Legal is the professional services vertical where generative AI has moved fastest and where the failure modes have been most publicly embarrassing. Lawyers have been sanctioned for filing briefs containing citations to cases that did not exist, produced by models asked to find supporting authority and obliging by inventing some. Against that background, a serious educational programme aimed at legal professionals is badly needed, and Harvey, which has become the dominant legal AI vendor by a wide margin, has built the most substantial one. Harvey Academy splits into a few things.
There is a free certification course covering foundational AI literacy, aimed at building conceptual understanding of what these systems do, where they are reliable and where they are not. There is platform training on using Harvey in legal workflows. There is a Certified Legal Engineer path covering the competencies of a role that sits between legal practice and AI systems work. Completion issues verified digital badges through Credly by Pearson, which means the credential is externally verifiable rather than a PDF from a vendor.
There is also a course version developed for law schools, which matters because it moves this out of customer enablement and into professional education. The free AI literacy course is the part I would recommend broadly, and I say that having low expectations going in. It is better than the AI training most law firms have assembled internally, which typically consists of a partner who read some articles presenting slides about prompt engineering. The Harvey material is built by legal practitioners and the examples are recognisable legal tasks, reviewing a document set, summarising a deposition, drafting a first pass at a clause, rather than the generic content marketing examples that fill most AI literacy training.
Understanding why a model produces a confident wrong answer is taught conceptually rather than as a warning label. The legal engineering framing is interesting and I think broadly correct. Firms are discovering that deploying AI into legal work requires somebody who understands both the practice and the systems, who can decompose a matter into steps a model can do reliably and steps it cannot, who can design a verification workflow, and who can tell a partner why a particular use case will not work. That role exists now at most large firms under various titles.
Naming it and building a curriculum for it is a genuine contribution, and the fact that a vendor did it first says something uncomfortable about how slowly legal education moves. The structural problem is the one that recurs through every vendor academy and is unusually sharp here. Harvey teaches verification and cautions about hallucination, and the coverage is present and not cursory. It is also the case that the company's commercial interest is in lawyers feeling confident enough to use the product extensively, and the appropriate level of professional paranoia about AI-generated legal content is higher than any vendor will comfortably teach.
The public record of fabricated citations reaching courts is not a hypothetical risk, it is a documented pattern with named practitioners and real sanctions, and training that engaged with those cases directly and unflinchingly would serve lawyers better than training that discusses verification in the abstract. The jurisdictional gap matters too. Professional responsibility rules about competence, confidentiality and supervision differ between jurisdictions, and questions about whether submitting client material to a model implicates privilege have been answered differently in different places and are still moving. A lawyer needs their own bar's guidance, not a vendor's general treatment, and the academy does not substitute for that.
Three point eight. The best structured legal AI education currently available, genuinely useful in its free tier, naming a real emerging role, and carrying the unavoidable limitation that the party teaching caution is the party selling the thing caution is needed about.