Israel building resilient sovereign AI

OPINION:

For years, the global technology industry treated resilience as a background feature, something handled by cloud redundancy and distributed architecture.

That assumption held as long as systems worldwide remained connected, stable and cooperative. Increasingly, however, they are not.

In recent weeks, missile strikes, cyberattacks and infrastructure disruptions launched by Iran have struck across Israel and the Middle East, affecting financial systems, shipping routes, cloud services and more.

In these moments, systems rarely fail dramatically. Instead, they degrade, forcing decisions under uncertainty and without guarantee of external support.

Resilience is no longer about preventing failure. In artificial intelligence systems, it is about controlling how they fail and whether they continue to function when they do.

Israeli institutions have operated critical systems, including advanced AI, defense infrastructure and civilian services, under sustained, multidomain pressure: simultaneous cyberattacks, infrastructure disruption and constrained access to external services.

These are conditions most countries simulate in planning exercises but rarely face in combination.

This is what it means to stress-test AI systems in adversarial conditions, not in controlled environments.

In this environment, a question the rest of the world treats as strategic becomes existential: What happens when the systems you rely on lose their external dependencies?

The answer, it turns out, is that you build differently.

Different design philosophy

What has emerged from Israel’s operating reality is both a battle-tested technology and a design philosophy that the global AI industry is only now beginning to recognize it needs.

Systems are designed to operate within sovereign environments: on-premise deployment, air-gapped execution where required, and local inference without dependency on external application programming interfaces or cloud connectivity.

This is what sovereign AI means in reality, not a flag on a data center or an analysis of AI autonomy. True sovereign AI is the ability to keep operating and deciding, and to retain control under all circumstances.

Today, we believe it is vital that AI systems are deployed inside sovereign environments, structuring fragmented data, and operating under constraints such as air-gapped networks and limited connectivity. The architecture is not adapted from enterprise AI; it is built around these conditions from the outset.

The world is catching up

As conflict exposes these vulnerabilities in real time, other regions are beginning to respond. Europe has committed $235 billion to sovereign AI through its InvestAI initiative. This is in addition to the deployment of nine new supercomputers across 17 member states and a new federated cloud infrastructure through projects such as GAIA-X.

Yet there is a gap between sovereign AI infrastructure and resilient sovereign AI infrastructure, and Israel is one of the few environments in the world where these systems have been tested under sustained, real-world pressure.

Gartner estimates that just 5% of European enterprises use localized AI platforms. That’s a reminder of how far infrastructure still lags behind the realities for which it is being built.

The governance frontier

The global conversation has barely begun to address one more dimension: governance under stress.

Existing frameworks such as the European Union AI Act and the NIST AI Risk Management Framework are important instruments. Both assume functioning audit trails and available human oversight. They were designed for stable conditions, and the environments where sovereign AI matters most — defense, critical infrastructure, emergency response — are precisely the environments where those conditions cannot be guaranteed.

If your oversight system is disrupted or attacked, does your AI halt or continue ungoverned? If the human in the loop is unreachable, then what happens? These questions are the central design challenge for any nation serious about deploying AI where it counts, and they remain largely unanswered.

The nations that solve this — building governance into system architecture rather than layering it on top — will set the standard for the next generation of AI deployment.

Israel is operating at the sharp end of this challenge, and it is further along that path than most of the world realizes.

Agency under pressure

What Israel is building is proof that sovereign AI can work under some of the most demanding conditions any AI infrastructure has ever faced.

The future of AI will not be defined, therefore, by how powerful it is on a clear day. It will be defined by whether it still functions — reliably, securely and aligned to human direction — when the clear days end.

• Jodie Levy is global head of AI governance and policy at Dream, an Israeli sovereign AI company serving government and defense clients globally. She was a founding member of the British AI Security Institute, where she worked with major AI labs and delivered frontier AI security and governance research.