring) of large language models (LLMs) that leverage certification and conformal prediction techniques to offer probabilistic and adversarial robustness guarantees. Applications include detection of misaligned LLM intentions such as deception, harmfulness, jailbreaking, and power-seeking behaviours. 

You will work alongside Dr Nicola Paoletti and a core team that includes a PhD student from Dr Paoletti’s lab and Prof Osvaldo Simeone from Northeastern University London. You will join a vibrant research environment, with several researchers in Dr Paoletti’s lab working on related topics, within a department that has a strong track record in AI research. 

Applicants for the associate position should have already completed, or be close to completing, a PhD in a relevant discipline. Applicants for the fellow position should also have prior post-doc experience. Commensurate with their career stage, applicants should have an outstanding research track record in topics related to the project’s scope, such as mechanistic interpretability of LLMs, robustness verification of machine learning models, and conformal inference. 

Applicants should demonstrate scientific creativity, research independence, the capacity to support junior team members, and strong communication skills, both written and verbal. 

This is a full time post (35 hours per week), and you will be offered an a fixed term contract until 31 October 2027 and starting as soon as possible. 

Research staff at King’s are entitled to at least 10 days per year (pro-rata) for professional development. 

We expect to hold interviews in late March 2026.

Advert information