Israel Israel AI national strategy regulation

Israel Bets Its AI Strategy on Soft Law and a 'Safety Card' Instead of an AI Act

Israel's new National AI Strategic Plan skips standalone AI legislation for sectoral rules and a proposed mutual-recognition safety standard.

Israel's National AI Strategic Plan, by the numbers People of Internet Research · Israel 100,000 AI accelerator target National computing base commitment… ≤5nm Chip fab node target Advanced fabrication facility the … 5 Regional acceleration labs "Acceleration reactors" under the … ~$1.66B Est. annual budget Director Erez Askal's estimate for… peopleofinternet.com
Israel's National AI Strategic Plan, b… People of Internet Research · Israel 100,000 AI accelerator target ≤5nm Chip fab node target 5 Regional acceleration labs ~$1.66B Est. annual budget peopleofinternet.com

Key Takeaways

Israel's National Artificial Intelligence Directorate published its National AI Strategic Plan on August 13, 2026, the implementation blueprint for Government Resolution 4255, "Accelerating Artificial Intelligence in Israel and Establishing Global Leadership," which the cabinet approved on June 16, 2026 (Israel National AI Directorate, PR Newswire). The plan commits the state to a national computing base of at least 100,000 advanced AI accelerators, a domestically built quantum computer under the banner "Project Nexus," and a push to bring an advanced chip fabrication facility at 5-nanometer technology or below onto Israeli soil. What makes the plan a genuine policy document rather than an industrial wish list is the regulatory chapter buried inside it: Israel is explicitly declining to write a standalone AI statute, betting instead on sectoral "soft law" and a proposed "Israeli Safety Card" standard for autonomous systems that officials hope will be mutually recognized by the EU and US.

The case for a law, stated fairly

The skeptics have a real argument, not a strawman. A December 2025 peer-reviewed assessment of Israel's approach in the Cambridge Forum on AI: Law and Governance, by Amir Cahane and Michael Sierra, notes that Israel "employs sector-specific AI regulation rather than a comprehensive framework like the European AI Act," leaning heavily on regulatory sandboxes that are "frequently mentioned as a panacea" even though their actual effectiveness "remains understudied" (Cambridge Forum on AI: Law and Governance). Their worry is structural: without one statute assigning duties across the AI lifecycle, gaps open between regulators, and a company operating in an under-supervised gap can cause harm no single sandbox was built to catch. China's mid-2026 crackdown on "AI boyfriend" companion apps shows the opposite failure mode is real too — Beijing's sudden, blanket rules forced ByteDance, Alibaba and Tencent to shut down emotionally interactive chatbot features overnight, leaving users like the 21-year-old profiled by Rest of World grieving a relationship terminated by regulatory fiat (Rest of World). Both extremes — no coordinating framework, or an inflexible one imposed top-down — carry costs.

Why Israel's bet is still the right one

Israel isn't choosing lawlessness; it's choosing sequencing. The plan builds a National Institute for impact-driven AI applications with five regional "acceleration reactors" and places a National Ethics Council inside that institute as an advisory body on "algorithmic transparency, fairness, privacy, and human authority in decision-making" (Israel National AI Directorate, PR Newswire). That is governance infrastructure, just not codified in a single statute. The rationale Israel's Ministry of Innovation gave when it first adopted this posture in December 2023 remains sound: "a law, being a rigid form of regulation, may fail to synchronize with the globally evolving regulation, potentially leading to an escalating burden on Israeli companies" (IAPP). The EU AI Act's own compliance clock illustrates the risk of over-committing early: obligations drafted against 2023-era model architectures are already being renegotiated as capabilities and deployment patterns move faster than any legislature can redraft primary law.

The more interesting bet is the Israeli Safety Card. Rather than asking Brussels and Washington to accept Israeli AI products on faith, the Directorate wants a certification standard for autonomous systems that both jurisdictions would recognize reciprocally — the same logic Israel has used for decades in defense-adjacent export standards, where interoperability with allied procurement systems matters more than domestic branding. If it works, an Israeli-certified system could clear a US or EU market with less duplicate testing, and Israeli regulators would in turn have to accept the burden of building a credible, internationally trusted certification body rather than a rubber stamp. That's a harder path than passing a law and declaring victory, and it's precisely why it's more defensible: it forces Israel's soft-law bet to prove itself against external scrutiny, not just domestic sign-off.

What actually gets tested

The hard part isn't the announcement, it's the follow-through the plan itself invites. An initial 5,000-accelerator buildout is a fraction of the 100,000-unit target, the quantum computer tender is, by the Directorate's own account, a multi-year undertaking, and the sub-5nm fab is a capital-intensity problem measured in tens of billions of dollars, not a policy commitment. Director Erez Askal has put a rough figure on the ambition, estimating an annual budget near NIS 5 billion (about $1.66 billion) for the plan's compute and infrastructure pillar (Times of Israel). None of that spending forces the regulatory question, though. The Safety Card only earns its name once a foreign regulator actually recognizes it, and the Ethics Council only matters once it has adjudicated a real dispute rather than issued a framework document. Cahane and Sierra's critique should be read as a to-do list, not a rebuttal: build the reverse sandboxes that let regulators proactively probe risk instead of waiting for sandbox applicants to self-select, and publish enough case outcomes that "soft law" stops being a synonym for no enforcement. Israel has chosen the more flexible, more pro-innovation route into AI governance. It now has to spend the next eighteen months proving that route was disciplined, not just cheap.

Sources & Citations

  1. Israel National AI Directorate — National AI Action Plan announcement
  2. Cahane & Sierra, 'Nascent regulatory sandbox frameworks for AI in Israel', Cambridge Forum on AI: Law and Governance
  3. IAPP — Proactive caution: Israel's approach to AI regulation
  4. Times of Israel — quantum computer tender and AI push
  5. Rest of World — China's AI companion ban