TL;DR:
In Joel Gavalas, as Personal Representative of the Estate of Jonathan Gavalas v. Google LLC and Alphabet Inc., the United States District Court for the Northern District of California entered on November 6, 2026 a stipulated protective order, as modified by the court, that creates a two-tier confidentiality regime for discovery and imposes strict restrictions on the use of generative AI in handling confidential information. Key requirements prohibit input of protected material into consumer grade AI tools, compel the use of enterprise-grade “Confidential AI Tools” with contractual safeguards against model training, data disclosure, and third-party access, and mandate deletion of AI-generated outputs and training data at the close of litigation. The order also limits expert access to non-competitor users under controlled conditions and requires explicit post-disposition return or destruction of protected materials. For trial teams, the practical effect is a concrete, court-backed framework to govern AI-assisted review, data-security, and the handling of sensitive information in discovery, with clear consequences for violations. Counsel should audit current AI vendors, update discovery workflows, and implement protective measures now to avoid later disputes or sanctions.
What happened
On November 6, 2026, Magistrate Judge Susan van Keulen entered a [Stipulated Protective Order] in Gavalas v. Google LLC, No. 5:26-cv-1849-EKL-SVK, ND Cal, San Jose Division. The order is based on the district’s model for litigation involving confidential information and trade secrets, but it was modified by the court to address AI-related discovery practices. The proceedings concern Joel Gavalas, acting as personal representative of the Estate of Jonathan Gavalas, against Google LLC and Alphabet Inc. The document makes explicit that the protective regime applies to confidential information and data generated, copied, or extracted from protected materials, and it sets up a framework for how AI tools may be used in connection with the case. The court’s entry confirms the parties’ compromise and the court’s explicit supervision of key AI-related disclosure issues. (govinfo.gov)
The regime at a glance
The order establishes a two-tier confidentiality framework for Protected Material designated as “CONFIDENTIAL” or “HIGHLY CONFIDENTIAL – ATTORNEYS’ EYES ONLY.” It extends protection to copies, summaries, and derivatives that reveal that material, and it limits access to specifically designated recipients, with a mechanism to challenge designations. A central and novel feature is Section 12.9, which prohibits inputting, uploading, or submitting Protected Material to consumer-grade AI tools or publicly accessible AI platforms. The order envisions use of enterprise-grade Confidential AI Tools only, conditioned on contractual safeguards that (a) prohibit training on inputs to improve the AI model, (b) restrict disclosure of inputs and outputs to third parties, and (c) permit deletion of protected data upon request. The court’s framework requires that any AI tool used in handling Protected Material must be a “Confidential AI Tool” with these protections, and it obligates the receiving party to document the contractual protections before using such tools. (govinfo.gov)
Practical consequences for trial teams
Concrete ban on training data risk: The order treats inputs, outputs, and derived artifacts as protected, and it ties these protections to ongoing obligations for enterprise AI tools. If a team feeds a protected document into a consumer AI service, that action could constitute unauthorized disclosure with stipulated remedies. Practitioners should audit every AI workflow used in discovery, including e-discovery review, document summarization, and translation, to ensure no protected material is exposed to consumer tools. (govinfo.gov)
Strict criteria for AI tooling: Only enterprise-grade, “Confidential AI Tools” may be used, and such tools must have contracts prohibiting model training on inputs, restricting disclosure of inputs and metadata, and enabling deletion of data upon request. Firms should map their AI stack to confirm that each tool complies with these requirements before processing any protected data. Documentation of these safeguards is not optional; the order requires it as a condition of access to protected materials. (govinfo.gov)
Clear post-disposition obligations: The order contemplates returning or destroying Protected Material and AI-generated outputs within 60 days after final disposition, with certification to that effect. This creates a measurable lifecycle for digital materials and reinforces the need for robust data-retention and destruction procedures in litigation teams. (govinfo.gov)
Expert-access constraints with risk-aware handling: The order defines Experts to include certain AI developers and other professionals, with access governed by explicit safeguards and potential court review if access is contested. It also requires that designated experts comply with confidentiality terms and that access be tightly scoped to necessary tasks. This has practical implications for how outside consultant work intersects with privileged or confidential information during trial preparation. (govinfo.gov)
Enforcement and professional-responsibility implications: Noncompliant submissions to AI tools are treated as unauthorized third-party disclosures, triggering the order’s remedial duties, including notice and corrective actions. In practice, this elevates the risk calculus around AI-assisted review and could influence the risk of sanctions or other consequences for missteps in data handling. (govinfo.gov)
How to adapt litigation practice now
Conduct an AITools audit: Litigation teams should inventory all AI-enabled workflows used in the case, including document review, deposition summarization, translation, and trial graphics preparation. Identify which tools are consumer-grade versus enterprise-grade and ensure all use complies with the order’s requirements. If any tool cannot be configured to prohibit data training and external disclosure, discontinue its use for Protected Material. (govinfo.gov)
Replace or configure tools accordingly: Where enterprise-grade tools are in use, verify their data-handling terms align with the order. Ensure vendor contracts explicitly prohibit training on inputs, limit access to authorized personnel, and enable deletion of data after the litigation closes. Maintain readily auditable documentation demonstrating compliance for in-house counsel and opposing counsel. (govinfo.gov)
Update internal playbooks and training: Create or refresh internal guidelines that reflect the protective order's requirements, including a do-not-use list of consumer AI tools for any Protected Material, a standardized process for designating Expert access, and a checklist for verifying tool compliance before any data enters an AI workflow. Training should emphasize the consequences of noncompliance and the procedure for handling suspected violations. (govinfo.gov)
Align with post-disposition protocols: Establish a documented process to return or destroy Protected Material and AI-generated artifacts within 60 days after final disposition, including certification to the court. This minimizes lingering data exposure risks and supports a clean closure of the matter. (govinfo.gov)
Prepare for potential future disputes: The order shows the court’s willingness to tailor AI protections within a protective order. Anticipate questions from opposing counsel about access for experts, the scope of permissible AI-assisted work, and what constitutes a compliant “Confidential AI Tool.” Having a ready set of compliant workflows and contract language can streamline future disputes or requests for modification. (govinfo.gov)
Takeaway for trial teams in 2026
The Gavalas protective order represents a concrete and enforceable blueprint for governing AI in discovery within a high-stakes civil action. For trial teams, the practical impact is a shift from general caution about AI to a precise, court-endorsed standard: no inputs to consumer AI platforms, mandatory use of enterprise-grade AI tools with strict non-training obligations, and clear post-litigation data-handling duties. As AI continues to permeate litigation workflows, orders like this signal the legal profession’s move toward sophisticated data governance in discovery, with real consequences for teams that neglect to implement them.
Sources:
- United States District Court for the Northern District of California, Gavalas v. Google LLC et al., Stipulated Protective Order as Modified by the Court, Case No. 5:26-cv-1849-EKL-SVK, entered November 6, 2026. GovInfo PDF: USCOURTS-cand-5_26-cv-01849/USCOURTS-cand-5_26-cv-01849-3.pdf. (govinfo.gov)
- LexSummary, Joel Gavalas, as Personal Representative of the Estate of Jonathan Gavalas v. Google LLC and Alphabet Inc. — Court-approved protective order restricts confidential discovery material to secured enterprise AI tools and bars consumer AI uploads, updated November 7, 2026. (lexsummary.com)
- Justia Dockets & Filings, Order Re 46 Stipulated Protective Order, Gavalas v. Google LLC, 5:26-cv-1849-EKL-SVK (ND Cal), filed November 6, 2026. (docs.justia.com)