Anthony Hughes
PhD student at the University of Sheffield, about to join Arcadia Impact as an Alignment Researcher.
about
I'm a Ph.D. student at the University of Sheffield, supported by a UKRI scholarship and supervised by Ning Ma and Nikos Aletras. I'm about to join Arcadia Impact as an Alignment Researcher. I previously interned at the Vector Institute with Gautam Kamath and Jacob Imola, on unlearning and privacy in LLMs, and related topics.
Research direction: I work on the security and privacy of machine learning, largely focused on the relationship between a model and its training data.
Recent work: My EMNLP 2025 paper examines information leakage in abstractive summarisation; my EACL 2026 paper uses activation patching and circuit discovery to understand and mitigate PII leakage in language models; and most recently I've been studying membership inference attacks against safety classifiers. As a SPAR AI safety fellow I worked with Andrew Draganov on backdoor detection in LLMs, covering both black- and white-box techniques — we've written up our findings in a LessWrong post. Separately, I worked on building better activation probes and on whether LLMs can evade them — blog post here.
publications
blog posts
On building better activation probes and whether LLMs can evade them — arguing for standardised evaluations of activation faithfulness: whether models can detect and circumvent monitors tracking their internal states.
Joint work with Andrew Draganov on whether defenders can detect implanted backdoors in LLMs and recover their trigger phrases without access to the training data.
research experience
Supervisor: Gautam Kamath. Research on certified unlearning and differential privacy for question answering in LLMs.
Mentor: Andrew Draganov. Research on detecting backdoors in LLMs — whether defenders can find implanted triggers using black- and white-box techniques, without access to the training data.
Collaborators/Mentors: Vasisht Duddu, N. Asokan. Research on mechanistic approaches to understanding PII leakage in language models.
industry experience
Led development of NLP-based products including automated text classification SaaS, data visualization tools, and integration of language models into data platforms. Built production systems serving FTSE clients.
Software Engineer working on services solving data silo issues with linked data.
Worked on a new digital platform centered around semantic web technologies.