From Siloed Tools to Seamless Workflows: How AI Is Reshaping Disputes and Investigations

This second installment in our blog series about AI in litigation and investigations focuses on how to mitigate risks introduced by AI’s shifting role in legal practice: from point solutions for one-off tasks in a wider context to end-to-end workflows that can run from early risk detection to final work product. As detailed below, thoughtful design and implementation of mitigation strategies can limit risk associated with AI solutions while also offering gains in efficiency, consistency, and strategic insight for companies managing complex and often critical disputes and investigations.

The Shift from Tools to Infrastructure 

Until recently, the legal industry relied on standalone tools for specific tasks. E-discovery platforms, legal research databases, and summarization tools each accelerated their respective functions within the broader disputes and investigations workflow. Each has also become more powerful for specific tasks with the addition of built-in AI capabilities. 

The advent of agentic AI and workflow orchestration marks a break with that task-based model. AI is no longer just a tool to speed up particular tasks—deployed correctly, it has the potential to serve as the strategic infrastructure for the management of entire legal matters. For instance, an AI-enabled workflow can streamline an entire case by ingesting data, classifying materials based on that data, proposing review priorities, suggesting key issues and themes, and ultimately generating draft pleadings.

So what risks should companies and counsel look out for?

• Potential for cascading errors: While traditional, manual processes—including those enabled and improved by AI—often involve a human safety net at each stage, automated AI-driven workflows may not. One omission or mistake early on, such as a missed or mislabeled set of documents, can create a flawed, powerful point of influence that impacts how a team prioritizes certain facts, themes, or strategies in their court filings or presentations to regulators.
• Impact of hidden distortion on case theory, factual development, and advocacy: Because AI systems are trained on data that may reflect incomplete information, existing human prejudices, or other flaws, AI workflows may operate based on certain priorities or distortions that affect output but are not readily apparent to or detectable by users. Even without an obvious omission or mistake, this distortion can materially influence case theory, defenses, and strategic decision-making, affecting which claims or defenses are pursued or abandoned, which facts are emphasized in pleadings, and how findings are presented to regulators. By influencing what is reviewed first, which issues are prioritized, or how facts are summarized, AI-enabled workflows may craft or skew a case theory in a manner that crowds out nuances a human team would otherwise consider and in a manner that ultimately finds its way into deliverables.
• Questions around accountability and explainability: As AI becomes increasingly embedded in the end-to-end process of bringing and defending claims, conducting internal investigations, and responding to regulators, courts and regulatory authorities are focusing more closely on how attorneys conducted their review or prepared their deliverables. Counsel relying on AI-enabled workflows to drive their litigation and investigations must therefore be prepared to demonstrate technological literacy and defend their approach: how they selected and configured tools, how they prompted or trained the AI, and how human judgment and/or oversight were involved in the process. As the legal industry shifts away from siloed AI-powered tools, this scrutiny will only intensify as automation expands across core litigation, arbitration, and investigative functions. 

Well-designed workflows for disputes and investigations can manage these risks, while enabling companies and counsel to reap the benefits of new capabilities. 

So what can companies and counsel do?

• Implement defined and systematic human checkpoints: Before and during deployment of AI workflows in disputes and investigations, it is essential to identify specific points where counsel can pause, assess, and—as needed—override AI outputs to keep the rest of the process running smoothly and accurately.
• Ensure transparent documentation: Teams should consider keeping clear, contemporaneous records of their AI deployment in disputes and investigations, explaining the design and implementation of their workflow architecture, from models and prompts used; to training approaches; to sampling, quality control, and human-in-the-loop measures taken; and beyond.
• Test AI workflow performance in specific contexts: Teams may wish to test the workflow infrastructure under varied, realistic scenarios to identify potential gaps, shortcomings, or failure points in the context of their disputes and investigations. What performs well in a domestic dispute may not work in a cross-border investigation involving multiple languages, strict data protection regimes, or other unique aspects. Pressure testing will help ensure that the AI-enabled workflow being deployed is robust, adaptable, and aligned with teams’ unique needs and risk profiles well before filing day. 

By mindfully designing and governing AI workflows in the context of disputes and investigations, companies and counsel can benefit from the strategic upside of this technology—achieving better, more efficient outcomes in their legal matters—while taking steps to identify and contain potential risks.

To receive the latest insights on US legal developments, subscribe to the Freshfields A Fresh Take Blog.