grantmaking.ai Launch Round
Please find our budget here.
The problem / why now
AI is making it easier to engineer biology. Frontier models now outperform expert virologists at troubleshooting wet-lab experiments (virologytest.ai). Open-source tools can predict which mutations let a pathogen escape existing immunity (evescape.org). As models become more biologically capable, the number of bad actors able to cause harm with biology will rise dramatically.
Fortunately, anyone designing a threat must still gain access to biological material and equipment. Experts see securing access to synthetic DNA as the most promising chokepoint to defend. This idea has widespread support from the CEOs of frontier AI labs, the DNA synthesis industry, broader biotech, and the policy community, who have recently signed an open letter calling for mandatory customer screening (screendna.org). Regulation has been proposed in the US (S.3741) and the EU (Biotech Act).
Progress so far
safely.bio is building Verify, software that runs know-your-customer checks on the biologists placing DNA orders ahead of this regulation. Our software identifies whether they are legitimate researchers or bad actors. We recently ran a study showing Verify outperforms the current state-of-the-art AI customer screening tool; the preprint will be released shortly.
We’re two full-time co-founders. Phil Palmer holds a Cambridge PhD in computational biology, worked on biosecurity safeguards at CLTR and the Nuclear Threat Initiative, and was a visiting researcher in Kevin Esvelt's group at MIT. Joey Bream is a third-time founder and software engineer who worked on customer screening with the former White House NSC Director for Biodefense and Pandemic Preparedness at Sentinel Bio. We are three months in, incorporated as a UK Ltd, have spoken to 30+ potential customers, and are currently sprinting to sign design-partner pilots with DNA synthesis companies.
What we'll do
Concrete output over the next four months:
Preprint published showing Verify outperforms the current state-of-the-art AI customer screening tool
Pilots signed with 3+ DNA synthesis providers
Regulatory engagement in the UK and EU ahead of incoming mandatory KYC requirements (EU Biotech Act, US S.3741)
We think that biorisk is the most immediate x-risk posed by frontier AI.
Our theory of change is as follows:
We have already built the current state-of-the-art AI customer screening tool
DNA synthesis providers adopt our tool
We are actively seeking pilot programmes with DNA synthesis companies
Regulation will likely mean that 90%+ of DNA synthesis companies screen their customers in regulated jurisdictions in the next two years
Note: we do not need 100% adoption for screening to be effective because widespread screening is itself a deterrent for bad actors seeking to cause harm with biology
Co-founder
Please find our budget here.
Private comment. Only shown to approved funders and grant reviewers.
I endorsed this grant proposal because:
I recommend funding this the maximum $50k. Even if Phil can get cG funding soon, every week counts and biosec is an urgent priority.
Also, it's possible that cG won't be funding DNA synthesis screening?
Private comment. Only shown to approved funders and grant reviewers.
Awesome!
We previously made some impact estimates, which can be found here. We found that even with conservative assumptions (e.g., ignoring x-risks), this project could save thousands of lives per year in expectation.
Private comment. Only shown to approved funders and grant reviewers.