Legal AI has had a legitimacy problem. High-profile incidents of AI systems hallucinating case citations — fabricating court decisions that didn't exist and submitting them in legal briefs — created warranted skepticism about AI's role in legal work. Eighteen months later, the picture has substantially clarified. Some legal AI applications are reliably valuable; others remain too risky for unsupervised use.
What works today
Contract review and analysis is the most mature and reliably effective legal AI application. Tools like Harvey, CoCounsel, and LexisNexis Protégé can review standard commercial agreements — NDAs, SaaS agreements, employment contracts, vendor agreements — for missing clauses, problematic provisions, and inconsistencies against a defined playbook faster and more consistently than a junior associate. For in-house legal teams managing high volumes of routine contracts, this is a genuine productivity multiplier.
Legal research assistance, when constrained to retrieving and summarizing existing case law from verified databases (not generating citations), is also valuable. Tools integrated with Westlaw and LexisNexis databases retrieve and synthesize relevant precedents accurately — the hallucination problem arises when AI generates citations rather than retrieving them from verified sources.
Document drafting acceleration — generating first drafts of standard documents from templates and parameters — is effective when rigorous attorney review is built into the workflow. The AI produces a workable starting point; the attorney provides the judgment, strategy, and final review.
What remains genuinely risky
Adversarial legal work — predicting litigation outcomes, strategic positioning in disputes, assessing jury perception — remains firmly in the domain of experienced human judgment. AI systems lack the contextual understanding of local legal culture, judge temperament, and litigation dynamics that experienced litigators develop over careers.
Novel legal questions, cross-jurisdictional issues, and regulatory interpretation in rapidly evolving areas (AI regulation, crypto, emerging technology) are high-risk AI applications. These areas require synthesizing incomplete, sometimes contradictory guidance with judgment about regulatory intent — a task where AI hallucination risk is highest.
The integration approach that works
The most effective legal AI deployments we've seen treat AI as a junior associate: it does the first draft, finds the relevant cases, flags the issues, and produces a workable starting point. An experienced attorney reviews, evaluates, and takes responsibility. This workflow captures the productivity benefit of AI while maintaining the judgment and accountability that legal work requires.
Found this useful? Share it.