Opinion
The war already taught us what AI can't do. Now comes the hard part.
Israel does not need to win the race to build the most autonomous weapon. It needs to win the race to understand how to make autonomous systems trustworthy, and to build an ecosystem that turns this understanding into companies.
The United States just made its largest single commitment to autonomous warfare in history: $54.6 billion for the newly created Defense Autonomous Warfare Group, a 24,000 percent jump from the previous year's budget. The message from Washington is unambiguous. Autonomous AI systems are no longer the future of warfare. They are the present.
Here is the question that $54 billion cannot answer: do we actually know how to build AI systems we can trust on a battlefield? The honest answer is that we are not there yet, and Israel is better positioned than almost anyone to understand why.
We have accumulated something rare: years of operational experience deploying software in contested, high-pressure physical environments, often with limited or no GPS, while maintaining reliability under jamming, and making decisions on incomplete and/or contradictory sensor data. These constraints reveal exactly where current AI falls short. Israel has a detailed understanding of where the walls are, while most of the world is still discovering them.
The gap shows up clearly in the field. Ukraine has become the most comprehensive real-world laboratory for autonomous warfare, and while AI-assisted targeting has meaningfully improved strike precision over manual operation, reports from the frontlines have repeatedly highlighted the vulnerabilities of these systems. This is a gap many in the procurement process do not want to discuss: systems that perform impressively under expected conditions, and fail unpredictably the moment conditions shift. The problem is not hardware, it is the AI itself. We do not yet possess the tools and frameworks to characterize when a physical AI system will generalize correctly to a complex situation it has never encountered before, and when it will not. Years of my own research were spent precisely on this question, and our findings are not reassuring: the gap between "works in training" and "works reliably in the unpredictable real world" is not something you can always patch with better engineering. It requires entirely new types of AI.
Israel’s operational knowledge is a strategic asset that has not yet converted into technological leadership in physical AI. Our defense sector has built unparalleled experience working under real-world constraints. Our academic community, with its strong tradition in the mathematical foundations of AI, is positioned to turn this experience into rigorous understanding. Our entrepreneurial ecosystem knows how to bridge the two. What is missing is the connective tissue between these three forces, which currently operate largely in silos. Breaking the silos can be the difference between leading the physical AI revolution and watching it pass us by, the way we did with generative AI.
The properties that matter most in physical AI: defense-grade reliability, real-world experience, theoretical depth, and an engineering culture forged under pressure- already exist here. The recent establishment of a national AI authority within the Prime Minister's Office signals that the field is finally becoming a strategic priority. What we need now are structural connections between defense, academia, capital and the entrepreneurial ecosystem. That can create a flywheel: technology born in the lab, hardened in deployment, commercialized into systems the world buys, and every battlefield lesson flows back to the lab to sharpen the next generation.
According to Fortune Business Insights, the autonomous weapons market is projected to grow significantly, reaching tens of billions of dollars per year over the next decade. For Israel, this is not only a strategic imperative. It is a significant investment opportunity, for the state and for private capital alike. Seizing the opportunity requires government action on two fronts: lowering the regulatory and tax barriers that slow physical AI ventures from evolving, and expanding targeted support through bodies like the Israel Innovation Authority, thereby seeding the next generation of companies in this space. Capital and talent will follow that combination.
Israel does not need to win the race to build the most autonomous weapon. It needs to win the race to understand how to make autonomous systems trustworthy, and to build an ecosystem that turns this understanding into companies. We have the operational experience, the theoretical depth, and the engineering culture to do it. What we need is the will to connect them before someone else does.
Prof. Nadav Cohen is an expert in physical AI and the theoretical foundations of deep learning at Tel Aviv University, and co-founder of IMUBIT.
