AI tools for lawyers — what's safe, what's risky
A practitioner-grade survey of AI tools for legal work, with explicit notes on what's safe to deploy and what's malpractice waiting to happen.
The legal AI category has a credibility problem. Half the tools confidently hallucinate case citations; the other half cost $1,000/seat/month with unclear ROI. Here's a practitioner-grade read in 2026, with explicit risk notes.
Tier 1 — Safe for daily work
SeekFiles AI
Use for: querying your own case files, contracts, codals, reviewers.
- Chunk-level citations back to the exact page.
- Refuses to answer when your materials are silent.
- Folder scoping per question (constrain to a single case or matter).
- Mobile for court-day research.
Risk: any tool that retrieves from your documents is structurally safer than one that "knows" jurisprudence from training. Always verify cited chunks against the source.
Westlaw + AI-assisted search (or Lexis equivalent)
Use for: case law research with built-in citation validation.
- Citations come from a verified legal database.
- AI summary layer is improving fast.
- Expensive but auditable.
Risk: the AI summary can still mischaracterise — use it as a starting point, then read the cases.
Tier 2 — Useful with strict guardrails
Casetext CoCounsel (or competitor)
Use for: memo drafting, document review at scale.
- Built for legal workflows.
- Better than generic LLMs for legal phrasing.
Risk: review every citation. Multiple sanctions in 2023–2025 traced back to lawyers who trusted CoCounsel-style tools without checking.
Harvey (BigLaw)
Use for: if you're at a firm with the contract, the integrations are real.
- Tight workflow integration in M&A and litigation.
Risk: seat cost is high; outside BigLaw the ROI math is questionable.
Tier 3 — Risky without verification
ChatGPT for legal research
Use for: only drafting and brainstorming. Never citations.
- Will confidently invent case names that don't exist.
- The 2023 Mata v. Avianca sanctions case was a wake-up call; "but I checked!" is not a defense.
Risk: malpractice. Every cited case must be independently verified in Westlaw/Lexis/PACER.
Generic "AI contract review" tools
Use for: triage only, never as final review.
- Many will flag standard clauses as "risky" because they over-index on training data.
- Few have the firm-specific context to know what your clients accept.
Risk: false positives that scare clients; false negatives that miss real risk.
Workflow we recommend
- Client-document research: SeekFiles AI (your matter files).
- Case law research: Westlaw or Lexis with their built-in AI summary; verify cases yourself.
- Drafting: ChatGPT or Claude for first draft; your own editing pass; partner review.
- Brief checking: human + Casetext-style cite-checker; never AI alone.
What never to do
- Submit a brief with AI-generated case citations you haven't verified.
- Use a public AI tool for confidential client documents without a Business / Enterprise tier that contractually excludes training.
- Trust an AI summary of a long case without reading the case yourself if it's going in a filing.
- Promise clients "AI-powered" anything without being able to explain what the AI actually does.
A note on professional responsibility
Wherever you practise, treat AI the same: useful for your own files (statutes, reviewers, case digests), risky for jurisprudence research, malpractice if you cite uncited. Your bar's competence obligations apply to AI-assisted work just as much as to traditional research.
AI in legal is a useful junior associate at best. Always supervise it.
Like this? Get the next one in your inbox.
Weekly tips on getting more out of your file library — RAG, retrieval tricks, and product updates. No spam.
Try it free
Ask your files anything. Get answers with citations.
50 welcome credits. 3 assistants. No credit card. Upload your first file in under two minutes.