Global-wide: A Litigation Services Overview

Defensibility Under Pressure: Litigation Risk, AI and Operational Reliability in 2026

Litigation clients in highly regulated industries are operating under three converging pressures. The regulatory environment has fragmented across jurisdictions and become harder to predict. Organisations are under sustained pressure to do more with less while demonstrating measurable efficiency gains from AI investment. At the same time, the AI tools deployed to meet those expectations introduce the risk of failures that often do not obviously reveal themselves as errors and whose liability, when something goes wrong, rests with clients and counsel rather than vendors.

Each of these pressures would independently affect litigation strategy. Their convergence is now reshaping how legal teams evaluate risk, structure workflows, allocate resources and assess defensibility. The organisations navigating this environment most effectively are not simply the ones adopting AI most aggressively or avoiding it altogether. They are the ones redesigning litigation workflows to remain reliable and defensible when regulatory pressure, operational complexity and AI-assisted execution intersect.

A fragmented regulatory landscape

The regulatory pressure is easy to misread. Some organisations have interpreted shifts in federal enforcement priorities under the current administration as a sign that the overall regulatory environment has softened. In reality, many state attorneys general and state agencies have moved aggressively to fill the resulting gaps, creating an enforcement landscape that is spread across more jurisdictions and harder for legal teams to anticipate and manage.

The result is not reduced enforcement, but fragmented enforcement. The 2026 Norton Rose Fulbright Annual Litigation Trends Survey found that more than eight in ten general counsel reported increased state enforcement activity in response to shifting federal priorities. That fragmentation affects not only where enforcement originates, but also how predictable litigation exposure becomes across industries and jurisdictions.

The US Supreme Court’s decision overturning Chevron deference has compounded that complexity. Agency interpretations are now more vulnerable to judicial challenge; governance teams are recalibrating risk assumptions; and litigation strategy in some sectors has shifted toward more direct judicial engagement on regulatory questions. Even where overall case counts may decline in particular categories, the environment has become more difficult to assess and harder to prepare for. Fragmentation also increases discovery complexity, making it more difficult to determine proportionate discovery and requiring earlier and more rigorous case assessment.

Operational pressure and the AI liability gap

Running alongside this fragmented regulatory landscape is a sustained operational pressure that legal teams now share with the broader enterprise. Organisations are expected to deliver greater efficiency while demonstrating measurable returns from AI investment. In many companies, anticipated AI-driven productivity gains have already been incorporated into budgeting and resourcing assumptions. The question is no longer whether legal teams will adopt AI, but whether legal operations will deliver the expected economic impact.

The tools available to support those expectations are genuinely capable. AI-assisted document review, early case assessment, issue tagging, summarisation and identification of privileged communications have materially changed the economics of large-scale discovery. Adoption reflects this shift. The Association of Corporate Counsel’s 2025 Chief Legal Officer Survey found that generative AI use by in-house legal departments more than doubled between 2024 and 2025.

But beneath that adoption curve sits a structural feature of the market that many organisations have not fully absorbed. AI vendors frequently disclaim accuracy obligations and cap liability at the fees paid for the software. When AI-assisted work product fails because of a missed obligation, an inaccurate summary, inconsistent treatment of similar provisions, or incomplete review coverage, the vendor generally does not bear responsibility for the consequences. Counsel and their clients do.

This is not a hypothetical concern. Courts and disciplinary bodies are already confronting these issues, and organisations using AI-assisted workflows should expect scrutiny to increase. The economic case for AI is real. So is the reality that responsibility for the reliability and defensibility of work product ultimately remains with the lawyers and organisations relying upon it.

Invisible failure modes in AI-assisted workflows

That responsibility is harder to discharge than many legal teams initially assume because the failures most likely to create litigation risk are often not the ones that receive the most public attention. The profession’s focus has understandably centred on hallucination: fabricated citations, invented authorities and false quotations. Courts have imposed sanctions; bar associations have issued guidance; and lawyers have learned to verify what AI systems generate. These are visible failures, and the profession’s response has been appropriate.

In complex litigation workflows, however, the failures most likely to create material risk are often the ones that appear correct on their face. The problem is frequently not what the system produced, but what it failed to surface.

Three patterns appear repeatedly.

The first is a coverage failure. An AI system may accurately process every document it was given while failing to identify a critical provision or communication that sat outside the defined review boundaries. The output appears complete, and no obvious signal reveals the omission. In practice, that may mean accurately summarising communications within one custodian set while missing contradictory evidence held elsewhere.

The second is failure caused by an overly narrow review scope. An AI system may correctly answer every question posed within a defined document set while missing qualifying or contradictory information located in adjacent materials that were never included in the review. The issue is not that the analysis was inaccurate. The issue is that the workflow itself failed to examine the broader context. In litigation, those problems often do not become visible until deposition preparation, expert analysis, or motion practice exposes gaps sitting immediately outside the original review parameters.

The third is inconsistency across large-scale reviews. Across large document populations reviewed over multiple sessions or by different users, AI systems may apply materially different standards to equivalent provisions or fact patterns. No single output necessarily appears incorrect in isolation. The inconsistency becomes visible only through deliberate comparison across the broader workflow. The result can include inconsistent privilege determinations, divergent responsiveness decisions across review teams, or uneven issue tagging that affects downstream litigation strategy and factual development.

These are not edge cases. They are structural characteristics of how AI systems operate in complex legal workflows, particularly in the high-volume, high-pressure matters where the efficiency rationale for AI is strongest. They also create risks that standard end-stage quality checks often fail to detect because the problem frequently lies in review boundaries, workflow calibration and operational consistency, rather than isolated output errors.

Designing AI-assisted workflows for defensibility

Legal teams successfully navigating this environment are redesigning litigation workflows around operational defensibility rather than relying solely on downstream review. The critical question is no longer whether AI-generated outputs receive human review. It is whether the workflow itself was structured to detect omission, inconsistency and boundary failures before work product is relied upon.

In practice, that requires legal teams to treat workflow design decisions as substantive legal judgments. Review scope, system instructions, escalation protocols, validation measures and consistency controls all shape the reliability of the resulting work product. Organisations managing these risks effectively are embedding verification throughout execution rather than applying review only at the end of the process.

Professional responsibility rules increasingly intersect with these operational design decisions. The American Bar Association’s Formal Opinion 512 confirms that the use of AI does not diminish counsel’s duties of competence, supervision, or candour. But as AI systems become more integrated into litigation workflows, the harder question is no longer whether lawyers reviewed the output. It is whether the workflow itself was designed in a manner capable of producing reliable and defensible work product at scale.

A sign-off process alone has limited value if those providing it cannot meaningfully evaluate the completeness, consistency and reliability of the underlying work product. Under scrutiny, legal teams may increasingly need to explain how review boundaries were determined, why workflow assumptions were reasonable, what validation measures were applied, and how consistency was maintained across the matter.

The operational risks created by these failures are not theoretical. They directly affect discovery proportionality decisions, privilege assertions, deposition preparation, regulatory response strategy, and the defensibility of representations made to courts, regulators and counterparties.

The organisations best positioned under judicial, regulatory and client scrutiny will not necessarily be those adopting AI most aggressively. They will be the ones able to demonstrate that their litigation workflows were operationally designed to remain reliable, consistent and defensible under pressure.