Sedona Canada and the New Reality of Technology-Assisted Review in 2026

For years, the question Canadian litigators asked about technology-assisted review was whether they were allowed to use it. In 2026, that question has quietly inverted. The more pressing concern is whether a party that ignores capable review technology is still meeting its discovery obligations. The Sedona Canada Principles have always pointed in this direction, and the arrival of reliable generative AI review has simply made the destination obvious.

This shift matters because it changes how you justify your choices. A decision that once needed defending now needs documenting. If you run discovery in Canada, the framework you already cite to opposing counsel supports modern AI review more naturally than it ever supported manual, document-by-document slogs through hundreds of thousands of files.

What Sedona Canada Actually Asks Of You

The Sedona Canada Principles were never a technology mandate. They are a set of expectations about reasonableness, and two of them carry most of the weight in any discovery dispute. The first is proportionality. The volume, cost, and effort of preserving and reviewing electronic information should be proportionate to what is genuinely at stake in the matter. The second is cooperation. Parties are expected to communicate early and in good faith about how electronic discovery will be conducted, rather than ambushing each other later.

Read those two principles together and a clear standard emerges. You are expected to choose review methods that are reasonable for the size of the matter, and you are expected to be transparent about the methods you chose. Neither principle says anything about which tools you must use. Both, however, ask you to justify your approach against the size of the problem in front of you.

That is the lens through which AI review should be evaluated. The question is not whether the technology feels novel. The question is whether your chosen method is a proportionate, defensible response to the document population you actually have.

From Predictive Coding to Generative Review

Canadian courts accepted technology-assisted review long before generative AI arrived. Predictive coding, built on supervised machine learning, has been recognized in Canadian and comparable common law decisions for over a decade as a reasonable way to prioritize and cull large review sets. The premise was straightforward. A human teaches the system what relevance looks like across a sample, and the system extends that judgment across the wider population.

Generative review changes the mechanics without abandoning the premise. Instead of learning relevance from a coded sample, a system like Claira reads each document against instructions written in plain language and explains its reasoning for the decision it reaches. The reviewer is no longer training a statistical model and hoping the boundary holds. The reviewer is giving direction, reading the rationale Claira returns, and correcting it where the reasoning is wrong.

We have written before about the practical philosophy behind this approach, and the continuity with earlier technology-assisted review is the part most worth emphasizing here. Courts did not approve predictive coding because the math was elegant. They approved it because it was a reasonable, supervised method that a competent professional could explain and defend. Generative review meets the same test, and it produces something predictive coding never could, which is a readable explanation for each call.

Why AI Review Is Now the Defensible Baseline

The defensibility argument used to run in one direction. A party using technology-assisted review had to show that the method was sound. That burden has not disappeared, but a second question now sits beside it. Can a party reviewing six or seven hundred thousand documents by hand still claim its method is proportionate when faster, more consistent, and equally explainable alternatives exist?

Manual linear review is not inherently more accurate. Reviewer fatigue, inconsistency across large teams, and the sheer cost of human hours all cut against it. When a matter is large enough, exhaustive manual review can become the disproportionate option, not the safe one. That is the reversal worth internalizing. The defensible baseline in a high-volume Canadian matter in 2026 is a supervised, well-documented AI review, with human oversight concentrated where judgment actually matters.

This does not mean AI replaces lawyers, and it does not mean every decision is delegated to a machine. It means the proportionality calculus has moved. A reasonable litigator now starts from the assumption that capable review technology will be part of the workflow, then explains the human controls layered on top of it.

Documenting Your Process for Opposing Counsel

A defensible method is only as strong as your ability to describe it. This is where the cooperation principle becomes a practical advantage rather than an obligation. If you can explain your process clearly and early, you convert a potential dispute into a routine disclosure.

Start by recording the inputs that shaped the review. When you give Claira background on the matter through Case Context, you are creating a documented, repeatable foundation that every scan draws on. That record is exactly what opposing counsel and the court will want to understand. It shows the review was directed by the issues in the case, not improvised document by document.

From there, the elements of a defensible record are familiar. Capture the instructions and prompts you used. Note where humans reviewed, sampled, and corrected the system's output. Record how you validated results before relying on them. Because generative review returns a rationale for each decision, you can point to the reasoning behind a classification rather than asking anyone to trust an opaque score.

The goal is not to overwhelm the other side with technical detail. The goal is to be ready to answer the only questions that matter. Was the method reasonable for the matter, and can you show the work. Sedona Canada asks for cooperation and proportionality. A documented AI review answers both at once.

The New Default

The center of gravity in Canadian eDiscovery has moved. Technology-assisted review is no longer the choice you defend against a skeptical court. It is increasingly the choice you would have to defend abandoning. The Sedona Canada Principles, read honestly, have always rewarded the party that pursued proportionate methods and communicated about them openly.

Generative review fits that framework better than any tool before it, because it pairs scale with an explanation a human can read. If you want to see how this works inside the platform your team already uses, you can explore Claira and the way it runs review directly within Nuix Discover. The principles have not changed. The reasonable response to them has.