What AI Means for Legal Work
Cut through the hype and understand what large language models can and cannot do in a legal context — and why that distinction matters for your practice.
In practice: Contract first-pass: 2–4 hours → 20 minutes
Step 1 of 4
Read
AI in the Legal Profession: The Honest Picture
AI tools like Claude, ChatGPT, and specialised legal AI platforms are large language models (LLMs). They generate text by predicting likely next words based on patterns in billions of documents. That simple fact explains both their remarkable utility for legal work and their very real limitations.
What LLMs Are Genuinely Useful For
- First-draft generation. Drafting a non-disclosure agreement, a memo, or a client communication starts with AI producing a solid foundation. You refine, not create from scratch.
- Summarisation. A 200-page contract can be summarised into key commercial terms in minutes. A deposition transcript becomes a structured fact summary.
- Pattern recognition in text. AI can scan dozens of contracts for a specific clause type, flag non-standard language, or identify missing provisions across a document set.
- Research starting points. AI can outline the legal landscape on an issue, surface relevant doctrines, and suggest search terms for primary source research.
What LLMs Cannot Do
- They cannot cite reliably. AI tools hallucinate case citations — generating plausible-looking references to cases that don't exist or that say something entirely different. Never cite a case without verifying it independently.
- They do not know your jurisdiction's current law. Training data has a cutoff date. Recent statutory changes, new regulations, and recent case law may not be reflected.
- They are not privilege-aware. The AI doesn't know which information is privileged or confidential unless you explicitly design your prompts and workflows to protect it.
- They cannot exercise legal judgment. Risk assessment, strategy, and advice require human legal reasoning grounded in context the AI cannot fully understand.
The Mental Model That Helps
Think of AI as a very well-read paralegal who works at extraordinary speed but must never be sent to court alone. They can do the research and drafting; you review, verify, and take professional responsibility for everything that leaves your department.
Practical Starting Point
Begin with tasks where errors are catchable before they cause harm: summarising documents you've already read, drafting internal communications, or generating initial clause checklists for your own review. Build confidence before using AI on higher-stakes external documents.
The prompt
Claude's response
Quick check
A colleague says AI hallucinations are not a concern in legal work because the AI sounds very confident and uses correct formatting. What is wrong with this view?
Key Takeaways
- 1LLMs predict text — they do not retrieve law or exercise legal judgment
- 2AI hallucinations are a documented risk: always verify case citations and statutory references
- 3Use AI for drafting, summarising, and pattern recognition — not for final legal conclusions
- 4Start with lower-stakes internal tasks before applying AI to client-facing work
- 5Confidentiality must be actively managed — it is not automatic
Your challenge this week
Apply what you learned in a real task
Pick one document you are currently working on — a contract clause, a board minute, or an internal memo. Use AI to generate a first draft, then compare the time taken and what you needed to correct. Note both where AI saved time and where your professional judgment was essential.
Before you practise
Think about the last week of legal work. Which tasks required your deepest professional judgment — and which were largely mechanical? If AI could handle 60% of the mechanical work, what would you do with that recovered time, and how would it change the value you deliver to clients?
Next step
Put it into practice
You've read the lesson — now apply it in a guided hands-on exercise. It takes about 5 minutes.