Durham students and academics for reducing catastrophic risks from advanced AI
Whatever your expertise, you can contribute to the conversation and make a difference.
Proudly supported by
What is AI Safety?
“AI Safety” means keeping artificial intelligence from causing harm. We focus on the most catastrophic risks, up to and including human extinction, because we believe AI capabilities are advancing fast enough to make them real. Find out more.
Our Events
Loading calendar...
We'll update this calendar with upcoming events. You can add it to yours:
Click on and RSVP to any event to receive email updates for that event.
What we do
AIS101
Regular sessions exploring core AI safety concepts in an accessible, interactive format.
Learn More
In-Depth Reading Groups
Deeper discussion on AI safety papers with PhD students specialising in the areas. Open to anyone — aimed at those wanting to get into the nitty-gritty technical details of alignment or policy.
Learn More
Research Project Support
We support you to pursue AI safety research. We can support applications for fellowships and funding, and advise students on dissertation and independent study projects.
Learn More
Speaker Events
Insightful talks by leading researchers and industry professionals on AI safety and the latest research.
Learn More
Social Events
Relaxed socials to connect and build community.
Learn More
Free pizza after almost every event
Research
Sparse Autoencoders Can Learn Graded Latents for Relational Composition
Farrell, T., Leask, P., Al Moubayed, N.
Ghost Marks in the Machine: A Critical Review of SynthID for Code Provenance Monitoring
Sherratt-Cross, E., Farrell, T., Ogden, S., Ryley, O.
Probing by Analogy: Decomposing Probes into Activations for Better Interpretability and Inter-Model Generalization
Leask, P., Al Moubayed, N.
Order by Scale: Relative-Magnitude Relational Composition in Attention-Only Transformers
Farrell, T., Leask, P., Al Moubayed, N.
