News Briefing 4-18 March 2025

Autonomy in weapons systems, and AI in security and defence

• International: Message for the International Day for Disarmament and Non-Proliferation Awareness: 5 March, marked the International Day for Disarmament and Non-Proliferation Awareness. Amongst a number of  pressing disarmament matters,  the UN Secretary General also articulated that the ‘weaponization and the use of lethal autonomous weapons’ needs to be addressed. 
• International: Eric Schmidt Suggests Countries Could Engage in Mutual Assured AI Malfunction (MAIM): In their new paper titled “Superintelligence Strategy”, former Google CEO Eric Schmidt and Scale AI founder Alexandr Wang warn against the U.S. government creating a Manhattan Project for Artificial General Intelligence (AGI). They argue that such a project could lead to dangerous global tensions, with countries racing to develop the most powerful AI technologies for warfare.
• South Korea: South Korean defense giants bet big on AI-powered military tech: The Chosun Daily reports that ‘South Korean defense companies are ramping up investments in artificial intelligence (AI) software technology to boost unmanned warfare capabilities.’ The piece cites contracts and collaborations of Hanwha Aerospace, Korea Aerospace Industries, and HD Hyundai to develop autonomous systems for militaries. 
• Norway: The 2025 Conference on Disarmament. High Level Segment. Statement by Mr Andreas Kravik State Secretary, Ministry of Foreign Affairs, Norway: Andreas Kravik State Secretary, Ministry of Foreign Affairs, Norway, spoke at the International Conference for Disarmament and highlighted the ‘growing impatience’ among the states at the group of governmental experts on lethal autonomous weapon systems (GGE on LAWS). He stated “We must accelerate our work by agreeing in 2025 on a set of significant elements of an instrument to ensure meaningful human control over such weapon systems”.
• Russia: Russian scientists fuse reasoning models with drone-control models for thinking drones: This newsletter by Import AI features a research paper which ‘shows how one might staple together multiple different advances – basic VLA (Vision Language Action) models, general purpose reasoning models – to create new capabilities for drones.’  The paper looks at methods to train algorithms to work with drones on human recognition, symbol understanding, and reasoning to allow for greater autonomy in communication deprived circumstances. 
• US: Advancing Department of Defense Test and Evaluation for AI and Autonomous Systems: The study focuses on developing and deploying AI and autonomous systems within the Department of Defense (DoD) while addressing their unique challenges. The recommendations are divided into two main areas: research and system development and test and evaluation (T&E). ‘This study yielded findings that apply across the lifecycle of AI and autonomous systems.’  The study is available here. 

AI, algorithms and autonomy in the wider world

• International: AI red lines: the opportunities and challenges of setting limits: This article published by the World Economic Forum speaks to the need for behavioural red lines in AI systems to prevent harmful actions. These include prohibitions on unacceptable uses (like surveillance misuse) and undesirable behaviours (such as unauthorized surveillance or physical harm). Effective red lines should be clear, widely accepted, and universally applicable. Compliance mechanisms include pre-deployment certification and post-violation penalties, alongside continuous monitoring and fail-safe measures. The aim is to promote proactive AI safety to ensure that AI technologies serve society positively and avoid causing harm. 
• International: China has a new ‘completely autonomous’ AI agent. Here’s what to know about Manus: Manus AI, developed by Chinese researchers, is being touted as the world’s first general AI agent, distinct from chatbots and workflow assistants. Manus AI represents a significant step forward in AI development, showcasing the potential for truly autonomous agents capable of performing complex tasks with minimal human input. However, it still faces skepticism from some experts who see it as an advanced workflow tool rather than a complete AI agent.
• International: Israel uses AI to spy on Palestinians: The Israeli military’s Unit 8200 has developed an AI system that is said to utilise intercepted Palestinian communications to analyze intelligence and track individuals under surveillance. Trained on Arabic language data, the system has raised concerns about human rights abuses, including the criminalization of innocent people and biased decision-making. The AI has been used to monitor activities in the West Bank and target Palestinian activists. U.S. tech companies have quietly supported these efforts, providing AI tools that enhance Israeli surveillance and targeting capabilities, contributing to civilian casualties in conflict zones like Gaza and Lebanon.
• US: DOGE threat: How government data would give an AI company extraordinary power: The Department of Government Efficiency (DOGE) has gained access to sensitive U.S. government databases, raising concerns about privacy and the potential use of this data to train AI models for Musk-owned companies like xAI. The data is highly valuable for AI development, giving private companies the power to influence key sectors and public policy, posing risks to privacy and democracy.

Facial recognition, biometric identification, surveillance

• UK: Police Could Use Driving License Data for Facial Recognition in New Bill: The UK government’s Crime and Policing Bill (introduced earlier this year could provide law enforcement access to the driving license database (DVLA), this could potentially enable its use for facial recognition. Human rights group Liberty warns this could allow the police to access biometric data of millions of people, including those who have never committed a crime. Although the bill doesn’t explicitly mention facial recognition, Liberty argues that it may lead to its use. The Home Office has denied these claims, stating the provisions will not impact facial recognition technology. The bill also introduces other measures, such as prohibiting face coverings at protests and expanding police powers over surveillance and anti-social behavior.
• International: IBM stirs controversy by using Flickr photos for AI facial recognition: IBM is generally known to have used Creative Commons-licensed Flickr photos to train its AI facial recognition system, sparking concerns over privacy and ethics. While the use followed licensing rules, photographers were unaware their images had been utilised. The controversy raises questions about AI ethics, privacy, and the implications of Creative Commons licenses.
• Georgia: GYLA: MIA’s Use of Facial Recognition Cameras Against Protesters Raises Serious Privacy and Legal Concerns: The Georgian Young Lawyers’ Association (GYLA) has raised concerns about the Interior Ministry’s use of facial recognition cameras to monitor protesters, arguing it violates privacy and legal standards. They highlight the lack of proper identification procedures and the invasive nature of targeted surveillance. Despite requests for investigation, the Personal Data Protection Service has yet to provide a final decision. GYLA warns that this growing surveillance, especially in the context of protests, could have a chilling effect on freedom of expression and civil liberties.

Research and reports 

• Ukraine: Ukraine’s Future Vision and Current Capabilities for Waging AI-Enabled Autonomous Warfare: This report examines Ukraine’s advancements in ‘AI-driven unmanned systems’, which aim to reduce direct human involvement in combat while improving ‘effectiveness’. While fully autonomous warfare is a future goal, Ukraine has made significant progress in partial autonomy, particularly with aerial systems. However, ‘human oversight remains crucial, especially in engagement decisions.’