Are Global AI-Biosecurity Safeguards Really Global?

The goal of this project is to test whether safeguards designed to prevent AI-enabled biosecurity risks are effective in a resource-constrained environment, using Nigeria as a case study to investigate vulnerabilities that could undermine global catastrophic risk.

To do this, we shall evaluate the AI safety ecosystem in Nigeria at three levels. The first level will be to look at the AI-enabled biosecurity policies and regulations in Nigeria that have been designed to prevent the misuse of AI-frontier systems or the training of new models, and also to evaluate institutional preparation against pandemic risks. Secondly, we shall evaluate if frontier models remain robust using linguistic expressions and terminology that reflect our cultural peculiarities in Nigeria. Lastly, we shall evaluate if DNA synthesis safeguards against X-risks are effectively applied to ordering services from a resource-limited setting. The central question we are asking is, "Do AI biosecurity safeguards remain robust when deployed in untested and resource-constrained settings like Nigeria?"

At the end of this project, 1. We hope to have a clear landscape assessment of AI biosecurity governance in Nigeria. To understand the laws governing AI-biosecurity policy in Nigeria. 2. A confidential assessment of how consistently relevant DNA synthesis service providers apply safeguards for research requests originating from a resource-constrained setting. We shall use our regular research work done in our lab at Plateau State University to order oligonucleotide synthesis from various companies across the world and document how they respond to customers for KYC. Lastly, we shall make available datasets and evaluation reports of the red-teaming evaluation of the robustness of select frontier models in the Nigerian scientific, linguistic, and cultural variations. Playing out different scenarios.

Our Approach:

We shall place a random DNA synthesis request from our lab to several DNA synthesis providers across regions and see how they respond to the KYC of our requests. We hypothesise that not all regions of the world that provide DNA synthesis services value customers through KYC. We shall also look at laws and policies enacted within Nigeria to prevent misuse of LLMs. As part of our approach, we shall interview key stakeholders involved in the AI governance structure in Nigeria. There are several frontier models, and from our use, we have found that they handle queries differently. We plan to use a set of structured queries that reflect the cultural variations of our communities to evaluate these models

Our Team

Nnaemeka Nnadi, https://www.linkedin.com/in/nnaemeka-emmanuel-nnadi-a39298b5/

A microbiologist and faculty member at Plateau State University, with experience in the genomics surveillance of pathogens

Aayush Gandhi (https://www.linkedin.com/in/aayush-g-441a4a1bb/)- researcher in AI safety, aaygan29@gmail.com. Aayush builds jailbreak defenses and dual-use risk evaluations for protein and genomic design models, grounded in SIPRI/EU WMD nonproliferation training and hands-on molecular biology.

Juliet Joseph (https://www.linkedin.com/in/juliet-joseph-3332a62b/).

Is a public health analyst and IFBA-certified bio-risk manager with over 13 years of experience as a safety lead at Nigeria’s apex military facility in Nigeria.