Chris Stamper

Projects

Research Direction: Frontier Biological Risks Research

Mirror Life

A literature review of available achiral carbon and nitrogen sources in various environments, animals, and plants combined with modeling to give rough estimates of how fast mirror life could spread and what resources it could utilize.

• Bio or math background could be helpful

• Combines literature deep dives and some modeling work

• Could likely feed into a larger project and potential publication

An examination which barriers on the path to mirror life could be lowered by AI tools, either one or some combination of LLMs, BioFMs, or AI-BTs

• Bio or SynBio experience could be helpful

• Literature review that blends synthetic biology with AI risk research

• Could focus on top-down or bottom-up pathway or some combination of both

• Output depends on findings but could be infohazardy (or not)

AIxBio

(Infohazardy) Mapping out the framework of an eval / making a toy eval to explore combining frontier LLMs and open-weight models (potentially abliterated or uncensored) and the uplift the combo could provide

• Python experience important

• Can open-weight models give answers or plans a closed-source model will not.

• Can frontier, closed LLMs then help with related lab tasks framed as benign

• Can the OW models help with benign framing, jailbreaks, etc (either with the human intermediate or automated)

• Chance for public output likely limited

Applying mech interp techniques to BioFMs - either protein or DNA FMs.

• Python experience or past experience in mech interp helpful

• This could take a few directions - understanding what is firing when a model starts to produce a pathogenic / dangerous sequence or mapping internals back to biological pathogenic features

• Can this knowledge be used to be safeguards and classifiers around these models etc

• Depending on direction and focus there is a chance for public output

Building a test DNA or Protein sequence based classifier

• Python experience or Bio experience with some python

• Explore the possibility of building a classifier to wrap around LLMs that would only operate when a lighter weight input/output classifier sees a biological sequence

• This classifier would aim to ID sequences of concern, basically synthesis screening but on the model end

• This project could also take a less tech heavy and higher level approach to explore the feasibility of just adding a synthesis screening step with existing or upcoming functional based screening tools to LLMs.

Detection

Explore the feasibility and tractability of point of collection based DNA sequencing

• Various backgrounds could be valuable here

• What tech exists or is required to build small, durable 'labs in a box' that could periodically sequence in real time and then transmit data for analysis.

• How would this look building this into a wastewater plant? Attaching one to the waste collection trucks at airports? Using it for sequencing air in major transit hubs?

• Could solid-state nanopore sequencers be useful here? (Bonus angle, if these nanopores are achiral what is needed on the software side to add mirror DNA detection?)

What I'm looking for in a Mentee

This is somewhat project dependent but fellows with more technical backgrounds, whether in something biological or computational, would likely be suited best to many of the projects. Several projects would require familiarity and comfort with python while others could be structured as more high level literature reviews. Based on my mentorship style I think high agency fellows who would like some amount of independence would be a good pairing. I support when fellows are creative and a bit bold and want to have ownership of the project.

What I'm like as a Mentor

I tend to be somewhat hands-off and feel research flourishes when researchers are given time to think, explore, and attempt various approaches. However, I am also happy to be more hands-on and give feedback more often if a fellow prefers and am often quite available async through whatever method is preferable (email, slack, signal, etc). One hour weekly check in calls are good to go over overarching goals and project progress and to plan actionable next steps and identify any key blockers. My communication style tends towards warm, kind, friendly and supportive — even when offering more critical feedback.

Bio

Chris Stamper is an independent biosecurity researcher funded by Coefficient Giving, working on risks at the intersection of synthetic biology and frontier AI systems. He holds a PhD in Immunology from the University of Chicago and completed a postdoctoral fellowship at Karolinska Institutet, his academic work has bridged wet-lab virology and immunology with computational biology and machine learning. His biosecurity contributions include work on vaccines, rapid MCMs, LLM evaluations, mirror life, and various other projects while an RM with the ERA AIxBio Fellowship.