For the past fourteen years, the phrase "Da Silva Moore moment" has been shorthand among e-discovery practitioners for the instant courts stop treating a review technology as exotic and start applying ordinary proportionality doctrine to it. Predictive coding earned that moment in 2012 when Magistrate Judge Andrew Peck approved its use in Da Silva Moore v. Publicis Groupe (S.D.N.Y. 2012). On July 1, 2026, generative artificial intelligence ("GenAI") received its own. In Schulte v. LinkedIn Corp., No. 22-cv-00237-HSG (LB), 2026 WL 1905851 (N.D. Cal. July 1, 2026), Magistrate Judge Laurel Beeler upheld LinkedIn's use of Relativity aiR to make final responsiveness calls in litigation review — expressly declining to impose any heightened or novel standard on the technology (DLA Piper; WilmerHale).
The ruling is consequential not because it resolved deep doctrinal questions about artificial intelligence, but precisely because it refused to. Courts will not build a new AI-specific discovery regime; they will enforce the existing one with full rigor. This article examines the three operative holdings in Schulte, explains why ESI protocol language has become load-bearing infrastructure for any GenAI-assisted review, and provides practical guidance for litigators and in-house counsel negotiating those protocols today.
What the Court Actually Held — and What It Did Not
Judge Beeler's analysis rested on three discrete rulings, each of which deserves clear-eyed reading (Arnold & Porter; CSDISCO).
First, the court upheld LinkedIn's right to pre-cull the document population with negotiated search strings before feeding documents into the AI platform. Pre-culling is not evasion; it is proportional project design, provided the methodology is disclosed and consistent with what the ESI protocol permits (Relativity).
Second, the court denied plaintiffs' motion to compel disclosure of the AI tool's validation metrics. To obtain that kind of transparency — "discovery on discovery" into the review methodology itself — a challenger must point to a concrete, particularized deficiency in the actual production. Generalized skepticism about a GenAI tool's reliability does not suffice (HaystackID).
Third, and perhaps most important structurally, the court found LinkedIn's disclosures sufficient because they satisfied what the court-ordered ESI protocol required — not because they met some free-standing AI-transparency standard that exists in the ether. The protocol was the measure of adequacy. That framing is the holding that will shape every negotiation going forward.
The Emerging 2026 Judicial Framework
Schulte does not stand alone. Read alongside two earlier 2026 rulings, it completes a coherent architecture. In Morgan v. V2X (D. Colo. Mar. 30, 2026), the court addressed work-product protection for AI-generated outputs, confirming that existing doctrine travels with the technology (Akin Gump). A separate Q1 2026 ruling — commonly referenced as Jeffries — imposed a ban on uploading discovery materials to open, non-enterprise AI tools, grounding the prohibition in protective-order obligations rather than any new AI rule (eDiscovery Today).
The pattern is unmistakable: courts are not drafting AI-specific discovery law. They are applying FRCP 26(b) reasonableness-and-proportionality, protective-order duties, and work-product doctrine to AI workflows with the same analytical tools they deploy everywhere else. The open question for practitioners is not whether courts will accept GenAI review; Schulte answers that. The question is whether counsel have negotiated ESI protocols that protect their clients once that review is scrutinized (Legal Tech News).
Practical Guidance for Litigators and In-House Counsel
The following steps should be taken before any GenAI-assisted review workflow is deployed and, critically, before the ESI protocol is entered as a court order.
1. Negotiate GenAI-specific disclosure provisions into the protocol before it is entered. Because Schulte made the protocol the sole measure of LinkedIn's disclosure obligations, boilerplate TAR language will not adequately address GenAI workflows. Requesting parties should seek provisions specifying which AI tool is used, the prompt architecture or configuration parameters disclosed at a level sufficient to assess methodology, and what pre-culling steps preceded the AI review. Producing parties should define those obligations narrowly and precisely so that compliance is demonstrable.
2. Document cost and quality-control steps regardless of which side you represent. Proportionality doctrine operates in both directions. If GenAI materially reduces review cost, opponents will argue that the traditional "undue burden" shield is correspondingly smaller. Producing parties should preserve contemporaneous records of: (i) per-document review costs before and after AI deployment; (ii) quality-control sampling protocols applied to the AI's determinations; and (iii) validation steps taken to confirm production completeness. That record is the foundation of any proportionality defense.
3. Identify a concrete production deficiency before seeking discovery on discovery. Challengers who wish to compel disclosure of AI validation metrics must move beyond general distrust of the technology. Counsel should conduct a careful gap analysis of the production — comparing custodian counts, date-range coverage, and document-type distributions against what was represented — before filing any motion to compel review-methodology disclosure. Abstract skepticism about machine learning will not carry the motion under Schulte.
4. Confirm that the ESI protocol governs AI tool selection and any mid-review platform changes. Parties sometimes switch review platforms or upgrade AI configurations after the protocol is entered. The protocol should address whether such changes require notice or re-negotiation, and producing parties should treat any significant workflow change as a potential disclosure event.
5. Address open-tool prohibitions in every protective order. Following the Jeffries line, any protective order governing confidential or attorney-eyes-only materials should expressly prohibit upload of those materials to non-enterprise, publicly accessible AI services. Counsel should audit their firm's and client's AI subscriptions to confirm that discovery materials processed through GenAI tools remain within enterprise-controlled, data-isolated environments.
Conclusion
In conclusion, Schulte v. LinkedIn is not a green light for uncritical AI-assisted review; it is a calibration. Courts have confirmed that GenAI responsiveness determinations are evaluated under FRCP 26(b) proportionality — the same framework that has governed technology-assisted review for fourteen years. That doctrinal stability is genuine progress, but it places the full weight of client protection on the ESI protocol and the workflow documentation that surrounds it. Litigators and in-house counsel who negotiate those terms carefully, and who build quality-control records before a dispute arises, will be positioned to defend their review methodology when challenged. Those who carry over generic TAR language into GenAI workflows without revision will discover, as producing parties sometimes did in the TAR era, that the protocol they agreed to is the one they are held to. The 2026 judicial framework is now clear enough to act on; the practitioners who act on it now will be the ones who shape the next generation of precedent rather than respond to it.