Israel's National Artificial Intelligence Directorate, the body inside the Prime Minister's Office established under Government Resolution 4255, published its National AI Action Plan on August 13, 2026. The Directorate calls the five-year program "Lived AI" — a bet that Israel's advantage will come not from training the largest foundation models, but from being the first country where AI is fully embedded in daily public life, from hospitals to classrooms to government service counters.
The headline commitments are large: a national computing base of at least 100,000 advanced AI accelerators (an initial 5,000-unit buildout, with the rest "led by the private market"), a domestically built quantum computer procured under a "state as customer" model, and a push toward an advanced chip fabrication facility at 5-nanometer technology or below. A National Institute for Impact-Driven AI Applications will run five regional "acceleration reactors," at least two sited outside the Tel Aviv core, alongside AI-security laboratories built jointly with Israel's National Cyber Directorate (Israel's National AI Directorate press release).
The Regulatory Core: "Right of Way for AI"
The plan's most consequential line isn't an infrastructure target — it's a regulatory default. It states that an action an AI system can perform better than a human "should not be prevented by regulation unless a specific and material risk has been proven," and it commits to compatibility with "leading Western privacy regimes, chief among them the GDPR" (Unite.AI). That is a genuinely pro-innovation starting posture: it puts the burden of proof on the regulator to show harm, not on the deployer to prove safety in advance. Most jurisdictions, the EU AI Act chief among them, run the opposite way — risk classification first, deployment second.
The steelman for a more cautious posture is real, and worth stating plainly. AI systems embedded in healthcare triage, benefits eligibility, and policing — exactly the "daily public life" use cases this plan targets — can encode bias or fail silently at a scale no individual human decision-maker could match, and a burden-shifted regulatory default means those failures surface only after deployment, not before. A National Ethics Council makes sense as an answer to that concern. The plan places one inside the new Institute, with a mandate covering "algorithmic transparency, fairness, privacy, and human authority in decision-making" (Unite.AI).
But the Council is advisory, sitting inside the implementing body rather than outside it with independent standing, binding review power, or a public reporting mandate. Paired with a "right of way" default that requires proven material risk before regulation kicks in, the ethics function reads as a compliance layer bolted onto a deployment engine, not a check on it. For a government that is simultaneously building the compute, curating the acceleration reactors, and running the ethics board, that structure invites the same criticism GDPR's own regulators have long made of self-assessment regimes: oversight that reports to the deployer rarely constrains the deployer.
The Money Doesn't Match the Ambition
The bigger problem so far isn't the ethics architecture — it's the budget. Israel's National AI Directorate has estimated the plan requires an annual budget of roughly NIS 5 billion (~$1.66 billion), while the 2026 allocation actually secured is NIS 3 billion (~$1 billion) (Calcalist). Globes reported the plan contains no itemized budget for several of its named initiatives, using terms like "dozens" and "hundreds of millions of shekels" rather than committed figures — a gap one industry source called "too little and too late; a drop in the ocean" against the scale of the 100,000-accelerator target, which independent estimates put at $20–30 billion in hardware costs alone (Globes; ynet). Globes also noted this is the fourth Netanyahu-government AI plan announced in seven years, and that the current NIS 3 billion figure is a fraction of a 2024 proposal from Brig. Gen. (Res.) Jacob Nagel that had called for NIS 25 billion over five years, including an NIS 18 billion national supercomputer.
That gap matters more than the ethics-council question for assessing whether "Lived AI" is real policy or aspirational branding. A permissive, deployment-first regulatory default is defensible precisely because it trades away precautionary friction for speed — but only if the state actually delivers the compute, chips, and quantum capacity it promises. An underfunded plan with light-touch rules is the worst of both configurations: it neither protects against the harms a stricter regime would catch, nor builds the infrastructure a well-funded one would.
What to Watch
The Directorate published a Request for Information on sovereign foundation models and opened public consultation through September 6, 2026, ahead of selecting the acceleration-reactor sites and quantum tender ("Project Nexus") bidders (The Quantum Insider). The near-term signal to watch isn't the 100,000-accelerator target — that's a five-year horizon, and largely privately funded by design. It's whether the 2027 budget round closes the gap between the NIS 5 billion the Directorate says it needs and the NIS 3 billion it has, and whether the Ethics Council publishes anything before the reactors are already operational. A "right of way for AI" default is the correct instinct for a small, innovation-dependent economy competing on deployment speed. It only earns public trust if the oversight body riding alongside it has more than an advisory voice.