For the last couple of years, every conversation in this industry has started with the same question, which tool, which model, whose GenAI is smartest at finding the needle in the haystack. That conversation is getting stale. The tools have got good enough, fast enough, that "does it work" isn't really the interesting question any more. The interesting question, the one nobody wants to sit with, is whether anyone can actually govern what they've just switched on.

The adoption numbers alone tell you how quickly this crept up. Active use of generative AI in eDiscovery has jumped from around 12% of professionals to 37% in the space of two years, according to Everlaw's 2025 Ediscovery Innovation Report (produced with ACEDS and ILTA). That's not a slow rollout, that's a stampede. And from what I'm hearing in conversations with firms and vendors, the tools themselves have moved a long way past simple document review too, some are now drafting privilege logs with reasoning attached, flagging sensitivity issues, building chronologies, doing work that used to sit firmly with a human reviewer.

Here's the bit that should worry people more than it seems to. The latest EDRM/ComplexDiscovery Business Confidence Survey found roughly three in ten organisations running these tools in production are doing so with inconsistent or, in some cases, no documented rules at all. Courts haven't settled the basics yet either, whether AI can be relied on for final responsiveness determinations, how error rates get measured and reported, what counts as adequate human oversight. We're deploying the answer before we've agreed the question.

I wrote about data governance in eDiscovery last year, and the argument hasn't changed, it's just got more urgent. Back then it was about managing growing data volumes responsibly. Now it's about managing decisions an algorithm is making on your behalf, in a way you can stand behind if a judge, a regulator, or opposing counsel asks you to explain it.

And this isn't a 2026 problem that resolves itself by January. Agentic AI, tools that don't just suggest but act, is forecast by Deloitte to go from roughly a quarter of GenAI-using enterprises to half of them by 2027. The firms getting real value out of any of this aren't the ones with access to the flashiest model, they're the ones who've invested properly in the governance infrastructure sitting underneath it. That's not a hunch, PwC's research suggests nearly three quarters of AI-generated economic value is being captured by the fifth of organisations investing most heavily in governance and responsible AI.

So if you're evaluating a vendor right now, the sales pitch about capability is the easy part, ask about the harder part instead. Ask how they document decisions. Ask what their security architecture actually looks like, not just what the marketing page says. It's part of why I've been comfortable putting my name behind Merlin's Alchemy platform, single-tenant architecture and SOC 2 Type II certification aren't the most exciting words in a product demo, but they're the ones that matter when someone asks you to prove what happened and why.

The AI stopped being the story a while back. Whether we can govern it properly is the one that's going to define the next couple of years.

Sources: Everlaw 2025 Ediscovery Innovation Report: https://www.lawnext.com/2025/07/report-shows-37-of-e-discovery-professionals-now-using-ai-with-cloud-adopters-leading-the-charge.html

EDRM/ComplexDiscovery 1H 2026 Business Confidence Survey: https://complexdiscovery.com/from-deployment-to-discipline-ai-and-governance-in-the-1h-2026-ediscovery-business-confidence-survey/

Deloitte TMT Predictions, agentic AI: https://www.deloitte.com/us/en/insights/industry/technology/technology-media-and-telecom-predictions/2025/autonomous-generative-ai-agents-still-under-development.html

PwC value-capture research: https://evolvancemarketresearch.com/statistics/ai-governance-statistics/

#eDiscovery #GenAI #AIGovernance #LegalTechnology #DataGovernance #TheFortedBunker #Forted