At Relativity Fest, I sat down with Cristin Traylor, Senior Director of AI Transformation and Law Firm Strategy at Relativity to talk about how generative AI is rapidly transforming the legal landscape. What once felt like tentative exploration has shifted to widespread adoption, and Traylor offered compelling insights into the practical and strategic considerations firms face at this time.  

Defensibility and Discoverability of Generative AI

One of the biggest questions in the legal community right now is how generative AI will be treated in court, especially when it comes to disclosure and potential challenges from opposing parties. The discussion below outlines a clear strategic framework for establishing defensibility.

The Argument for Prompts as Work Product

A significant point of debate is whether prompts used to instruct a generative AI model should be discoverable. The argument is made that they should be protected as privileged work product.

The analogy is straightforward: AI prompts aren’t just search terms defining what to review – they’re more like a review protocol, instructing the system on what counts as responsive. In Relativity aiR for Review, the AI acts as a “super reviewer,” and just as training materials and review protocols for a team of human attorneys are considered privileged work product, so too should the instructions given to the AI.

Traylor notes, “no one ever asks for your review protocol when you’re training humans, so why are we asking for it over here?”

The Primacy of Validation Over Process Disclosure

Traylor suggests that the industry may be asking the wrong question when it comes to generative AI. Instead of focusing on whether prompts are discoverable, she argues that the real emphasis should be on the validation of the AI’s results. This shifts the focus from “how” the review was done to “how well” it was done.

A defensible AI process relies on robust validation. This involves both quantitative metrics and qualitative checks to ensure the tool is being used responsibly. Transparency can be achieved by sharing key performance metrics with the opposing party. As Traylor put it, “I’m going to validate in a proper manner and I’m going to tell you that I got 95% recall and 75% precision and 1% elusion.”

This approach aligns with the longstanding legal principle that discovery must be reasonable, not perfect. If questions arise about missing documents, the usual inquiry and follow-up processes still apply.

The Role of the Courts and Technical Expertise

Traylor cautions against relying on court cases to set broad precedents for AI use, calling it both inefficient and reminiscent of the early days of technology-assisted review (TAR). Expecting courts to approve every new tool doesn’t make sense – especially with AI evolving so rapidly. Judges simply aren’t positioned to act as technology assessors.

When AI issues do reach a judge, Traylor emphasizes the importance of bringing in technical expertise. As Judge Andrew J. Peck puts it, legal teams need to “take your geek to court.” An expert can demystify the process, explain the technology’s reliability, and show that AI is not some indecipherable black box.

Law Firm Adoption and AI Integration

The adoption of generative AI among law firms has become widespread but varied, falling along a spectrum outlined below.

Adoption Category	Characteristics
Early Adopters	These are “tech forward” firms that are “all in” on AI. They are using tools like Relativity aiR for Review and Relativity aiR for Privilege in every matter by default, unless a specific reason exists not to.
Middle Adopters	These firms are interested but more cautious. They are “getting their toes wet” by testing AI on older, dormant cases, on opposing party productions, or as a quality control check after a human review. They often need more guidance to get started.
Laggards	This group consists of more conservative firms that may not even be on cloud platforms. Technology adoption is not a high priority, and they are slower to embrace new tools.

Key Adoption Metrics and Trends

The adoption of AI in e-discovery has grown dramatically. According to Traylor, aiR for Review has already processed over 30 million documents and recorded 120 million review decisions.

Looking ahead, she expects usage to climb even higher as AI becomes a standard,“baked in” part of the RelativityOne offering. The value provided by AI-generated rationale, considerations, and citations for downstream tasks like case strategy is a significant driver of this anticipated growth.

Ethical Framework for AI in Legal Practice

Responsible AI adoption in law starts with ethics, and Traylor emphasizes that existing rules provide a strong foundation. She points to ABA Formal Opinion 512 as core guidance, citing this opinion as a useful resource that connects AI usage to foundational ethical duties like confidentiality, security, and the duty to supervise.

Traylor describes ABA Rule 1.1 as “the keystone.” Attorneys have a duty of competence, which requires them to understand the AI tools they’re using, but this does not mean they need to be experts in how large language models are built. They do need to understand what these tools are doing, how they’re working, and if they’re working.

Practical safeguards are essential, like ensuring protective orders contain language that prevents client documents from being uploaded to public AI models. Supervision and quality control remain critical. An attorney cannot simply put it in, press the button, and walk away. Traylor emphasizes that rigorous quality control and validation are essential ethical obligations.

The Challenge of AI Hallucinations in Legal Filings

AI can sometimes generate false or nonexistent case citations, and Traylor warns this is a real concern, especially in U.S. legal filings. Addressing this challenge requires what she calls a “new type of lawyering.” Every AI-generated assertion or citation must be carefully verified against the source material.

Hallucinations can be difficult to catch. Sometimes an AI tool will correctly identify a case but then invent quotes or holdings within it, making a surface-level check insufficient. Ultimately, the responsibility remains with the lawyer to ensure the information included in briefs is accurate and reliable.  

Comparative Analysis: Generative AI vs. TAR

Generative AI is often compared to earlier forms of TAR. While there are overlaps, key distinctions make generative AI a significant step forward.

One school of thought takes an evolutionary view, seeing generative AI as simply another form of TAR. If this perspective is accepted, existing case law permitting the use of TAR could apply, potentially removing the need for a new precedent-setting case.

A major advantage of generative AI over older TAR models is its transparency. Traylor explains that with generative AI, users can see the rationale, considerations and citations behind each decision, so they understand why a document was flagged. This contrasts with older TAR processes where users would receive documents without a clear explanation of why they were deemed relevant.

Looking ahead, while traditional TAR may still have niche applications or appear in hybrid workflows, Traylor expects its use to decline significantly as generative AI adoption rises. For new matters, her preference is clear: generative AI all the way.

The Strategic Horizon: AI for Knowledge Construction

The most exciting frontier for AI in law lies beyond document production – into high-level case strategy. Traylor emphasizes that the ultimate goal of discovery isn’t just producing documents but truly understanding the data so attorneys can advise their clients on the best path forward.

Relativity aiR for Case Strategy brings AI in at the very start of a case to handle tasks that were once labor-intensive and manual. It can create timelines and chronologies, develop witness summaries and statements, and automatically extract key facts from a dataset.

By automating these knowledge-construction tasks, AI empowers legal teams to visualize the critical elements of a case or investigation, learn from their data more efficiently, and build a more informed, effective strategy.

Conclusion

As generative AI continues to mature, its impact on the legal industry is becoming undeniable. From streamlining document review to enabling high-level case strategy, these tools are reshaping how attorneys work.

Traylor’s insights make it clear that the real challenge lies beyond the technology itself. It’s about how law firms integrate it responsibly – validating outcomes, maintaining ethical standards, and ensuring attorneys remain competent and engaged with the process.

The future of AI in law is not about replacing human judgment but amplifying it. By turning data into actionable knowledge, tools like aiR for Case Strategy allow legal teams to move faster, make smarter decisions, and ultimately provide more effective guidance to clients.

In the end, effective implementation of generative AI in law requires a shift in mindset. Those who embrace its power alongside invaluable human expertise may soon look back and wonder how they ever managed without it.

So, what do you think? What do you think the state of eDiscovery is? Please share any comments you might have or if you’d like to know more about a particular topic.

Note: This write-up from a recorded interview was prepared with the assistance of NotebookLM from Google and reviewed and edited by both Cristin and me for accuracy.

Disclaimer: The views represented herein are exclusively the views of the author, and do not necessarily represent the views held by my employer, my partners or my clients. eDiscovery Today is made available solely for educational purposes to provide general information about general eDiscovery principles and not to provide specific legal advice applicable to any particular circumstance. eDiscovery Today should not be used as a substitute for competent legal advice from a lawyer you have retained and who has agreed to represent you.