Loading...
Loading...
The Southern District of New York is the most litigated federal forum in the country, and the firms that live there — Cravath Swaine & Moore, Sullivan & Cromwell, Wachtell Lipton Rosen & Katz, Davis Polk & Wardwell, and Skadden Arps — did not arrive at their market positions by being late to anything. Technology-assisted review entered Biglaw in New York well before it entered most jurisdictions: SDNY Magistrate Judge Andrew Peck's 2012 Da Silva Moore opinion endorsing predictive coding set a precedent that New York litigators adopted faster than anywhere outside Delaware. In 2026, the conversation has moved well past whether TAR is acceptable — it is table stakes in any SDNY case with a document universe above 100,000 records. The current frontier is deploying AI for continuous active learning on privilege logs, foreign-language document populations (New York matters routinely generate Mandarin, Korean, and Portuguese document sets from financial-sector counterparties), and hallucination-resistant deposition preparation. Simultaneously, the New York Department of Financial Services Part 500 cybersecurity regulation — revised in November 2023 with expanded AI and third-party risk provisions effective through 2025 — has created a dense compliance-advisory market at firms and in-house law departments serving JPMorgan Chase, Goldman Sachs, and the 3,000+ DFS-licensed entities headquartered in New York. The legal AI opportunity in New York is not about whether to deploy — it is about which capabilities to prioritize in a market where the counterparty across the table is running the same tools.
Ask any litigation partner at a Midtown Manhattan firm about AI in discovery and they will describe TAR as infrastructure, not innovation. The real competitive question in 2025 SDNY practice is what happens after first-pass review. Continuous active learning platforms — Relativity Active Learning, Reveal, and Brainspace are the dominant stacks in large New York litigation shops — are now being extended into privilege log automation, foreign-language classification, and pre-deposition document clustering. Sullivan & Cromwell and Davis Polk both run large SDNY-resident litigation departments where the average deal-related dispute generates 500,000 to 5 million documents; privilege review on those populations without AI would be economically prohibitive at New York billing rates. The New York State Bar Association issued guidance in 2024 affirming that attorneys using AI in discovery maintain full Rule 1.1 competence obligations and that inadvertent production of AI-misclassified privileged documents does not automatically constitute a waiver — but the guidance also makes clear that over-reliance on first-pass AI without attorney review protocols creates ethics exposure. Firms that have built internal review pipelines with checkpointed human review at defined confidence thresholds are clearing that bar; firms that set TAR cutoffs at 85% recall without a documented validation protocol are not. In the SDNY context, where sophisticated opposing counsel will challenge TAR protocols in discovery disputes, a defensible validation methodology is the document every AI-using litigator needs and not every AI vendor helps you build.
New York DFS Part 500 (23 NYCRR 500) was amended in November 2023 to add explicit requirements for AI system risk management, third-party due diligence for technology vendors with access to non-public information, and annual certifications from covered entities' chief information security officers. The DFS enforcement record through 2024 — including an $11 million consent order against a major bank's mortgage subsidiary for Part 500 lapses — has sharpened in-house legal teams' attention. JPMorgan Chase, Goldman Sachs, and Citigroup each run legal departments of 800 to 1,200 attorneys with dedicated regulatory-compliance practice groups, and those groups are currently working through the legal-ops implications of Part 500's AI provisions: does deploying a contract-AI tool on counterparty documents create a third-party data-processing relationship that triggers DFS vendor due diligence? In most configurations, yes. In-house teams at Goldman and JPMorgan are requiring AI vendors to complete DFS-compliant third-party risk assessments before provisioning contract tools on non-public deal documents. Law firms advising DFS-regulated clients — and in New York, that includes virtually every financial-institution matter — need to understand the Part 500 vendor-diligence requirements themselves, because the client's legal department will ask whether the firm's AI tools are covered-entity compliant. Operators report that the DFS compliance-advisory work generated by Part 500's November 2023 revisions has created a new practice sub-segment at firms like Debevoise & Plimpton and Covington & Burling's New York office that did not exist at meaningful scale three years ago.
M&A due diligence automation is the highest-volume AI deployment in New York transactional practice, and the economics are straightforward: a $2 billion acquisition with a 60-day timeline and 200,000 documents in the data room is a $600,000 to $1.2 million associate-hour problem without AI and a $150,000 to $300,000 problem with a well-configured contract-review platform. Firms like Wachtell and Skadden do not publicize their internal AI stacks, but their deal timelines have compressed in ways that the associate headcount alone does not explain. The current upgrade cycle in New York M&A practices is moving from clause-identification AI (flagging change-of-control provisions, MAC definitions, reps and warranties) to deal-comparison AI — tools that compare the draft agreement against the firm's closed-deal database to identify deviations from the firm's standard negotiating positions. That second capability requires a proprietary training corpus, which means the larger New York firms with multi-decade closed-deal archives have a durable AI advantage over smaller competitors that cannot build an equivalent training set. For capital-markets work — debt issuances, IPO prospectuses, structured-product documentation — AI is being used for regulatory-disclosure comparison against SEC comment-letter databases and prior-period filings, a use case where the EDGAR corpus provides a large, high-quality training set that generic NLP models can exploit without custom fine-tuning. The cost range for a New York transactional-practice AI deployment at the firm level runs from $200,000 annually for a mid-size shop to $2 million or more for a full-platform enterprise deployment at a Magic Circle or Biglaw firm, with implementation timelines of six to eighteen months.
Strategic planning for AI adoption, readiness assessment, and roadmap development
Workflow automation using AI, including Make.com-style automation and RPA
Text analysis, document automation, sentiment analysis, and language processing
Bespoke AI solutions, model fine-tuning, and custom model development
Ongoing IT support, managed networks, helpdesk, cybersecurity, and infrastructure management enhanced with AI-driven monitoring and automation
A defensible SDNY TAR protocol documents the seed set construction, training rounds, elusion testing methodology, and the recall target used to set the review cutoff — typically 75% to 85% recall with a statistically valid elusion sample. Magistrate judges in the Southern District have increasingly required parties to exchange TAR protocols in electronically stored information stipulations before review begins. Firms using Relativity Active Learning or Reveal should build the protocol documentation into their matter-opening workflow, not as an afterthought. The 2024 NYSBA AI guidance reinforces that attorney oversight at each validation checkpoint is non-negotiable — the protocol needs to show where humans reviewed the AI's outputs, not just what the AI did.
For DFS-regulated covered entities, yes — any third-party vendor that accesses non-public information (which most contract-AI tools do) triggers the Part 500 third-party service provider due diligence requirements under 23 NYCRR 500.11. Law firms using AI on documents from DFS-regulated clients should confirm whether the AI vendor has completed a SOC 2 Type II audit and can provide a DFS-compliant vendor risk assessment. Covered entities that have received DFS examination letters since the November 2023 amendments have reported that AI vendor oversight is a top-three examination focus area. Firms advising JPMorgan, Goldman, or any of the 3,000+ DFS licensees need to treat their own AI vendors as a potential regulatory question, not just a procurement decision.
Publicly available information suggests a hybrid model: most New York Biglaw firms use commercial platforms (iManage, NetDocuments, Relativity, Harvey, Ironclad) for high-volume tasks while exploring proprietary fine-tuning on closed-deal and brief archives for competitive differentiation. Wachtell's M&A practice, which handles some of the highest-value deals in the country, has been reported to use custom AI configurations for deal-comparison work. Davis Polk has publicly discussed its Davis Polk Datasite integration for due diligence. The firms that will have durable AI advantages are those with large proprietary corpora — decades of negotiated agreements, brief libraries, and deposition transcripts — that competitors cannot replicate by simply buying the same vendor platform.
White-collar defense in SDNY — securities fraud, FCPA, AML, bank fraud — generates document populations in the millions on large matters, and the government's own document review is now AI-assisted. Defense firms that are not using comparable TAR and NLP tools risk being at an informational disadvantage when the government's production analysis flags patterns that manual review would miss. Skadden's white-collar group and Sullivan & Cromwell's criminal defense practice are among the New York firms known for sophisticated ESI practices on large SDNY matters. The cost delta between AI-assisted and manual review on a 2-million-document SDNY white-collar matter can reach $800,000 to $1.5 million — which concentrates rapidly into a client billing and competitive retention issue.
A mid-size New York litigation firm — 80 to 200 attorneys — can reach competitive SDNY TAR capability for $150,000 to $350,000 annually, covering a Relativity or Reveal subscription, a Harvey or similar AI drafting assistant, and initial implementation and training. That is the floor. Firms that also want DFS-compliant document-handling for financial-sector clients, a proprietary brief database for motion drafting, and deposition-prep AI should budget $400,000 to $700,000 annually. The shortlist criterion for any New York litigation AI vendor is SDNY familiarity — ask whether they can walk through the firm's TAR validation protocol and whether their platform generates the audit trail that SDNY Magistrate discovery orders require.