Work in progressAgentsMultimodal

Projects I'm working on right now.

Ongoing research that hasn't been published yet. Details and results will be shared once the work is out. Happy to talk about any of it: get in touch. Published work is on the home page.

• In progressSince Sep 2026University of Edinburgh

  Safety of LLM Agent Harnesses

  Evaluating the safety of language models deployed inside agent harnesses.

  AI safetyAgentsEvaluation
  Updates & citation →
• In progressSince Jun 2026University of Edinburgh

  Safety Evaluation of Computer-Use Agents

  Better safety evaluation for agents that operate computer interfaces.

  AI safetyAgentsEvaluation
  Updates & citation →
• In progress2026SPAR · Co-mentor

  Multimodal Safety & Interpretability in Encoder-Free VLMs

  Safety and interpretability of unified, encoder-free vision-language models. A SPAR project I co-mentor.

  Multimodal safetyInterpretabilityMentoring
  Updates & citation →