Beyond the Ethics Board: Using AI to Identify Dual-Use Research of Concern Before It Causes Harm

In 2011, two independent research groups — one led by Ron Fouchier at Erasmus University and one by Yoshihiro Kawaoka at the University of Wisconsin — announced that they had engineered H5N1 avian influenza to become transmissible between ferrets via respiratory droplets. The implications were immediate and profound: for the first time, a pathogen with H5N1's lethality (approximately 60% case fatality rate in humans) had been shown to be capable of airborne transmission in a mammalian model. The US National Science Advisory Board for Biosecurity (NSABB) initially recommended against full publication of the methodology, triggering a global debate about the boundaries of dual-use research that continues to this day.

That debate crystallised into policy. In 2012, the US Department of Health and Human Services and the US Department of Agriculture jointly published the first formal framework for identifying and overseeing Dual-Use Research of Concern (DURC) — research that, while conducted for legitimate scientific purposes, could be misused to pose a significant threat to public health, safety, security, or national security. The framework defined 15 specific categories of concern, ranging from enhanced pathogen transmissibility to the reconstitution of eradicated pathogens.

The challenge has always been implementation. Identifying whether a given piece of research falls within one or more of these 15 categories requires deep domain expertise, careful reading of methodology sections, and an understanding of how specific experimental approaches map onto biosecurity risk profiles. For institutional review boards (IRBs) and ethics committees — which are typically staffed by generalists rather than biosecurity specialists — this is a genuinely difficult task. BioScreens is building AI infrastructure to make it tractable.

The 15 DURC Categories

The US DURC policy framework identifies research that is reasonably anticipated to create, transfer, or use enhanced potential pandemic pathogens (ePPPs) or that involves any of the following 15 categories of concern:

Applying this framework to a 40-page research paper — with its methods sections, supplementary data, and implicit assumptions about experimental context — is not a task that can be reliably performed by a non-specialist reviewer in the 30 minutes typically allocated to IRB pre-review. It requires the kind of systematic, comprehensive analysis that large language models are increasingly well-suited to provide.

How BioScreens DURC Analysis Works

The BioScreens DURC Analysis tool accepts research papers, lab protocols, or grant proposals in PDF or DOCX format, as well as pasted plain text. For scanned documents, the platform uses Tesseract OCR with support for English, French, German, and Arabic, with confidence scoring per language — a feature that is particularly significant for international research institutions and for documents submitted to multilateral bodies like the Biological Weapons Convention (BWC) Implementation Support Unit.

Once the document is ingested, a large language model evaluates the content against all 15 DURC categories, identifying specific passages, methodological descriptions, and experimental outcomes that correspond to each category of concern. The output is a risk score from 0 to 100, classified as Low, Medium, High, or Critical, accompanied by a visual gauge for immediate comprehension and a detailed breakdown of the evidence supporting each identified concern.

The platform's handling of a hypothetical H5N1 transmission study illustrates the system's capabilities. Such a study receives a Critical Risk classification (87/100), with three primary concerns identified: enhanced transmissibility via gain-of-function methodology, airborne transmission potential in a mammalian model, and vaccine resistance implications from antigenic characterisation data. The platform then generates prioritised safety recommendations — in this case, escalation to BSL-4 containment, restriction of access to cleared personnel, redaction of transmission methodology before publication, and submission for institutional ethics review.

Crucially, these recommendations are not generic. They are derived from the specific content of the submitted document and ranked by priority based on the severity and specificity of the identified concerns — a level of contextual analysis that generic checklist-based DURC review cannot provide.

The Gap Between Policy and Practice

The DURC framework has been US policy since 2012, and the European Union has developed parallel guidance through the European Centre for Disease Prevention and Control (ECDC). Yet implementation remains inconsistent. A 2019 review by the Johns Hopkins Center for Health Security found that fewer than 20% of US research institutions had fully implemented DURC review procedures, and that the quality of reviews at institutions that had implemented them varied enormously.

The reasons are structural. DURC review requires biosecurity expertise that most IRB members do not have. It requires time that review boards do not have. And it requires a systematic approach to a 15-category framework that is genuinely difficult to apply consistently without decision-support tools. BioScreens addresses all three constraints simultaneously: it provides the expertise through its AI engine, it reduces the time requirement to minutes rather than hours, and it enforces systematic application of the full 15-category framework on every submission.

Implications for Global Biosecurity Governance

The significance of AI-powered DURC analysis extends beyond institutional compliance. As synthetic biology capabilities spread globally — to universities, startups, and national research programmes in countries with varying levels of biosecurity infrastructure — the ability to perform credible DURC review becomes a prerequisite for responsible scientific participation in the international research community.

For institutions in Kenya, Nigeria, South Africa, and across the developing world, BioScreens offers something genuinely new: access to biosecurity review infrastructure that was previously available only to well-resourced institutions in high-income countries. At $10/month, the platform makes credible DURC analysis accessible to any institution with an internet connection — a meaningful contribution to the goal of universal biosecurity compliance that international frameworks like the BWC have long aspired to but never achieved.

Conclusion

The H5N1 controversy of 2011 demonstrated that the line between transformative science and existential risk can be crossed inadvertently, by researchers acting in good faith, working within the norms of their discipline. The DURC framework was designed to prevent that from happening again. BioScreens is building the AI infrastructure to make that prevention systematic, scalable, and accessible — moving biosecurity review from a compliance checkbox to a genuine early-warning system for research that could cause catastrophic harm.

---

Reference: BioScreens Biosecurity Intelligence Platform — https://www.bioscreens.org/