AI and Agentic eDiscovery,
read by litigators, sourced from public records.
Most of what vendors call “AI eDiscovery” in 2026 is TAR 2.0 with a generative-AI scoring layer. Genuinely LLM-native review, the kind that reasons about privilege, narrative coherence, and cross-custodian patterns, is real and narrower than the marketing suggests. This site names the line.
Pricing Reset
At Relativity Fest 2025, Relativity announced that aiR for Review and aiR for Privilege would be included in RelativityOne subscriptions at no additional charge from early 2026. Everlaw responded by making core EverlawAI Assistant features free for existing subscribers. AI capability is no longer a premium add-on at the two largest platforms.
The ripple effects, DISCO holding its position, Logikcull maintaining flat-fee, smaller vendors under pressure, are not yet consolidated on any single reference page. See the full pricing model analysis →
Section 01 // Taxonomy
The five-tier review taxonomy
Pre-2010 baseline
Terms-and-connectors search. Still used in 40 percent of matters, especially early case assessment. Defensible when counsel certifies the search design. Fast, but misses conceptual synonyms.
Da Silva Moore era, 2012
Attorneys code a fixed seed set; a classifier trains on that seed set and predicts document relevance. Approved in Da Silva Moore. Fell out of favour because the fixed seed set became stale on rolling productions.
Post-2014, Rio Tinto era
The classifier retrains continuously as reviewers code documents. Handles rolling productions, converges faster, and is the current industry default. Most vendor 'AI review' branding.
2023 to 2026
An LLM relevance scorer is layered on top of CAL. The lawyer writes a natural-language issue description; the LLM scores documents against it. Relativity aiR, EverlawAI, DISCO Cecilia, Reveal Ask all sit here.
Emerging 2025 to 2026
LLM agents that reason about issues, custodians, and timelines; retrieve and score in multi-step chains; produce structured reasoning traces with per-document explainability. Narrow, real, and ranking quickly.
Section 02 // Vendor Capability Log
Ten platforms, indicative capability
Indicative pricing from public sources. Verify directly with each vendor. Full capability matrix →
| Code | Platform | AI Product | Pricing Model | Indicative Cost | Buyer Focus |
|---|---|---|---|---|---|
| REL | Relativity | aiR for Review / Privilege | per GB | $11-$13/GB/mo, aiR included | AmLaw 200, lit-support |
| EVR | Everlaw | EverlawAI Assistant | per user | $250/mo base + per user | Mid-market, boutique |
| DSC | DISCO | Cecilia AI | custom | Custom quote, cloud-native | Mid-market, in-house |
| RVL | Reveal | Reveal Ask + Brainspace | custom | Custom; concept clustering | AmLaw, enterprise |
| NUX | Nuix | Nuix Neo AI | hybrid | Per user + GB blended | Government, regulator |
| LGK | Logikcull | AI Tagging | per matter | $1,500-$3,500/matter | Solo, small firm |
| EXT | Exterro | Exterro AI | custom | Hold + review bundle | In-house legal ops |
| CSP | Casepoint | CasepointAI | custom | Mid-market, integrated | Mid-market, government |
| LHT | Lighthouse | Lighthouse AI Hub | service bundle | Service-led + platform | AmLaw, enterprise |
| EPC | Epiq / Consilio | Service AI | service bundle | Service-led, pass-through | AmLaw, large in-house |
Last verified Apr 2026 // Confirm with vendor
Section 03 // Workflow Exhibits
Where AI does work in eDiscovery
Predictive Coding 2.0
TAR 2.0 workflows, CAL implementation, statistical validation, Sedona Principle 6 compliance.
OPEN EXHIBIT →
BATES_EREV_001
AI Privilege Review
LLM-assisted privilege detection, accuracy benchmarks (85 to 97 percent), Rule 502(d) orders, privilege log automation.
OPEN EXHIBIT →
BATES_EPRV_001
Redaction Automation
NER and LLM redaction of PII, PHI, and privileged content. Validation methodology and stratified sampling.
OPEN EXHIBIT →
BATES_ERED_001
Early Case Assessment
Data analytics and conceptual clustering before full review. Cost-benefit under FRCP 26(b)(2)(B).
OPEN EXHIBIT →
BATES_EECA_001
Deposition Preparation
AI-assisted timeline reconstruction, document clustering, narrative summarisation for deposition prep.
OPEN EXHIBIT →
BATES_EDEP_001
Pricing + Procurement
Per-GB vs per-user vs per-matter modelled across 500GB, 5TB, and 50TB matters. Interactive calculator.
OPEN EXHIBIT →
BATES_EBUY_001
Section 04 // Case Law Reporter
The cases that matter
The defensibility framework for AI in discovery is an accumulating line of opinions. Four you need to know on a single screen.
287 F.R.D. 182 (S.D.N.Y. 2012)
Da Silva Moore v. Publicis Groupe
Judge Peck became the first U.S. court to approve Technology Assisted Review, holding it 'acceptable' and potentially 'preferable to manual review'. The gold standard for TAR defensibility.
306 F.R.D. 125 (S.D.N.Y. 2015)
Rio Tinto PLC v. Vale S.A.
Judge Peck endorsed TAR without requiring disclosure of the seed set, reinforcing that transparency in process, not document disclosure, is the defensibility standard.
2013 WL 6405156 (N.D. Ind. 2013)
In re Biomet M2a Magnum Hip Implant
Applied FRCP 26(b)(2)(B) proportionality analysis to approve cost-burden shifting in TAR-driven review. Established cost-benefit as a central defensibility factor.
EEOC v. Tesla (N.D. Cal. 2024-2025)
EEOC v. Tesla
The first public-record case involving GenAI document review. The court accepted the review process subject to validation, signalling judicial acceptance of LLM-assisted review.
Section 05 // Ethics Hold
ABA 512 in the GenAI era
ABA Formal Opinion 512 (29 July 2024) confirmed that lawyers using generative AI have the same duties of competence, confidentiality, and candor. In eDiscovery, that means verifying vendor zero-retention terms, understanding how prompts and documents move through a vendor LLM, and disclosing AI use to clients where state-bar guidance requires it.
California, Florida, DC, and New York have all issued AI guidance as of April 2026. The duties scale with the volume of client data crossing into AI systems, which in eDiscovery is the entire production set. Read the full ethics reference →
Section 06 // Common Questions