Something jumped out at me early on at the recent LegalTech Connect Law Firm Research and Innovation Conference with the legal research community. Greg Lambert asked how many participants have encountered partners who don’t know their Westlaw or Lexis passwords. This question resonated throughout the room as most had experienced it.
It’s common knowledge that partners rarely conduct legal research, but if they don’t know how to connect to a legal research platform, what process do firms use to allow partners to access the underlying sources in a brief?
Over the past year, we’ve seen a steady stream of headlines about AI hallucinations where false cases, fabricated quotes, and erroneous facts have managed to pass through a review process and be filed in court.
Technology is not the problem; It’s a process problem
Even though lawyers are responsible for their work, there is a tendency to attribute this failure to the introduction of AI technology into the process. I think a better explanation is that AI revealed a flaw that already existed in most processes.
For decades, law firm writing relied on associate advice and associate writing, with a senior associate earning a passing grade and then the associate reviewing in a supervisory role. Citation tools such as Shepard’s or KeyCite are used by the associate throughout the process. But in practice, how, what and when does the supervising attorney review the work?
If a case was reviewed earlier in the process, is it checked again at the end of the process, including the citation taken from the case?
Before AI, an error that pierced a gap in the process did not matter as much because potential errors were minor and could be corrected as an errata if necessary. But it doesn’t work anymore. Human authorship is enhanced by AI-generated content and editing that can change facts, fabricate quotes, and invent citations to nonexistent cases.
AI reveals gaps that have always existed
Anyone who regularly uses AI grammar checking software or Microsoft Copilot has had the experience of “uncorrecting” their revisions that are more grammatically correct but also change the meaning of a statement. Associates are still learning and may miss subtle errors introduced by these powerful new tools.
What was correct at an earlier stage of a document may well be changed by AI in a later version.
Errata against manufacturing
The central problem here is that an AI hallucination is a fabrication. False cases are a problem, but more recent issues in the cases include fake quotes. A full review of everything, including the basic facts, should be done at the end of the process before filing.
Tried-and-true workflows that were once assigned to humans need to change.
Writing and editing must be inseparable
Remember your high school science class? There are protons, neutrons and electrons in an atom, but the atom itself cannot be separated. AI highlights this in document creation. The writing and review processes are atomic. They cannot be separated without some sort of nuclear reaction.
AI enables editorial efficiencies, but it places new burdens on the review process and exposes assumptions about who reviews what and when.
The process must now address who checks citations, when the check occurs, and how it is communicated within the team writing and reviewing the document. The same should happen for quotes and fact-checking.
Zero Trust Principles
In cybersecurity, there is a concept called Zero Trust. It’s basically “never trust, always verify.” Unfortunately, AI systems are probabilistic. They are literally guessing the next part of a word that matches a pattern. This means that hallucinations are a feature of the system design rather than a bug. There is techniques to minimize hallucinationsbut they cannot be eliminated.
“Back to basics” review process
The consequences of integrating AI into writing are that errors can become fabrications that go well beyond the scope of errata.
Before the Internet, it was common to print all authorities and keep them in binders for access and reference. But the digital equivalent of this physical filing cabinet has been lost in some companies.
There are solutions available to help organize and improve processes. Creating hyperlinks in documents to a repository of referenced authoritative sources can help speed up review. If there is no link to a case, that is a red flag that it may not exist. Clicking on a link can also help review and verify that quotes or facts are correct.
Businesses should act now
AI policy is not a process. Firms must implement how lawyers can comply with the policy and provide the processes and tools necessary to facilitate compliance with the standards of Rule 11 on Ethical Responsibility.
The California Bar recently comments requested on the evolution of ethical rules in matters of AI. And there are discussions proposing a real Hyperlink rule to ensure that sources are linked in records. Businesses should not wait for regulators to tell them what to do.
Conclusion
AI may have introduced crude fabrication into legal work, but it also removed the illusion that review processes were always airtight.
There are more advanced solutions possible, some of which I have suggested in previous writings on hallucinations. But before the industry considers more advanced solutions, it needs to make sure to cover the basics first.
Companies that recognize and rebuild their review processes and tie them closely to the editorial process will be in a much better position than those that treat this as a problem caused by technology or as a problem to be solved by policy alone.
AI was used in the revision of this article.
Ken Crutchfield has over forty years of experience in the legal, tax and other industries. Throughout his career, he has focused on business growth, innovation and transformation. His consultancy advises investors, legal technology startups and others. As a strategic thinker who understands markets and creates products to meet customer needs, he has worked at start-ups and large companies. He has held general management positions in six companies. Ken has his finger on the pulse of trends affecting the market. Whether it’s the Internet in the 1980s or generative AI, he understands technology and its impact on business. Crutchfield began his career as an intern at LexisNexis and has worked at Thomson Reuters, Bloomberg, Dun & Bradstreet and Wolters Kluwer. Ken holds an MBA and BS in Electrical Engineering from The Ohio State University.
