In June 2026, Microsoft published the concluding update of its external review — led by the law firm Covington & Burling — into how Israeli military intelligence used its cloud. The review confirmed the core of what The Guardian, +972 Magazine and Local Call reported on August 6, 2025: the Israel Ministry of Defense, through the signals-intelligence Unit 8200, consumed Azure storage in the Netherlands and Microsoft AI services to hold and analyze intercepted Palestinian phone calls, in violation of Microsoft's acceptable-use terms prohibiting mass civilian surveillance (CTech). Microsoft pledged stronger pre-contract review of national-security engagements and enhanced human-rights due diligence in conflict-affected areas — while keeping its broader commercial relationship with the Israeli government.
The most important fact in this story is institutional, not technical: the mechanism that finally constrained one of the world's largest single-population surveillance archives was a vendor's contract, enforced from Redmond. Israeli privacy law — freshly overhauled — never touched it.
The check that worked was contractual
According to the joint investigation, the system had been operational since 2022, capturing recordings of millions of daily phone calls from Palestinians in Gaza and the West Bank, with Microsoft engineers reportedly helping move up to 70 percent of Unit 8200's sensitive data onto Azure servers in the Netherlands and Ireland (Al Jazeera). Microsoft opened a review on August 15, 2025, and on September 25, 2025, Vice Chair Brad Smith announced the company had found evidence supporting the reporting — including "IMOD consumption of Azure storage capacity in the Netherlands and the use of AI services" — and would "cease and disable specified IMOD subscriptions," including specific cloud storage and AI technologies (Microsoft On the Issues).
Notably, the review never examined customer content. It relied on Microsoft's own business records, billing data and internal communications. The de facto regulator of Israel's largest intelligence database operated with less investigative access than a civilian privacy authority would demand of a supermarket loyalty scheme.
What Israeli law couldn't see
This happened in the same window in which Israel completed its most ambitious privacy reform in decades. Amendment 13 to the Protection of Privacy Law, 5741-1981 — the statute itself dates to 1981 (WIPO Lex) — passed the Knesset on August 5, 2024 and took effect on August 14, 2025, arming the Privacy Protection Authority with administrative fines, cease-processing orders and criminal investigation powers (IAPP).
But the amendment deliberately routes security bodies around that machinery. Under Section 23U, authorities "dedicated to security" — the police, the IDF, the Shin Bet, the Mossad — are supervised by an internal privacy supervisor appointed from among the agency's own employees, who reports to the head of the security body and works to an annual plan approved jointly with the PPA (FPF Israel Tech Policy Institute). The PPA cannot directly inspect, fine or enjoin them. And because the law applies territorially, to data subjects within Israel's jurisdiction, Palestinians in Gaza and the West Bank — the population whose calls filled those Dutch data centers — sit largely outside its protective reach altogether.
The steelman — and why it fails here
The case for the carve-out is genuine. Signals intelligence is a core sovereign function; exposing classified collection systems to a civilian regulator's inspectors creates real leak and counterintelligence risks. Every democratic privacy regime makes some version of this trade — the GDPR itself excludes national-security processing from its scope. And Amendment 13's internal-supervisor model, with PPA guidance, is more oversight than Israeli security agencies faced before.
But the Unit 8200 episode is a controlled experiment in whether that model works, and the result is unflattering. Internal supervision did not surface, slow or scope a program whose own engineers described its ambition as storing "a million calls an hour." Journalists surfaced it; a vendor's legal department stopped parts of it. That arrangement should trouble security hawks as much as civil libertarians: it means operational continuity for sensitive state workloads now depends on the reputational calculus of a foreign corporation, applied through terms of service, with no due process in either direction. A company can under-enforce for years, then over-correct under pressure. Neither failure mode is accountable to anyone.
A proportionate fix
The answer is not less cloud. Hyperscale infrastructure and AI analytics are now baseline capabilities for any serious security establishment, and forcing intelligence workloads back into opaque sovereign silos would reduce external visibility to zero. The answer is independent oversight proportionate to the sensitivity: a security-cleared judicial or Knesset-level body with authority over bulk processing by exempt agencies, on the model of allied democracies' intelligence-oversight commissioners — supervision that travels with the data, wherever it is hosted.
Microsoft's new commitments — pre-contract national-security review, periodic reassessment when political circumstances change, employee escalation channels — are welcome, and EFF is right that Google and Amazon should match the disclosure precedent rather than hide behind silence (EFF). But voluntary vendor governance is a stopgap, not a system. The lesson for every government negotiating sovereign cloud and AI deals is simple: where the law leaves a gap, the gap gets filled by whoever owns the data center. Legislatures should prefer to fill it themselves.