While nearly 77% of legal teams have jumped on the AI bandwagon by 2025, they might want to pump the brakes. The reality behind those shiny automated legal tools? They're making stuff up. A lot.
Lexis+ AI and Ask Practical Law AI are hallucinating incorrect information over 17% of the time. Westlaw's AI? Even worse at 34%. That's more than one-third of responses containing fabricated nonsense. For an industry where accuracy genuinely determines whether someone wins or loses a case, these numbers are terrifying.
The hallucinations aren't picky either. They affect doctrine questions, jurisdiction-specific queries, false premises, and basic factual recall. Fundamentally, AI tools are confidently wrong about everything from simple facts to complex legal principles. These discriminatory outcomes mirror the broader problem of AI systems inheriting human biases that plague other industries.
AI tools are confidently wrong about everything from simple facts to complex legal principles.
Sure, some tools like Harvey Assistant and CoCounsel are outperforming human lawyers in tasks like document Q&A and chronology generation. Harvey leads in five out of six evaluated areas, while CoCounsel consistently scores between 73.2% and 89.6% in document tasks.
But here's the kicker: humans still crush AI in complex work. EDGAR research and contract redlining remain firmly in human territory because they demand actual contextual understanding, not pattern matching. The Vals Legal AI Report establishes benchmarking frameworks that consistently show lawyers maintaining superiority in these sophisticated analytical tasks.
The industry knows there's a problem. Independent benchmarking frameworks are popping up everywhere, with calls for transparency growing louder. Vals AI's auto-evaluation system represents one attempt at impartial assessment, though methodologies remain controversial due to high error margins. Over 25 federal judges now require disclosure of AI usage in legal contexts, signaling the judiciary's growing concern about accountability.
Despite the risks, adoption continues surging. Legal teams use AI for document review (77%), legal research and summarization (74%), and drafting briefs and memos (59%). Rising workloads and AI mainstreaming drive this growth, even as selecting appropriate tools remains challenging.
The productivity promises are real. AI can potentially free up 240 hours per legal professional annually by automating routine tasks. Document review, contract analysis, and legal research see significant improvements.
But those time savings mean nothing if the output is wrong. With no fully hallucination-free performance standard in sight, human legal expertise isn't just helpful—it's crucial. The technology isn't ready to fly solo, regardless of what the marketing materials claim.