Federal court applies traditional rules to AI document review

U.S. Magistrate Judge Laurel Beeler of the Northern District of California ruled that generative AI document review must be governed by existing technology-assisted review (TAR) standards, rejecting arguments that the technology requires novel discovery rules WilmerHale.

The decision, issued June 30, 2026, in *Schulte v. LinkedIn Corp.* (No. 22-cv-00237-HSG), addressed how courts should treat AI-powered document workflows in litigation. LinkedIn deployed Relativity aiR, a generative AI tool, to review documents in the case Complex Discovery. Judge Beeler's order treated the AI review process as analogous to traditional predictive coding and keyword filtering—established practices in federal discovery—rather than as uncharted legal territory.

TAR principles apply to generative AI

The court allowed parties to use search terms to pre-cull documents before feeding them into the generative AI review system HK Law. This mirrors longstanding TAR workflows, where keyword searches narrow the document universe before machine-learning models rank results by relevance. Judge Beeler found no reason to treat AI-generated rankings differently from algorithmic predictions courts have validated for nearly two decades.

Court rejects expanded discovery demands

Plaintiff's counsel sought expansive disclosure about how the AI model was trained, validated, and tuned—essentially demanding discovery about the discovery process itself. Judge Beeler rejected these requests as impermissible "discovery on discovery" absent a specific showing of deficiency in the AI review WilmerHale. Under TAR case law, a party challenging AI-assisted review must demonstrate concrete evidence of inadequate precision or recall, not merely request methodological transparency as a matter of course.

The ruling preserves the burden allocation that has governed technology-assisted discovery for years: the producing party bears responsibility for reasonable validation; the requesting party bears the burden of proving failure.

Implications for enterprise adoption

The decision signals that courts will not create special evidentiary or procedural regimes for generative AI in discovery, even as vendors like Relativity integrate large language models into litigation workflows Exterro. This removes one legal uncertainty that has slowed enterprise adoption of AI-assisted review tools—companies can now deploy these systems under familiar frameworks rather than negotiate novel protocols in every case.

The decision does not eliminate discovery burdens or validate any particular AI implementation; it simply means that generative AI tools will be measured against the same standards TAR systems have faced since predictive coding emerged in the early 2010s.