Adam Harrison

Email: A.Harrison-15@sms.ed.ac.uk

LinkedIn: www.linkedin.com/in/adam-harrison-ai

Research keywords: Open-weight LLMs, tamper-resistant safeguards, SLMs

Bio:

Adam grew up in Poland before moving to Scotland to study. He completed a BSc in Computer Science and Mathematics at the University of St Andrews.
He then completed a part-time MSc in Artificial Intelligence at the University of Edinburgh. His dissertation explored how to make open-weight LLMs more resistant to harmful fine-tuning. Alongside studying he worked as a research assistant for the University of Southampton where he explored the trade-off between privacy and explainability in machine learning systems.

He has been leading the AI Safety Hub Edinburgh (AISHED) since January 2025, teaching members about risks from AI and discussing potential solutions to mitigate them. In AISHED, he has run discussion groups, AI Safety courses and paper replications.

PhD research:

Frontier AI models are developed and controlled by a handful of powerful companies. Since these LLMs closed source, these companies have significant control over their use and distribution. This gives these companies unprecedented power and poses risks to privacy, democracy and sovereignty.
Open-weight models provide an alternative, but come with significant risks, due to the complete lack of control over their distribution and modification.
Adam intends to mitigate these risks to make open-weight models a viable alternative to closed source models. In particular, he is interested in developing tamper-resistant safeguards to reduce the risks of harmful fine-tuning and in improving systems combining small language models and large language models in privacy preserving ways.

Supervisors: Ivan Titov