If there was a single message that echoed through the discovery-focused sessions at LegalWeek 2026, it was this: generative AI in disputes and investigations is now operational, not experimental. The days of proof-of-concept are over. Firms are deploying AI in real matters, with real data, producing real cost savings—and courts are starting to weigh in on what that means for legal obligations.

Real-World Deployments

Three case studies presented at LegalWeek illustrated the range of current deployments. In an internal investigation, a team used TAR 2.0 combined with GenAI to surface unusual activity in employee communications, identifying coded conversations and unusual expenditures within a subsidiary. In a privilege review, AI-powered domain intelligence classified domains and auto-generated privilege log entries, reducing first-pass privilege review volume by 50%, cutting review time in half, and reducing overall human review costs by 15%. And in an antitrust litigation, GenAI combined with TAR 1.0 was applied across approximately 7 million documents for responsiveness review with no search terms, projecting savings of hundreds of thousands of dollars versus traditional contract attorney review.

Real-World Discovery Deployments at Scale:

Internal Investigations: TAR 2.0 + GenAI for anomaly detection across communications Privilege Review: AI domain intelligence, 50% volume reduction, 15% cost savings Antitrust Litigation: GenAI on 7M documents, no search terms required, hundreds of thousands in savings projected

The Fraud Detection Breakthrough

The fraud detection angle was equally compelling. A session on fact-sheet intelligence in mass torts demonstrated how AI is transforming Plaintiff Fact Sheets from compliance documents into strategic intelligence assets. In major MDLs involving tens of thousands of claims, AI can now flag inconsistencies that humans simply cannot detect at scale: exposure timelines conflicting with diagnosis dates, geographic impossibilities, identical narratives across unrelated plaintiffs, and fabricated documents. A case study from the Uber MDL revealed 21 claims using fabricated receipts generated via third-party websites or AI. The ability to detect these patterns transforms defense strategy from one-off case arguments to systematic, pattern-based approaches.

AI-driven fraud detection isn't about reviewing claims faster. It's about seeing patterns that no human team can spot across thousands of cases simultaneously.

These aren't marginal gains. In a single MDL, identifying dozens of fabricated claims shifts the risk profile fundamentally. It changes settlement dynamics. It shapes defense strategy across the entire litigation.

The Case Law Is Catching Up

A workshop on significant discovery rulings covered cases addressing AI-related discovery obligations, ESI protocols and protective orders, hyperlinks in discovery, and—critically—GenAI hallucination cases including Lifetime Well v. IBSpot.com (2026) and Johnson v. Dunn (2025). The legal framework governing AI use in discovery is no longer hypothetical; it's being written in real time by courts across the country.

The emerging consensus from the case law is clear: firms have an obligation to understand their AI tools. If a tool produces outputs, you need to know whether those outputs are reliable. If you can't validate them, you shouldn't rely on them in discovery. This creates a disincentive for black-box tools and an incentive for transparent, auditable systems.

Workflow Design as Differentiator

A grounded assessment session emphasized that AI is now part of mainstream discovery practice, though adoption varies by matter type, risk profile, and forum. Most uses remain assistive rather than determinative—supporting relevance, privilege, and classification alongside keyword search and TAR. The central message was that thoughtful workflow design is critical, and professional responsibility obligations around competence and supervision apply equally to AI-assisted work.

What distinguishes the most effective deployments from the rest is whether AI operates with awareness of the full matter or in isolation. Tools that connect documents, facts, and strategy across the case produce defensible, pattern-aware results. Isolated tools that process documents one at a time without case context will always struggle to deliver the kind of intelligence that changes litigation outcomes.

The Inflection Point

The message from LegalWeek was unmistakable: AI in discovery has moved from experiment to infrastructure. The question is no longer whether to adopt AI, but how to adopt it responsibly. Firms that invest now in understanding their tools, building transparent workflows, and maintaining audit trails are positioning themselves to lead. Firms that delay will be playing catch-up when the next generation of discovery technology arrives.

The firms winning with AI in discovery aren't the ones with the most sophisticated tools. They're the ones with the most disciplined approaches to validation, workflow, and oversight.

This article draws on reporting from LegalWeek 2026, held March 9–12, 2026 in New York City. The views expressed are those of Advocacy.