Leaders wrestle with a potent mix: AI and weapons of mass destruction

BERLIN — Emerging technologies have radically reshaped the arms control landscape and pose a set of major challenges, though also some opportunities in curbing the spread of weapons of mass destruction, said representatives of five major UN-adjacent disarmament agencies.

Speaking Oct. 25 on the sidelines of the United Nations General Assembly’s First Committee meeting – the UN’s top disarmament body – representatives discussed how the emergence of artificial intelligence, accessible drones, new reactor technologies and others have impacted their task of controlling the proliferation of dangerous weapons and materials.

For example, the advent of widely accessible large language models such as ChatGPT may make it easier for terrorists or rogue states to access instructions for making chemical weapons , said Hong Li of the Organization for the Prohibition of Chemical Weapons.

Based in the Hague, Netherlands, the OPCW implements the nearly universally ratified Chemical Weapons Convention and has overseen the destruction of all declared stockpiles of chemical weapons across 193 countries – an effort for which it received the Nobel Peace Prize in 2013. Only three UN countries – Egypt, North Korea and South Sudan – have not signed the convention.

It is “critical” to track the changes in delivery systems, Li said. Inexpensive drones, particularly those built for agriculture with tanks for liquids and sprayers attached to them, “can be easily adjusted for the delivery of chemical weapons, which brings new challenges for us,” he added.

To keep up with the times, OPCW officials created a temporary working group on artificial intelligence, which will begin work in 2025 for two years. It will systematically evaluate this emerging technology’s impact on the world of chemical weapons while also taking into account how the organization could use it to further its goals of a chemical weapons-free world.

Meanwhile, the Vienna-based International Atomic Energy Agency is tasked with surveilling nuclear facilities globally to ensure no fissile material is diverted for the use in atomic bombs through a system collectively referred to as safeguards.

To stay ahead of the curve, the agency has a dedicated technology foresight team, said Tracy Brown, the agency’s liaison and public information officer. Its members keep tabs on new developments in the nuclear field, devising new tools and techniques to uncover illicit nuclear-weapons efforts.

In the past decade alone, the amount of nuclear material under the agency’s safeguards has increased by 25%, Brown said. With limited resources, this requires a more efficient allocation of inspectors’ time.

Machine learning has helped in the process, enabling “more efficient and effective video surveillance” of nuclear facilities, according to Brown. Computer systems can flag relevant events – such as when a cask carrying radioactive material is unexpectedly removed – and set off alarm bells, alerting humans to review the case manually.

The agency has also trained its own AI models to scour openly available information for material relevant to detecting illicit nuclear activities, Brown disclosed. Open-source data streams include news reports, scientific papers, satellite imagery and signals picked up by remote sensors, all of which would be time-consuming to mine manually.

Similarly, the Comprehensive Test Ban Treaty Organization, housed in the UN headquarters in Vienna, has been harnessing the power of machine learning to train its computer systems to use the data flowing from its global monitoring network to more quickly and accurately detect nuclear tests, a program it called NET-VISA. The CTBTO’s system of 306 stations around the globe was crucial in detecting and confirming North Korean nuclear tests from 2006 onward and in dispelling rumors about a possible Iranian test when earthquakes were recorded in the country’s heartland earlier this month.

NET-VISA will be made available to the treaty’s state parties to enhance their national abilities as well, said Jose Rosenberg, senior liaison officer at the organization.

“We are living in an age of accelerated technological change,” said Izumi Nakamitsu, the UN’s High Representative for Disarmament Affairs, who leads the organization’s Office for Disarmament Affairs. “This is also a time of heightened danger due to fraught and changing security environments.”

The convergence of technologies such as AI and 3D printing of biotechnology and nanotechnology can lower the barriers for terrorists or rogue states to gain access to weapons of mass destruction, she said.

“We need to adapt the existing nonproliferation and disarmament regime to the ever-evolving security landscape.”

Linus Höller is Defense News' Europe correspondent and OSINT investigator. He reports on the arms deals, sanctions, and geopolitics shaping Europe and the world. He holds master’s degrees in WMD nonproliferation, terrorism studies, and international relations, and works in four languages: English, German, Russian, and Spanish.

柏林——五个主要联合国附属裁军机构的代表表示，新兴技术已经彻底改变了军控格局，并带来了一系列重大挑战，但也带来了一些遏制大规模杀伤性武器扩散的机会。

10 月 25 日，在联合国大会第一委员会（联合国最高裁军机构）会议间隙，代表们讨论了人工智能、无人机、新型反应堆技术等的出现如何影响了他们控制危险武器和材料扩散的任务。

例如，禁止化学武器组织的洪利表示，像 ChatGPT 这样广泛普及的大型语言模型的出现，可能会让恐怖分子或流氓国家更容易获得制造化学武器的指令。

总部位于荷兰海牙的禁止化学武器组织（OPCW）负责执行几乎得到普遍批准的《化学武器公约》，并监督销毁了193个国家所有已申报的化学武器库存——为此，该组织于2013年荣获诺贝尔和平奖。只有三个联合国成员国——埃及、朝鲜和南苏丹——尚未签署该公约。

李表示，追踪投送系统的变化“至关重要”。他补充说，价格低廉的无人机，特别是那些为农业用途而设计、配备液体罐和喷雾器的无人机，“很容易改装用于投放化学武器，这给我们带来了新的挑战”。

为了与时俱进，禁止化学武器组织官员成立了一个人工智能临时工作组，该工作组将于2025年开始工作，为期两年。它将系统地评估这项新兴技术对化学武器领域的影响，同时也会考虑该组织如何利用这项技术来推进其实现无化学武器世界的目标。

与此同时，总部位于维也纳的国际原子能机构负责监督全球核设施，以确保没有裂变材料被挪用于制造原子弹，这一过程通过一个统称为保障监督的系统来实现。

为了保持领先地位，该机构设立了一支专门的技术前瞻团队，该机构联络官兼公共信息官特雷西·布朗表示。团队成员密切关注核领域的新发展，并设计新的工具和技术来揭露非法核武器活动。

布朗表示，仅在过去十年间，该机构监管下的核材料数量就增加了25%。在资源有限的情况下，这就需要更有效地分配检查人员的时间。

布朗表示，机器学习在这一过程中发挥了重要作用，使得核设施的视频监控“更加高效便捷”。计算机系统可以标记相关事件——例如，当装有放射性物质的容器被意外移走时——并发出警报，提醒工作人员进行人工审查。

布朗透露，该机构还训练了自己的AI模型，用于搜索公开信息，寻找与探测非法核活动相关的资料。开源数据流包括新闻报道、科学论文、卫星图像和遥感器接收到的信号，所有这些信息如果手动挖掘将非常耗时。

同样，设在维也纳联合国总部的全面禁止核试验条约组织（CTBTO）也一直在利用机器学习技术训练其计算机系统，使其能够使用来自全球监测网络的数据，从而更快、更准确地探测核试验，该项目被称为NET-VISA。CTBTO在全球设有306个监测站，自2006年以来，该系统在探测和确认朝鲜核试验方面发挥了至关重要的作用，并且在本月初伊朗中部地区发生地震时，该系统也帮助辟谣了有关伊朗可能进行核试验的传言。

该组织高级联络官何塞·罗森伯格表示，NET-VISA也将向条约缔约国开放，以增强其国家能力。

联合国裁军事务高级代表、联合国裁军事务厅厅长中满泉表示：“我们生活在一个技术变革加速的时代。同时，由于安全环境充满挑战且瞬息万变，这也是一个危险加剧的时代。”

她表示，人工智能、生物技术和纳米技术3D打印等技术的融合，可以降低恐怖分子或流氓国家获得大规模杀伤性武器的门槛。

“我们需要使现有的防扩散和裁军机制适应不断变化的安全形势。”

林努斯·霍勒是《防务新闻》的欧洲记者和开源情报调查员。他报道影响欧洲乃至全球的军火交易、制裁和地缘政治。他拥有大规模杀伤性武器不扩散、恐怖主义研究和国际关系三个硕士学位，并精通英语、德语、俄语和西班牙语四种语言。