HackTalk: Jonas Sandbrink - Verified Researcher Credentials for Bio-Capable AI Models

​HackTalk: Jonas Sandbrink

​Verified Researcher Credentials for Bio-Capable AI Models

​Pre-hackathon talk. Free and open. Recording available after.

​Bio-capable AI models are getting better fast. The field needs a way to give verified researchers access without handing the same keys to everyone. Jonas is working on that problem.

​Jonas Sandbrink is Entrepreneur-in-Residence at Sentinel Bio, where he works on researcher credentialing for access to bio-capable AI systems. Previously at the UK AI Security Institute, where he founded the chem/bio evaluations team, the first government program anywhere to assess whether frontier AI models could enable biological or chemical misuse. Before AISI, he advised the UK Cabinet Office and Google DeepMind on biosecurity. He holds a DPhil from Oxford's Nuffield Department of Medicine, where his research examined misuse risks from synthetic virology and AI.

​He'll share what the verification stack looks like in practice, where access control fits in the AIxBio threat model, and which parts of the problem are open for researchers and builders to tackle.

​Format: 10 min talk + 20 min Q&A and discussion.

​Part of the AIxBio Hackathon pre-event programming.