eDiscovery Decoder Newsletter
AI-enabled review, prompt-injection controls, adaptive TAR, enforceable ESI protocols, and AI recording in discovery
Five current developments spanning conversational review, prompt injection, CAL, ESI orders, and AI deposition recording.
Relativity Adds KPMG and Three Law Firms to claiR Advanced Access
Relativity added KPMG, Latham & Watkins, Ashurst Perkins Coie, and Troutman Pepper Locke to the advanced-access program for claiR, its conversational AI interface for legal matter data.
- What happened: Relativity added KPMG, Latham & Watkins, Ashurst Perkins Coie, and Troutman Pepper Locke to the advanced-access program for claiR, its conversational AI interface for legal matter data.
- Why it matters: The participant mix puts outside counsel and an advisory firm into product testing alongside existing law-firm participants, giving Relativity feedback on how conversational analysis must operate across large litigation datasets while preserving governance and defensibility.
- Watch: The capabilities and timing are described by Relativity; claiR remains in advanced access, with general availability planned for early 2027.
Prompt-Injection Sanctions Put Hidden Document Content on eDiscovery Risk Lists
A Connecticut trial-court memorandum sanctioned a self-represented litigant for placing hidden instructions aimed at AI systems in filings, turning indirect prompt injection into a concrete litigation and document-review issue.
- What happened: A Connecticut trial-court memorandum sanctioned a self-represented litigant for placing hidden instructions aimed at AI systems in filings, turning indirect prompt injection into a concrete litigation and document-review issue.
- Why it matters: Teams using generative AI for review, investigation, summarization, or translation need controls for manipulated inputs in addition to controls for hallucinated outputs. Existing processing and quality-control practices can be extended to identify concealed instructions before documents enter an AI workflow.
- Watch: The source characterizes the Connecticut memorandum as potentially among the first U.S. opinions imposing sanctions for prompt injection; that characterization is not a comprehensive survey of all state and federal decisions.
Hybrid Search Study Reports Higher Precision After LLM Judgments Train CAL
A Merlin Search Technologies study reported that using high-confidence LLM relevance scores to train continuous active learning improved precision beyond a static hybrid-search ranking.
- What happened: A Merlin Search Technologies study reported that using high-confidence LLM relevance scores to train continuous active learning improved precision beyond a static hybrid-search ranking.
- Why it matters: The experiment describes a practical bridge between search and TAR: search supplies seed evidence, an LLM supplies provisional relevance judgments, and CAL uses those judgments to reprioritize the collection. That design could reduce the amount of static result review while retaining keyword and semantic retrieval.
- Watch: This is vendor-authored research using one public collection, twelve topics, passage-level results, and LLM-generated relevance scores. The reported gains should be validated on matter-specific data with independent quality controls before operational use.
Federal Court Enforces Court-Crafted ESI Order Over Production Objections
In Hall v. Warren, the Western District of New York rejected objections to a court-crafted ESI order addressing metadata, native production, load files, and other production mechanics.
- What happened: In Hall v. Warren, the Western District of New York rejected objections to a court-crafted ESI order addressing metadata, native production, load files, and other production mechanics.
- Why it matters: An ESI protocol entered as a court order creates enforceable obligations, and objections based on routine production mechanics may carry little weight when modern tools and local rules address them. Counsel should identify jurisdiction-specific ESI requirements and negotiate operational details before the court must impose them.
- Watch: The analysis is a vendor publication summarizing the decision and related commentary; practitioners should consult the cited order and applicable local rules for the controlling language.
Nevada Court Finds No Notice Defect in Request to Use AI Deposition Recorder
A District of Nevada decision found no remaining procedural defect in a request to use an AI voice recorder as a secondary deposition record after the requesting party supplied Rule 30 notice.
- What happened: A District of Nevada decision found no remaining procedural defect in a request to use an AI voice recorder as a secondary deposition record after the requesting party supplied Rule 30 notice.
- Why it matters: AI recording tools can implicate both ordinary deposition-notice rules and unresolved reliability, confidentiality, transcription, and evidentiary issues. Teams considering such tools should address the additional method explicitly and preserve a conventional official record.
- Watch: The court resolved only the notice issue and mootness on the record described; it did not endorse the recorder's accuracy, admissibility, security, or use in future depositions or Rule 26(f) conferences.