A General, Not a Ministry, Now Runs Israeli AI Policy
On August 9, 2026, Prime Minister Benjamin Netanyahu and Brigadier General (res.) Erez Askal unveiled Israel's National AI Program, the operational plan under Government Resolution 4255, "Accelerating Artificial Intelligence in Israel and Establishing Global Leadership." The headline goal is ambitious: at least 100,000 advanced AI accelerators as a national horizon, with an initial deployment of 5,000 units, plus a National AI Institute, five regional "acceleration reactors," and a National Ethics Council.
What's structurally unusual is where the authority sits. The National AI Directorate that designed and now runs this plan is not a unit of the Ministry of Innovation, Science and Technology, which held this portfolio as recently as the 2021 Bennett-Lapid government. It is a support unit inside the Prime Minister's Office, with Askal reporting directly to Netanyahu. The Directorate's initial budget — NIS 120 million, mostly earmarked for 2026, plus 20 staff positions, rising to NIS 13 million a year in ongoing operating costs from 2027 — sits with the PMO rather than a line ministry with its own oversight rhythms.
The Appointment That Skipped the Queue
Askal was named Director on October 12, 2025, without a competitive tender. Israel's Civil Service Commission approved the exemption, citing the urgency of standing up the Directorate quickly and classifying the post as a security position given Askal's background — he previously led the IDF's Digital Transformation Directorate and Unit 9900, with senior roles in Air Force and Military Intelligence. To its credit, the Commission limited the exemption explicitly to this first appointment; subsequent hires to the post are expected to go through normal process.
The Case for Centralizing
There's a real argument here, and it deserves a fair hearing before the criticism. Government AI strategies elsewhere have stalled precisely because no single body had the political weight to force ministries to cooperate on data access, procurement, and infrastructure siting. The Civil Service Commission's own justification — that Israel's government "lagged" other systems on AI and needed to move fast — is not an empty talking point; it echoes what critics of slow-moving democracies have said for years about AI adoption races. Putting the Directorate in the PMO, reporting to the head of government rather than a cabinet minister who must negotiate coalition politics, is a plausible way to cut through exactly that kind of gridlock. If speed is genuinely the scarce resource, structural centralization is a defensible tool.
Where the Bet Gets Risky
The risk is that centralized authority without centralized accountability just relocates the bottleneck rather than removing it. Askal himself told a Knesset committee in July 2026 that fully implementing the national plan would require roughly NIS 5 billion a year — more than 40 times the Directorate's initial NIS 120 million appropriation. That gap between the stated ambition and the funded reality is the kind of thing a line ministry, subject to routine Knesset budget committee scrutiny, would normally be forced to reconcile in public. A PMO support unit has looser structural obligation to do so.
There is also a jurisdictional problem the launch didn't resolve. The Israel National Cyber Directorate (INCD), a mature body with its own statutory mandate, already handles national-level incident response — including, increasingly, cyberattacks that use or target AI systems. Analysts have flagged direct overlap between the AI Directorate's stated mandate ("formulation and implementation of national strategy," regulation, infrastructure, international negotiations) and INCD's existing turf, with the added complication that both directors come from IDF intelligence backgrounds and past INCD-Shin Bet jurisdictional disputes took years to settle informally rather than by clear delineation. Layering a second AI-focused authority on top of an already-contested cyber governance landscape, without a public memorandum spelling out who leads an AI-enabled hospital ransomware incident, invites exactly the kind of institutional friction that slows implementation rather than speeding it.
A Proportionate Fix, Not a Reversal
None of this argues for scrapping the centralized model. Several fast-moving AI programs abroad route through a single empowered coordinator rather than a committee of ministries, and there's a reasonable case Israel needed to break from a system that previously produced no comparable urgency. But legitimacy for that model has to be earned with transparency the launch hasn't yet supplied. Three fixes are consistent with the pro-innovation case for centralization rather than opposed to it: hold the Civil Service Commission to its own limit and put every subsequent Directorate hire through open tender; require the Directorate to report budget execution to a standing Knesset committee on the same cadence a ministry would; and formalize — in writing, not through the goodwill of two generals — where AI Directorate authority ends and INCD authority begins. Centralizing AI policy in the PMO can be the right call. Doing it without the paperwork that normally comes with ministry status is the part still owed to the public.