--- title: "AI can widen legal review" description: "AI can help a legal team examine more material and test more questions. Its value depends on traceable sources, measured performance and a lawyer prepared to stand behind the result." language: "en-CH" source: "/notes/ai-in-legal-work/" canonical: "https://jonashertner.com/notes/ai-in-legal-work/" content_id: "https://jonashertner.com/notes/ai-in-legal-work/#post" type: "article" status: "current" author: "Jonas Hertner" published: "2026-07-27" modified: "2026-09-05" version: "2026-09-05" citation: "Jonas Hertner, “AI can widen legal review” (version 2026-09-05), https://jonashertner.com/notes/ai-in-legal-work/." --- # AI can widen legal review 27 July 2026 · Revised 5 September 2026 > AI can help a legal team examine more material and test more questions. Its value depends on traceable sources, measured performance and a lawyer prepared to stand behind the result. A legal team may have more documents than it can read closely before the next decision. AI can help identify dates, compare accounts and find passages that deserve attention. The useful question is how much more the team can examine, and how reliably it can check what the system returns. ## Sources and findings Language models can summarise documents, translate them, extract stated dates, amounts and parties, and compare texts for possible inconsistencies. These tasks produce different kinds of output and need different checks. A date quoted from a letter is different from a deadline calculated from that date. A discrepancy between two documents is different from a conclusion that one is wrong. A reliable process preserves those distinctions: what the source says, what was calculated or inferred, and what a lawyer concludes. If a model says a notice was sent on 14 March, the reviewer should be able to open the evidence of sending. The date printed on the notice does not by itself establish when it was sent. That small distinction can matter more than the fluency of the summary. Rankings, clusters and classifications also need labels that show how and when they were produced. A document classified as irrelevant has been filtered by a system; its irrelevance has not thereby been established. ## Broader review AI can make it practical to ask the same defined questions of a much larger set of documents. It can link possible findings to passages and identify files that could not be processed. The resulting coverage report should show what was included, what was excluded, when the material was collected and where processing failed. Processing every file does not establish that every relevant point was found. A scanned page may be misread, a question framed too narrowly or a contradiction overlooked. Disagreements between sources should remain visible until a responsible person resolves them. The review must account for what may have been missed as well as what was found. ## When the model supplies the law A model asked for the law without specified authorities may combine sound propositions with obsolete, irrelevant or invented support. Its tone gives the reader no reliable way to tell them apart. The most dangerous error may look entirely ordinary: a plausible citation, a familiar legal test or a proposition that is right in another jurisdiction. It can survive a hurried reading precisely because it fits the answer the reader expected.[^1] An unsourced answer may suggest a line of inquiry, but the legal research remains to be done. Providing authorities makes verification possible. The reviewer must still establish that they exist, remain relevant and support the proposition in its full context. ## Working rules - Define the task, the material to be used and the required output. - Distinguish a source statement, a reproducible calculation and an assessment. - Link every material factual or legal proposition to its supporting sources. - Identify gaps in coverage, unreadable material, conflicts and uncertainty. - Test the process on representative material whose relevant answers have been checked independently. - Name the person who reviews the work and takes responsibility for its use. Confidentiality, professional secrecy, privilege, data protection and information security constrain the choice of system.[^2] Before entering sensitive material, establish where it will be processed, who can access it and how it will be retained, reused, transferred, logged and secured. Local processing may reduce some risks; access controls, backups and connected services still need assessment. Removing names is insufficient if the remaining details identify the people concerned. ## Test the system that will be used Evaluate the whole process, including retrieval, extraction and human review. A model may perform well on text that another part of the system fails to retrieve or read. Choose representative material, expected results and difficult cases before running the test. Measure at least: - relevant material missed; - statements unsupported by the specified sources; - citations that do not exist or do not support the proposition; - answers that stray beyond the question or permitted sources; - the time a reviewer needs to reach a reliable result. Models, settings and connected sources change. Repeat the tests after a material change, and investigate failures found in use. Success on one task is evidence about that task; it must not be treated as proof of reliability elsewhere. ## Responsibility Wider review is valuable when it improves the basis for advice. That requires a reviewer who can inspect the evidence, challenge the answer and explain what remains uncertain. The final question is whether the lawyer has sufficient grounds to act on the result and advise the client. ## Notes [^1]: [NIST AI 600-1, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile](https://doi.org/10.6028/NIST.AI.600-1), section 2.2 and actions MP-2.1-002, MP-2.3-001 and MS-2.5-003. NIST identifies confidently presented false content, including invented logic and citations, and recommends testing data transformations, comparing output with known ground truth and verifying sources and citations. [^2]: [Art. 13 of the Federal Act on the Free Movement of Lawyers](https://www.fedlex.admin.ch/eli/cc/2002/153/en#art_13) governs professional secrecy. The [Federal Data Protection and Information Commissioner](https://www.edoeb.admin.ch/en/ai-and-data-protection) states that the Federal Act on Data Protection applies directly to AI-supported data processing and requires transparency about its purpose, operation and data sources.