You've heard it in vendor pitches, conference panels, and cautious internal memos: "We're waiting for the case law to catch up." The assumption is that generative AI in discovery review needs its own judicial blessing before your team can deploy it with confidence. That assumption is wrong, and it's costing you time.
The legal precedent already exists. It's been settled for over a decade. The confusion comes from treating each technological iteration as if it requires a fresh stamp of approval, when the courts have consistently ruled on the principle, not the implementation details.
Myth 1: Da Silva Moore Only Approved Predictive Coding
The Reality: When New York Magistrate Judge Andrew J. Peck ruled in Da Silva Moore v. Publicis Groupe & MSL Group in February 2012 that "computer-assisted review is an acceptable way to search for relevant ESI in appropriate cases," he didn't write a permission slip for one algorithm. He approved the use of technology to assist review.
The protocol negotiated in that case used what we now call TAR 1.0. But Judge Peck cited the Grossman & Cormack study advocating for continuous active learning approaches within that same ruling. He wasn't limiting approval to the specific workflow the parties happened to use. He was establishing that courts will accept technology-assisted methods when they're appropriate to the case.
By May 2015, Judge Peck stated in Rio Tinto PLC v. Vale S.A. that "it is now black letter law that where the producing party wants to utilize TAR for document review, courts will permit it." That was over eleven years ago. The principle was settled then.
Myth 2: GenAI Is Different Enough to Need New Case Law
The Reality: Predictive coding and generative AI are both forms of artificial intelligence applied to document review. They're both technology-assisted review methods. The underlying legal question is the same: Can you use computational methods to identify responsive documents instead of having humans review every page?
The answer has been yes since 2012.
In Schulte v. LinkedIn Corp., decided in July 2024, California Magistrate Judge Laurel Beeler referred to Relativity aiR explicitly as "a form of Technology Assisted Review." She didn't treat it as a novel category requiring separate validation. She applied the existing TAR framework and ruled on whether LinkedIn's specific disclosure obligations were met.
That's how courts work. They rule on principles, then apply those principles to new circumstances. The principle that technology can assist review is settled. What varies case-by-case is whether your specific approach is defensible given your data volume, case complexity, and opposing counsel's concerns.
Myth 3: You Need a Court Order Before Using GenAI Tools
The Reality: You don't need judicial permission to use GenAI for review any more than you needed permission to use TAR 1.0 after 2015. What you need is a defensible process and appropriate disclosure to opposing counsel.
The Schulte ruling reinforced that the producing party's obligation is to disclose when they intend to use TAR to filter non-responsive documents. LinkedIn disclosed its use of Relativity aiR. The court evaluated whether that disclosure met the requirements of the interim ESI order. It did.
Your workflow decisions don't require pre-approval. Your disclosure and cooperation obligations do require attention. If your ESI protocol or standing order requires TAR disclosure, you disclose. If opposing counsel raises specific objections to your approach, you negotiate or litigate those objections. But you're not waiting for a landmark case to greenlight the technology category itself.
Myth 4: Each New TAR Approach Needs Its Own Precedent
The Reality: Continuous active learning didn't need a separate landmark case after TAR 1.0 was approved. Judge Peck's 2015 Rio Tinto ruling noted that the seed set in CAL "is much less significant" than in TAR 1.0, but he didn't treat CAL as requiring fresh judicial validation. He explicitly said the Da Silva Moore protocol wasn't meant to be "appropriate in all [or any] future cases that utilize [TAR]."
Courts evaluate whether your chosen method is reasonable for your case. They don't maintain a registry of approved algorithms. The Sedona Conference TAR Case Law Primer documented at least 85 TAR-related rulings as of May 2023. Those cases addressed specific disputes about seed sets, validation methods, disclosure timing, and quality metrics. None of them said, "This new variant of TAR is hereby approved for the first time."
The pattern is clear: courts rule on process disputes, not technology categories.
Myth 5: "AI" and "TAR" Are Separate Legal Standards
The Reality: The terminology confusion is understandable but irrelevant to the legal question. We use "TAR" as shorthand for older predictive coding methods and "AI" for newer generative approaches. But both are technology-assisted review. Both are forms of AI. The legal standard doesn't change because the marketing language does.
When Judge Beeler called Relativity aiR "a form of Technology Assisted Review," she wasn't coining a new legal category. She was applying the existing one. If your team is waiting for a case that says "generative AI is approved," you're waiting for a distinction the courts aren't making.
What to Do Instead
Stop waiting for permission you don't need. Start building defensible processes.
Document your GenAI workflow the same way you'd document any TAR approach. What's your sampling method? How are you validating recall and precision? What quality control steps catch errors the model misses? When and how do you disclose to opposing counsel?
If your case involves an ESI protocol or standing order, review its TAR disclosure requirements. Meet them. If it doesn't mention TAR at all, consider whether proactive disclosure reduces later disputes.
Train your team on the difference between the legal question (is technology-assisted review acceptable?) and the case-specific question (is our implementation defensible?). The first is settled. The second requires the same rigor you'd apply to any discovery method.
And when someone tells you we're still waiting for the "Da Silva Moore of Generative AI," you can correct them. It happened in February 2012.
