When AI searches thousands of legal sources, the signing lawyer's duty to verify grows

Lucio Labs describes a research mode that, by its own account, searches hundreds to over two thousand legal sources and also shows rejected and conflicting rulings, which increases coverage but does not produce a guaranteed correct answer. The signing lawyer must check every cited passage, jurisdiction and date, and record that.

What exactly has Lucio Labs announced and why is that news?

On 15 October 2026 Lucio Labs published a description of an in-depth mode for legal research. According to the company, that mode typically assesses roughly 800 to 1,500 sources, sometimes more than 2,000, and the system shows for each source whether it has been classified as highly relevant, relevant, flagged or rejected. These are the vendor's own figures and not independently verified performance.

The significance lies not in a smarter final answer, but in a shift in the economic boundary of research. Trade publication Artificial Lawyer placed the development in the context that research in practice often stops when time or budget runs out, not when every relevant authority has been found. In our assessment that is the core: the research can run further, including unfavourable and older rulings, and shows what it has left out.

Does searching more sources mean the legal answer is also correct?

No. Breadth is not correctness. A larger number of searched or cited sources says nothing about whether a proposition is actually supported by the cited passage, whether the ruling is still good law and whether it falls within the correct jurisdiction.

In the Journal of Empirical Legal Studies, researchers found substantial reliability problems among the specific commercial legal-AI tools they tested: performance varied by tool, with some producing inaccurate, hallucinated or incomplete answers in a substantial share of the questions examined. The result applies to those tools and the study's test design, not automatically to every legal-AI system. That result applies to the tools and test design examined, not automatically to every legal AI system. Earlier, researchers in the Journal of Legal Analysis described how general language models can fabricate or misrepresent legal authorities and are poorly calibrated about their own reliability. A randomised experiment reported by the University of Minnesota Law School found that AI assistance improved performance and speed in some structured legal tasks, while participants who began with stronger analyses sometimes produced lower-quality work after incorporating AI-generated revisions.

The lesson from those three studies is consistent: AI can broaden the candidate list of authorities, but does not shift the judgement. Those who leave the reasoning to the system rather than using it as a tool run the risk that strong analyses are diluted.

What does wider AI research mean for lawyers, directors and firms?

Our analysis: because the system now supplies hundreds to thousands of sources instead of a few dozen, the volume a human still has to read grows, and with it the liability for the final judgement shifts to the systematic checking of that larger volume; this is why a firm must record in advance which of the lawyers involved reads and ticks off which categories of sources, and not leave that to whoever happens to have time. Because the vendor itself states that sources can be filtered out as rejected or flagged, a relevant counter-ruling may stay out of view without anyone noticing; in our analysis the responsible lawyer must therefore open the rejected sources on a sample basis and document why exclusion was justified, before the memo goes out. Because independent research shows that these tools can attach incorrect propositions to real rulings, a citation list does not suffice as evidence; our reading is that the signing lawyer must therefore read back every load-bearing proposition in the full ruling and not in the summary or the heading, and sign off that check per proposition. And because the final advice rests on more authorities, that increases the reach of any error; this is why directors and partners must revise their internal liability and review protocol, with a recorded decision point at which a named senior determines which authorities govern the client's facts. This is editorial interpretation, derived from the sources named, not a statement by Lucio Labs or by the researchers themselves.

How does a firm set up verification and accountability with a larger source volume?

Use the wider search capacity as a first sift and as a counterweight, not as a finished product. The following steps follow directly from the findings named above and are our recommendation, not a prescription from the sources. See also our earlier pieces on the the signing lawyer's duty to verify AI citations and on AI summaries that drop load-bearing arguments.

• Record the jurisdiction, the reference date and the legal question before searching, so that breadth does not come at the expense of relevance.
• Force the system to show conflicting and negative authorities explicitly and open the category of rejected sources on a sample basis.
• Read the full ruling instead of relying on summaries or headings and check that every proposition is supported by the cited passage.
• Record the search query, the source collection, the exclusions, the model or product version and the review date as part of the file.
• Designate one responsible lawyer who decides which authorities govern the client's facts and who signs off the final judgement.

For organisations considering such systems, this belongs in a broader framework of AI in professional legal practice, including a sober assessment of what to verify before deploying a legal AI system. Under the EU AI Act, human oversight in professional use remains a requirement; searching more widely changes nothing about that. The difference between more available research and better-founded advice lies in who reads, checks and signs.

Sources: The article draws on the publication by Lucio Labs, the reporting by Artificial Lawyer and research from the Journal of Empirical Legal Studies, the Journal of Legal Analysis and the University of Minnesota Law School.