Conference Agenda

Recent years have seen multiple UN Member States express support for the proposal of a Group of Scientific and Technical Experts on Nuclear Disarmament Verification (GSTE-NDV) within the United Nations. The initiative’s proponents argue that the GSTE-NDV would provide unique and practical benefits in support of the long-term goal of nuclear disarmament and enable UN Member States to work collaboratively on nuclear disarmament verification in a multilateral setting. Following UN General Assembly (UNGA) Resolution 79/240, the UN is seeking views from member states on the possible merits, objectives, mandate and modalities of a GSTE-NDV, through written submissions as well as a series of informal consultations. The results of this process will be published in September as a report by the UN Secretary General, which will be submitted to the UNGA’s eightieth session in late 2025. During this year’s session, the UNGA may also be called to vote on a resolution on establishing the GSTE-NDV.

We will discuss the origin of the concept and some historical precedents; and what value its proponents consider it can have in the context of other work on nuclear disarmament and the wider geopolitical environment. We will then outline the key questions about a GSTE-NDV, including its objectives, possible mandate and modalities, format, location and reporting structures; as well as the Group’s membership. We will also examine similarly structured institutions, such as international panels or advisory bodies, that could be examined as possible models and for lessons learned and its role in capacity-building for the development of multilateral nuclear disarmament verification capabilities. Finally, we will outline options and recommendations for these as food-for-thought.

Exploring Antineutrino-based Safeguards for Naval Propulsion Reactors

In a world with an increasingly chaotic security landscape, the danger of nuclear proliferation is rising. In recent years, a new potential proliferation concern has emerged with the planned employment of nuclear propulsion in submarines by Non-Nuclear Weapon States under the NPT. A prominent example of this is the sale of conventionally armed, nuclear powered attack submarines by the US and the UK to Australia under the AUKUS agreement, effectively placing significant quantities (SQs) of special nuclear material outside of traditional safeguards regimes due to their military nature. Here, it is important to give safeguards inspectors powerful tools for a comprehensive and reliable safeguards regime, capable of detecting a potential diversion of weapons-grade nuclear material.

Here, we present research on the development and the detailed simulation of an antineutrino detector for the monitoring of a nuclear-powered submarine anchored in the base of a consenting host state. Nuclear reactors are the biggest manmade sources of antineutrinos, which are produced by the beta decay of the fission products. Their spectral distribution is directly dependent on the amount and type of nuclear material present in the reactor. Although they rarely interact with other matter, and thus are difficult to detect, this is offset by the huge number of antineutrinos produced in the reactor. Their detection via the inverse beta decay, whereby an incident antineutrino turns a proton into a neutron, thus enables stand-off monitoring without revealing sensitive information about the reactor’s design. To this aim, we will simulate a naval reactor with OpenMC, and the subsequent detection of its antineutrino emission a few tens of meters away using Geant4. The goal is to produce a model of a compact, tonne-scale detector suitable for use at a naval base of a consenting host state in the context of non-proliferation safeguards. While focusing on the type of nuclear reactors relevant for the AUKUS-class submarines, we will explore the plausible parameter space of nuclear reactors typically employed in other submarines in terms of enrichment fraction (focusing on HEU fuel), reactor power (from tens to a few hundred MWth), and burnup. Finally, we will explore deployment concepts and possible measurement and verification regimes to implement this approach in the context of safeguards

With the suspension of New START – the last standing U.S.–Russia nuclear arms control treaty, and prospects for its extension after expiry in 2026 uncertain, insights into nuclear deployments available from on-site inspections are lost. Moreover, the poor quality of U.S.–Russia and U.S.–China relations renders prospects for nuclear weapons agreements in the foreseeable future low. The consequences are reduced insights into nuclear deployments, increased worst-case assumptions and risks of nuclear weapons use, and loss of opportunities for creating shared understandings and forging mutual trust. Before this backdrop, it is expected that satellite remote sensing will be a critical tool for assessing nuclear forces and postures. Today, there is abundant satellite data generated from diverse imaging modalities and over multiple frequency bands with high temporal resolution available for perusal by governments and civil society. In view of the sheer size and diversity of datasets, data fusion and data analysis will increasingly be facilitated by machine learning (ML) / artificial intelligence (AI) techniques. This article takes as its entry point the urgency of conceptualizing satellite remote sensing, potentially involving the use of ML/AI, as a crucial component of future nuclear arms control founded on and promoting trust. However, the current data ecosystem exhibits factors that could impact the trustworthiness of compliance assessments based on the data workflows. Those factors are disparate and diffuse data sources and diverse actors. This article explores those factors in two steps. First, a case study in the use of ML/AI and satellite imagery is considered, demonstrating baselining and subsequent monitoring of activities at a chosen nuclear weapons deployment, production, storage, or dismantlement site. The case study is used to give a mixed quantitative–qualitative account of trustworthiness-impacting pressure points as encountered in data access, processing, and analysis, including from the potential use of synthetic imagery for model training. Second, the article looks at adjacent treaty regimes where the factors – disparate and diffuse data sources and diverse actors – have played a non-trivial role: the International Atomic Energy Agency safeguards regime and Comprehensive Test Ban Treaty Organization’s International Monitoring System and International Data Center. Technical reports and documented member state views articulated in the respective international forums are analyzed to draw relevant lessons for the arms control context. The article concludes by discussing the implications of the conducted two-step analysis for the prospects of future arms control as well as implications for the role – negative or positive – of ML/AI-facilitated satellite remote sensing in a future treaty verification regime, while fully cognizant that the certainty requirements for risk reduction and threat management differ from those for treaty verification or recently proposed “demonstrative verification.”

The presented research explores the novel biosecurity threats arising from the growing convergence of large language models (LLMs), biological design tools (BDTs) ), and automation technologies—including robotics— in the life sciences. A critical distinction is drawn between the varied risk profiles of LLMs and BDTs, underscoring how LLMs have the potential to democratize access to dual-use knowledge and reduce technical barriers to biological misuse. In contrast, BDTs could enable technologically sophisticated actors to design potential pandemic pathogens (PPP) or sophisticated biological weaponry. In addition, we assess the potential of AI-tools and automation technologies to diminish the barriers to entry for non-experts in bioengineering and analyze its implications for the design-build-test-learn cycle and its enabling capacity. In the realm of potential bioweapons, the analysis shows that these AI systems accelerate the entire development process - from knowledge access to experimental design to experimental optimization - while breaking down tacit knowledge barriers. As a result, previously experience-based expert knowledge is becoming increasingly technologizable and thus potentially usable by non-state, malicious actors. To prepare for a future in which AI can autonomously conduct tasks such as gene design, pathogen modeling, experimental planning, and execution—via self-driving laboratories and autonomous robotics equipped with decision-making logic—it is essential to critically assess the implications of such advancements. We therefore highlight the challenges of balancing innovation with responsible governance and propose possible pathways to mitigate risk. By elucidating these new frontiers in biosecurity threats, we aim to inform proactive policy-making and responsible innovation in the rapidly evolving landscape of AI-enabled biotechnology.