Connor Dunlop

Connor is Director of Policy & Strategy on the founding team of Lucid Computing, which builds hardware-rooted verification for AI. Previously he set up and led the EU and global governance programme at the Ada Lovelace Institute, where his research informed the International AI Safety Report, the EU Code of Practice on General-Purpose AI and the UK Government’s AI white paper. He supervises fellows at GovAI, sits on the OECD AI Expert Network, and has been featured in Al Jazeera, Euractiv and TechPolicy.Press. He has published a book chapter on general purpose AI models and systemic risk with Bloomsbury.

Projects

Research Direction: Identifying critical AI verification claims and designing solutions for them

Identifying and operationalising highest priority claims for pacing the frontier of AI, along with technical feasibility assessments for each

day to day - they would work on identifying the technical, legal, personnel prerequisites to 'pace the frontier'. Ideally when answers aren't available online, they should speak to people who may have good insights (e.g. at eval orgs, auditors, or companies)

The output could be a paper, a strategy doc (e.g. menu of options to verify certain claims under certain timeframes, or e.g. a 1 day, 1 week, 1 month strategy for a 'scramble' response to an AI accident), or several Substack posts (e.g. which externalises technical findings from Lucid's 18 month datacenter retrofit project in Sweden, which sees our verification designs built and red teamed by world class experts. The goal is to publish lab notes, experiments, early findings, new avenues for research as we run several cycles of building and breaking)

This matters because companies and governments are calling for pacing, but no one knows what that means in practice. without knowing the claims we want to verify, it is hard to design technical solutions. this reduces the chance the tech is ready before the time is needed.

the project is very much open to being shaped

What I'm looking for in a Mentee

Someone who has experience translating technical papers into policy facing outputs would be great, and even better if they have experience in doing this for AI verification specifically. If not familiar with AI verification, I'd like someone who is excited to get up to speed and document their reading (i.e. doing a good lit review). The ideal person would have the ability to be able to answer 'why does this matter' for someone at a lab or in government. I like mentees who are proactive and who set the agenda for how we will do impactful work together.

What I'm Like as a Mentor

I like to adapt to the mentee, so I usually let them decide how I can be most useful for them and try to structure the relationship around that. I appreciate if they bring an agenda to meetings, but I am happy to do this if they wish. I am usually most reachable on Signal for ad hoc comms. I like the mentee to set deadlines for when feedback is needed.