AI and TAR case law -- a working litigator's reference (April 2026)
Updated 21 April 2026 | Case citations verified against public court records | Not legal advice
The defensibility framework for AI in discovery is not a single case; it is an accumulating line of opinions that treat AI the same way they treated TAR: through the lens of process transparency, proportionality, and cooperation. Here is every case you need to know, with the specific holding and practical takeaway for each.
The core principle across all cases
Courts evaluate AI-assisted review through the lens of Sedona Conference Principle 6 and FRCP 26(g): the producing party is best situated to choose the review method, provided the process is documented, validated, and proportionate. No court has held that AI review is inherently non-defensible; no court has held that it is inherently sufficient without validation.
287 F.R.D. 182 (S.D.N.Y. 2012) | Judge Andrew J. Peck | 2012
Moore v. Publicis Groupe (Da Silva Moore)
The first U.S. judicial approval of Technology Assisted Review. Judge Peck held that predictive coding is 'acceptable' and may be 'more accurate than exhaustive manual review'. The opinion established the defensibility framework: process transparency, not document disclosure, is the standard. Seed-set disclosure was required in Da Silva Moore under the specific facts; subsequent cases (Rio Tinto) relaxed this.
Practical takeaway: TAR is an approved production methodology. Process documentation is the key defensibility element.
2013 WL 6405156 (N.D. Ind. 2013) | Judge Robert L. Miller Jr. | 2013
In re Biomet M2a Magnum Hip Implant Products Liability Litigation
Applied FRCP 26(b)(2)(B) proportionality analysis to a large MDL with a very large corpus. The court approved cost-burden shifting, finding that requiring full manual review of the entire corpus would be disproportionate. The opinion reinforced that TAR defensibility is evaluated on cost-benefit proportionality as well as process, and that courts will not second-guess the producing party's methodology choice when the process was reasonable.
Practical takeaway: Proportionality under FRCP 26(b)(2)(B) supports TAR even when recall is not perfect, as long as the cost-benefit analysis is documented.
306 F.R.D. 125 (S.D.N.Y. 2015) | Judge Andrew J. Peck | 2015
Rio Tinto PLC v. Vale S.A.
Three years after Da Silva Moore, Judge Peck addressed a motion to compel disclosure of the seed set used in the producing party's TAR process. He declined to require seed-set disclosure, holding that the producing party's documentation of its process was sufficient, and that TAR (including Continuous Active Learning) could proceed without disclosing the specific training documents. This significantly expanded the practical utility of TAR by removing the seed-set disclosure burden.
Practical takeaway: TAR/CAL is defensible without seed-set disclosure, provided process documentation is adequate.
2016 WL 4077114 (S.D.N.Y. 2016) | Judge Andrew J. Peck | 2016
Hyles v. City of New York
The requesting party moved to compel the producing party to use TAR rather than keyword search. Judge Peck denied the motion, holding that TAR is not mandatory -- even when it might produce more accurate results. The responding party retains the right to choose its review methodology (consistent with Sedona Principle 6), provided the choice is proportionate and the process is transparent. Courts will not mandate TAR over a reasonable producing-party methodology choice.
Practical takeaway: TAR is available but not mandatory. Courts follow Sedona Principle 6: the producing party chooses the method.
N.D. Cal. 2024-2025 | Ongoing (N.D. Cal.) | 2024-2025
EEOC v. Tesla
The first public-record U.S. case involving GenAI-assisted document review as part of the eDiscovery process. The court accepted the AI-assisted methodology subject to validation requirements -- the same framework applied in TAR cases. No new legal category was created; existing defensibility doctrine applied. This case is significant because it establishes that courts will treat GenAI review under the existing TAR defensibility framework, not as a novel untested methodology.
Practical takeaway: GenAI review is defensible under existing TAR case law. The same process documentation and validation requirements apply.
(N.D. Ill. 2024) | N.D. Ill. | 2024
In re Broiler Chicken Antitrust Litigation
An antitrust MDL in which the court addressed the discovery protocol governing AI-assisted review across multiple producing parties. The court required that each producing party disclose the AI methodology used, including whether GenAI tools were applied, and maintain an audit log of model versions and prompts. This case signalled that courts will begin requiring methodology disclosure for GenAI tools specifically -- going slightly further than the post-Rio Tinto standard for TAR.
Practical takeaway: In complex MDL matters, courts may require GenAI-specific disclosure including model version and prompt audit logs.
The FRCP framework
FRCP 26(b)(2)(B)
Limits discovery of electronically stored information from sources that are not reasonably accessible because of undue burden or cost. The party resisting discovery must show the burden; the requesting party must then show good cause. This is the proportionality hook that courts use to approve or deny TAR protocols.
FRCP 26(f)
Requires parties to confer at least 21 days before a scheduling conference to discuss the nature and basis of claims, the preservation of discoverable information, and any issues about disclosure or discovery, including the form of ESI production. The Rule 26(f) conference is the required opportunity to raise and agree on TAR and AI methodology.
FRCP 26(g)
Requires the attorney signing a discovery response to certify that after a reasonable inquiry, it is complete and correct. When AI assists in the review, the attorney signature on the response certifies the AI-assisted result. This is the 'reasonable inquiry' duty that requires attorney oversight and validation of AI outputs.
Fed. R. Evid. 502(d)
A court may order that inadvertent production of privileged or work-product material does not constitute waiver in the pending case or in any other federal or state proceeding. Essential backstop for AI-assisted privilege review where first-pass accuracy is 85-97%. The Peck model 502(d) order language developed through Da Silva Moore and Rio Tinto is the standard template.
FRCP 37(e)
Governs sanctions for failure to preserve ESI. A party that fails to take reasonable steps to preserve ESI that should have been preserved and cannot restore or replace it faces curative measures (on prejudice) or adverse inference instructions (on intent to deprive). AI legal hold systems must be documented under 37(e) standards.
The Sedona Conference framework
The Sedona Conference, a nonprofit legal policy research and education organization, has published a series of influential principles and commentary that courts frequently cite in eDiscovery disputes. The Conference is not a governmental body and its publications are not law, but they function as persuasive authority in courts that regularly handle complex commercial litigation.
Sedona Conference Principle 6
'Responding parties are best situated to evaluate the procedures, methodologies, and technologies appropriate for preserving and producing their own electronically stored information.'
Sedona Conference Cooperation Proclamation, 10 Sedona Conf. J. 331 (2009). Cited in Rio Tinto PLC v. Vale S.A., 306 F.R.D. 125 (S.D.N.Y. 2015) and Hyles v. City of New York, 2016 WL 4077114 (S.D.N.Y. 2016).
The Cooperation Proclamation (2009) established that adversarial conduct in discovery is inconsistent with professional obligations and the proper administration of justice, and that cooperation in discovery -- including agreement on ESI protocols and methodology -- is expected. This principle underlies the Rule 26(f) conference requirement and the expectation that TAR and GenAI methodologies be disclosed and agreed rather than sprung on opposing parties.
ABA Formal Opinion 512 (July 2024) -- brief
ABA Formal Opinion 512, issued 29 July 2024, addresses lawyers' ethical duties when using generative AI tools. While not case law, it is a formal opinion of the ABA Standing Committee on Ethics and Professional Responsibility and is directly relevant to AI-assisted eDiscovery. Key holdings: duty of competence (Rule 1.1) requires understanding AI capabilities and limits; duty of confidentiality (Rule 1.6) requires verifying vendor data handling including zero-retention terms; duty of candor (Rule 3.3) applies to AI-generated work product. See /ethics-confidentiality for the full eDiscovery application.
Frequently asked questions
What did Da Silva Moore hold?
Da Silva Moore v. Publicis Groupe, 287 F.R.D. 182 (S.D.N.Y. 2012) was the first U.S. judicial approval of TAR. Judge Peck held that predictive coding is acceptable and may be preferable to manual review. Process transparency was established as the defensibility standard -- not document disclosure.
What did Rio Tinto v. Vale hold?
Rio Tinto PLC v. Vale S.A., 306 F.R.D. 125 (S.D.N.Y. 2015) reinforced TAR defensibility and specifically held that TAR -- including CAL (TAR 2.0) -- could proceed without disclosure of the seed set. Process documentation is sufficient. This relaxed the burden significantly relative to Da Silva Moore.
Is AI review defensible in court?
Yes, subject to validation and process documentation. Courts following Da Silva Moore, Biomet, Rio Tinto, and the GenAI-era cases treat AI review under the same Sedona Principle 6 / FRCP 26(g) framework applied to TAR. EEOC v. Tesla (2024-2025) accepted GenAI review under this framework.
What does FRCP 26(g) require when AI assists in document review?
FRCP 26(g) requires the certifying attorney to attest that after reasonable inquiry the response is complete and correct. When AI assists, this duty requires the attorney to understand the AI methodology, review a statistically valid sample of outputs, and validate the AI results before signing. The attorney cannot delegate the reasonable inquiry duty to the AI system.