Why EDRM 2.0 Matters in the AI Review Era

On September 1, 2026, EDRM released EDRM 2.0, the first substantive update to the Electronic Discovery Reference Model since it absorbed the Information Governance Reference Model. Roughly 150 practitioners from across the global EDRM community contributed to the revision. For a diagram that has hung on litigation support walls since 2005, that is not a cosmetic refresh.

The question worth asking is not whether the new model looks different. It is whether the changes tell your team anything useful about how AI-assisted review should actually run. We think they do. The most consequential change is also the easiest to scroll past.

What actually changed

Four updates define the new model.

Information governance now sits underneath the entire lifecycle as a foundational layer rather than a phase you pass through on the way to collection. Identification, Preservation, Collection and Processing are grouped into a single Data Acquisition framework. Disposition appears as a discrete phase at the end of the lifecycle. And Analysis, formerly one box among many, is now represented as an activity spanning every phase of discovery.

EDRM describes that final change as making Analysis the connective tissue among iterative phases, reflecting the growing role of legal judgment, analytics, data science and AI throughout the lifecycle. Read quickly, it sounds like a diagramming preference. Read carefully, it is a claim about where the work of discovery now happens.

One practical note before going further. The EDRM 2.0 diagram remains licensed under Creative Commons Attribution 4.0, so your team can adapt it for internal protocols, training decks and client explainers without seeking permission. Attribution back to edrm.net is the only requirement, along with a note if you have modified it.

Continuous Analysis is the change that matters most

The previous model invited a particular mental habit. You identified, preserved, collected and processed. You reviewed. Then you analyzed what review had produced. Analysis was a downstream station, something that happened once the documents had been sorted and the expensive part was over.

Under EDRM 2.0, analysis runs alongside everything. That reframing describes what teams working with AI review have been doing in practice for some time, often without a model that acknowledged it.

Consider what a modern first pass produces. A single pass across a population can tag responsiveness, flag privilege risk, extract dates and authors, and summarize content at the same time. The output is not merely a sorted pile awaiting analysis. It is analysis, generated at the moment of first contact with the documents, available to inform scoping decisions that in the old diagram had already been made and locked.

This is where EDRM 2.0 and other recent frameworks converge. We have written before about how the Sedona Conference's TAR 1 Reference Model argues that generative AI does not require a new review framework, because it fits inside the defensible process litigation teams have used for over a decade. EDRM 2.0 makes a compatible point at a higher altitude. The phases have not been replaced. The relationship between them has become iterative, and analysis is what holds the iteration together.

Data Acquisition and the end of strictly sequential steps

Grouping Identification, Preservation, Collection and Processing reflects a reality most practitioners already recognize. Modern tooling lets these activities operate together rather than in strict sequence, particularly for producing parties as discovery work moves upstream.

The practical consequence is that decisions made early now carry more weight, because there are fewer distinct checkpoints at which to correct course. If your scoping is vague at acquisition, that vagueness propagates through a workflow that no longer forces a hard stop between processing and review.

This is an argument for fixing the definition of the matter before volume arrives, not after. In Claira, that work lives in Case Context, where your team records the background, the issues and the responsiveness criteria a matter turns on. That context then carries through every subsequent pass, so the standard applied to document ten thousand is the standard applied to document one. Under a model where phases blur together, a fixed and documented definition of scope is not administrative tidiness. It is the thing preventing drift.

Disposition, the phase everyone used to skip

Disposition earning its own box is overdue, and it has a wrinkle specific to AI-assisted review that is worth naming.

EDRM frames Disposition as the practical, legal and ethical responsibility to address data at the end of the lifecycle when it is no longer required. Most teams have some version of this in their retention policy. Fewer have thought carefully about what AI review adds to the pile.

An AI-assisted matter generates derived material that did not exist at collection. Summaries, extracted coding, chronologies, reasoning captured alongside each call. That material is genuinely useful during the matter and it is also data, subject to the same end-of-life questions as the documents it describes. If your disposition protocol was written before your review workflow produced these artifacts, it is probably silent on them. EDRM 2.0 gives you a reason to revisit it, and a named phase to hang the conversation on.

Mapping this to the work inside Nuix Discover

None of this requires a new platform. It requires knowing where each activity sits.

If your team runs review in Nuix Discover, the mapping is reasonably direct. Acquisition and processing happen in your existing environment. Claira performs the AI review inside it, which is the point at which continuous analysis becomes something you can actually demonstrate rather than assert. A bulk pass that codes for responsiveness while extracting objective fields and surfacing privilege risk is EDRM 2.0's continuous analysis expressed as a workflow rather than a diagram.

The defensibility argument improves too. When opposing counsel or a judge asks how your process worked, being able to point to a named phase for each stage, with a consistent scope definition and a visible record of how each call was reached, is a stronger position than describing review as a black box that produced tags.

Where this leaves your team

EDRM 2.0 is not a mandate and it does not change any rule. It is a consensus description of how discovery is practiced now, assembled by people doing the work across roles and jurisdictions. Its value is as a shared vocabulary, which matters most in exactly the conversations where precision is hardest, such as explaining an AI-assisted workflow to a client, a regulator or a court.

Our suggestion is modest. Pull the new diagram, put it next to your current review protocol, and look for the places where your protocol still assumes analysis waits until the end. Those are the places where your process and the profession's model have quietly diverged.

If you would like to see what continuous analysis looks like on real documents, book a fifteen-minute walkthrough and we will run a responsiveness criterion across a sample from one of your own matters inside Nuix Discover.