Litigation and eDiscovery teams have used analytics for years, including technology-assisted review (TAR) and predictive analytics. Generative AI and agentic workflows now give the technology a more autonomous role, and data volumes keep growing. At the UnitedLex Summit 2026 in London, the Litigation panel, “AI in eDiscovery, Evidence, and Ethics,” asked what that change means for defensibility, for ethics, and for the lawyers who stay accountable for the outcome.
In “Beyond the Courtroom: The Evolving Role of Litigation Leadership“, the 2025 panel argued that litigators need to be fluent with technology, that AI will support lawyers instead of replacing them, and that business partnership must never override a litigator’s duty to the court. This year’s panel moved from the litigator’s role to the evidence itself.
Nicole Colonna, SVP, Litigation and Investigations Operations at UnitedLex, hosted the panel, and Savitha Masson, SVP, Enterprise Strategy & GTM at UnitedLex, moderated for the second year in a row. The panelists were Amir Bashir, Vice President, eDiscovery at Swiss Re; Kaan Gürer, Partner at Linklaters; and Anthony Kenny, Assistant General Counsel at GSK.
Speed is proven, and completeness is the question
Last year, a panelist described how AI can help a team assess the evidential merits of a matter far faster than before. This year, Gürer, who works on competition law investigations, took that as a given and asked the next question.
“Can we trust what AI finds? And can we really be sure that it found what we asked it for ?” Gürer asked.
In one recent matter, an AI-assisted review kept returning the same plausible set of documents and missed a significant body of material the team knew existed. The provider blamed training bias: a system tuned to find strong examples in one category may lack the “fresh mind” needed to find the unknown. Authorities expect a complete production, so that black box is the problem.
AI is not a source of truth
Bashir described a measured approach in-house. The team uses AI to prioritize reviews, so a potentially relevant set comes forward before large volumes of non-relevant material. Results depend heavily on the quality of the prompt, and one principle is fixed.
“But one thing we’re really clear about with all our stakeholders is that the AI is not a source of truth,” Bashir said.
Bashir also warned against business users querying their own data with general-purpose assistants before the data reaches the eDiscovery team. Those tools tend to give people the answer they want instead of the actual answer.
Kenny described a different use. Combining personal expertise with internal AI tools, Kenny answered a regulatory question from the business in about half an hour, work that used to take half a day. Kenny also uses AI as a sparring partner, asking it to take a regulator’s position and argue against Kenny’s case. On one occasion, that surfaced a case Kenny had not yet considered.
So is AI a teammate? Gürer called it a mixture. AI can hallucinate, so lawyers first have to establish whether evidence is real. At the same time, agentic workflows can be more transparent than human work, because every step can be traced and corrected.
Build the foundation junior lawyers used to get from review
With the threat of training diminishing with AI, the panelists looked at what juniors may lose along the way. Document review used to teach the fundamentals. Gürer warned that lawyers who start their careers with AI may miss that foundation, along with the experience and strategic foresight it builds.
Kenny argued that the deeper skill is curiosity: knowing which questions to ask, including whether a piece of evidence may itself be AI-generated. Bashir expects tools to accept more plain-language questions over time. When that happens, detailed knowledge of the matter will count for more than search technique.
Be ready to disclose and explain
Last year, panelists cautioned that litigators who act as business partners must remember their duty to the court. That duty shaped this year’s discussion of disclosure. Panelists saw little need to disclose AI used to polish an email. More substantive use, especially use that changes the time and cost of work, should be disclosed to clients. As opposing counsel, Kenny would also want to know when AI was used and would expect disclosure to the court.
Bashir added that teams should be able to quantify and describe exactly what role AI played. In Bashir’s process, AI helps prioritize review and humans make the decisions.
Treat prompts as potential evidence
Gürer described agreeing on an AI-assisted search approach with the European Commission, which then required a detailed written explanation of the method. The panel worked on the assumption that prompts and AI interactions may be discoverable.
Bashir pointed to the spread of homegrown AI tools inside organizations and the informality of chat platforms. Bashir’s team repeats one message in its annual privilege training: if you do not want it discovered, do not write it down. Gürer predicted that authorities who once examined meeting notes and messaging groups may look at AI chat histories next.
On return on investment, Bashir’s team asks outside counsel whether and how they use AI, and tracks its own metrics to show value, in line with the 2025 panel’s call to measure litigation by business outcomes as well as wins and losses.
Colonna closed with a reminder that humans are not going away. A lawyer will always be in the room and accountable for what is put forward. Across both years, the message has held: technology changes how litigators work, and the lawyer’s judgment and duty to the court stay at the center.
Read the full UnitedLex Summit 2026 recap , or contact UnitedLex to discuss defensible, AI-enabled eDiscovery for your next matter.
