How secure is the future of dual-use defense? – Opinion

This gap has direct operational implications. Modern defense systems no longer function as closed, siled, air-gapped platforms. They operate as complex digital ecosystems built on software updates, remote connectivity, distributed sensors, and shared data pipelines. AI models are at the core and sometimes at the edges of this architecture. Correlate sensor inputs, recommend actions, prioritize threats, and support real-time decision making. Eventually, AI may be able to act on these decisions.

This is especially true in dual-use systems. The same computer vision algorithms that enable self-driving cars on public roads are now being built into unmanned aircraft systems. Commercial satellite imagery platforms provide feeds into military intelligence workflows. A cloud-based analytics engine supports command and control environments.

These technologies offer tremendous benefits. These reduce costs, shorten development cycles, and allow defense forces to benefit from the pace of commercial innovation. But they can also introduce commercial vulnerabilities directly into military systems.

AI expands the attack surface in unprecedented ways. Models can be manipulated through contaminated training data or adversarial inputs. Automated systems can be fooled by machine speed. Shared software libraries and open source components create hidden dependencies that attackers can exploit.

Historically, defensive superiority has been measured by platforms, munitions, and physical reach. There is now an increased reliance on the integrity of algorithms, data pipelines, and decision-making systems. Compromised sensor feeds or manipulated AI models can reduce situational awareness just as much as physical sabotage.

At the same time, AI is becoming deeply integrated into command and control environments. Today, decision support systems synthesize streams of information that cannot be handled by human teams alone. Autonomy allows one operator to manage multiple platforms simultaneously. While these features are essential for modern multidomain operations, they also introduce new failure modes.

If an AI system misinterprets context, amplifies noise, or obscures uncertainty, human operators may not immediately recognize the error. In a high-tempo environment, trust in automation can become a dependency.

It is important to emphasize that this is not an argument against AI. This is an argument against deploying AI without proper governance, resilience, cybersecurity, and validation.

WEF data highlights that many organizations remain unprepared. The combination of a high likelihood of incidents, poor governance, inexperience in responding to AI cyberattacks, and a lack of skills suggests structural vulnerabilities. In a rapidly escalating defense environment, this vulnerability could have strategic implications.

The key challenge lies in governance. Traditional defense certification models assume a deterministic system whose behavior can be thoroughly tested. AI systems don’t work that way. They can learn from data, adapt to patterns, and react unexpectedly to unfamiliar situations.

This makes verification, explainability, governance, and accountability difficult. Military commanders must be able to understand why the system has issued the recommendations. Engineers must be able to verify performance under adversarial conditions. Policy makers need to be confident that automated systems comply with doctrinal, legal, ethical and operational constraints. Without these guardrails, AI risks becoming a powerful but highly dangerous layer inserted between decision makers and reality.

The cybersecurity aspect makes this challenge even more acute. Attackers are already using AI to automate reconnaissance, create adaptive malware, and scale up their penetration attempts. Defenders increasingly rely on AI to detect anomalies and respond at machine speed. This accelerates feedback loops, making both attacks and defenses more automated, more complex, and less transparent.

For defense systems built on dual-use foundations, that loop becomes especially dangerous. Commercial AI platforms are not designed with the hotly contested battlefield in mind. They are optimized for performance, scale, and efficiency. Closing this gap requires intentional action.

Defense organizations need to invest in AI-savvy cyber expertise. The WEF finding that 54% of organizations perceive a skills shortage should serve as a warning. Securing AI systems requires experts who understand AI, cybersecurity, and adversarial threat models.

Explainability and continuous verification must be non-negotiable. AI systems deployed in defense environments must be transparent, testable, and continuously monitored for drift and tampering. Trust must be designed and verified. That cannot be assumed.

Finally, cooperation between government, industry and allies is essential. AI-centric cyber risks do not respect institutional or state boundaries. Threat intelligence, defensive techniques, and best practices must flow across sectors that increasingly share the same technology infrastructure.

Israel has a unique position in this landscape. The company’s defense division is deeply connected to a world-class technology ecosystem, and its experience operating under persistent cyber threats provides important perspective. Leadership in AI-driven defense depends not only on the speed of innovation but also on the ability to protect what is built.

The strategic competition unfolding today is not just about the companies that develop the most advanced AI models. It’s about who can reliably deploy them, protect them under pressure, and maintain confidence in their operations when the going gets tough.

The lessons from the World Economic Forum’s Cybersecurity Outlook are clear. The future of defense will also be defined by cyber and AI security. In the AI ​​era, strategic advantage will accrue to those who can ensure that the systems that guide decision-making are reliable, resilient, and secure when it matters most.