A Small State's Bet on Integration, Not Scale
On August 13, 2026, Israel's National Artificial Intelligence Directorate — housed in the Prime Minister's Office and headed by Brig.-Gen. (Res.) Erez Askal — unveiled the action plan implementing Government Resolution 4255, "Accelerating Artificial Intelligence in Israel and Establishing Global Leadership." The five-year vision, branded "Lived AI," concedes upfront that Israel cannot out-train the United States or China on frontier models. Instead it bets on being the first country where AI is threaded through daily public life — healthcare, education, government services — backed by a national computing base of at least 100,000 advanced AI accelerators (an initial 5,000-unit deployment first), a domestic quantum computer, a sub-5nm chip fabrication push, new AI-security labs built with the National Cyber Directorate, and a National Ethics Council advising on algorithmic transparency, fairness and privacy.
That framing is sound policy instinct. A country of under 10 million people has no realistic path to competing on raw compute with hyperscalers; competing on deployment — proving AI works inside a functioning state — is a coherent alternative lane, and one that plays to Israel's actual strength: a dense, well-funded tech sector that has already turned its cybersecurity niche into a $3.8 billion annual investment magnet, 40% of all private U.S. cyber investment in 2024.
The Governance Question Is the Real Story
The more consequential choice is institutional, not technological. The Directorate sits inside the Prime Minister's Office rather than the Ministry of Innovation, Science and Technology, which previously owned AI policy, and Askal was appointed without a competitive tender — the Civil Service Commission cited urgency and security considerations to justify the exemption.
There's a real case for that structure. The Nagel Committee, which reviewed Israel's AI readiness before the plan was drafted, concluded the country "is not in a good position" to move fast, and AI-security work genuinely benefits from sitting close to the Cyber Directorate and the PMO's coordinating authority rather than being negotiated across ministries. Centralizing early-stage strategy for speed is defensible in a way that centralizing permanent regulatory power would not be.
But speed and scrutiny aren't actually in tension here, and treating them as if they were is where the plan overreaches. A tender process for the director's role would have taken weeks, not months. Diverting a public budget line to the PMO rather than an existing ministry with established oversight machinery raises the ordinary accountability questions any concentration of discretionary spending should raise — questions that don't go away because the subject is AI. The Jerusalem Post has already flagged friction between the new Directorate and the existing Cyber Directorate as a live risk, echoing the turf disputes that followed the 2014 establishment of the cyber body itself. Proportionate governance means matching institutional scrutiny to the money and authority involved — not waiving it because the topic is urgent.
Ambition Outruns the Balance Sheet
The plan's numbers don't yet support its rhetoric. Globes reports the Directorate operates on roughly a NIS 3 billion 2026 allocation with 20 staff, against a 100,000-accelerator target that, at current Nvidia Vera Rubin pricing near $135,000 per unit, would cost on the order of $13.5 billion to fully build out — a gap Israel's own Ministry of Finance Budget Department has publicly called "bombastic" and "laden with... unfounded declarations," also noting duplicative procurement oversight spread across five separate agencies. Globes is blunter still, calling this the fourth Israeli national AI plan unveiled in seven years without one delivering results, and quoting an industry source who called the August rollout "a drop in the ocean."
"What the Directorate presented this week is too little and too late; it is a drop in the ocean." — senior industry source, quoted in Globes
That skepticism is earned, but it cuts toward a different conclusion than "the plan is too weak." A government that sets an aspirational compute target without matching capital isn't imposing an unfunded mandate on the private sector — it's simply failing to deliver its own infrastructure promise. The proportionate response isn't more state control over AI development; it's honest budgeting and a narrower, achievable first phase (the 5,000-accelerator initial deployment is the right unit to judge, not the 100,000 headline).
Getting the Ethics Model Right
Where the plan clearly gets the calibration right is the National Ethics Council itself: advisory, not a licensing or classification regime. That's the correct posture at this stage — Israel has resisted the instinct, visible in the EU's binding AI Act, to legislate a compliance apparatus before use cases and harms are well understood. An advisory body that flags bias, transparency and privacy concerns as AI enters public services, without gating deployment on pre-clearance, preserves room for the state to actually build the "Lived AI" integration the plan wants, rather than drowning early pilots in paperwork.
The test for this plan isn't whether Israel regulates AI too little — on current evidence it doesn't — but whether the Directorate can convert a rushed, PMO-centered launch into a funded, accountable second year. Watch the 5,000-accelerator deployment and the 2027 budget line, not the 100,000 headline.