The legal AI tool landscape

The legal AI landscape has two very different tiers, and confusing them is a common, costly mistake. Learn the categories and their risk profiles — specific products change constantly (a 2026 snapshot; verify features and pricing against vendor docs, and naming a tool is not an endorsement).

Tier 1 — general-purpose chatbots (highest misuse risk). Consumer tools like ChatGPT and similar general assistants are, per the ABA's surveys, the most-used AI among lawyers — and the source of most sanctions. They are not trained on or grounded in legal databases, so they fabricate citations freely, and their consumer tiers typically offer no confidentiality protection for client data. They can be useful for non-confidential, non-authoritative tasks (brainstorming, plain-language explanation, first drafts you will heavily verify), but they are not legal research tools and should never receive confidential client information without appropriate protections.

Tier 2 — legal-specific platforms. Purpose-built tools grounded in legal databases, with enterprise data protections:

• AI legal research — Thomson Reuters Westlaw Precision / CoCounsel, LexisNexis Lexis+ AI / Protégé, and vLex's Vincent, which run generative answers over actual case law with citation tools (Shepard's, KeyCite). Important: these still hallucinate a meaningful minority of the time (Module 2's Stanford study) — grounding reduces but does not eliminate the problem.
• Enterprise legal assistants — platforms like Harvey, built for firms with document analysis (Vault), drafting, and workflow features, with enterprise security.
• Task-specific tools — contract review and redlining, e-discovery/technical-assisted review, deposition and transcript analysis, and billing/timekeeping.

The critical distinction — grounded and secure, or not:

• Grounded vs. ungrounded: legal-specific tools retrieve from real legal sources; general chatbots generate from patterns. Grounded tools hallucinate less (not never); ungrounded tools hallucinate freely.
• Enterprise vs. consumer data handling: enterprise/legal tools typically offer no-training commitments, tenant isolation, and data-protection agreements; consumer tools often use inputs for training and offer no such protection — a confidentiality problem for client data (Module 3).

Why think capabilities, not vendors: products and features change every quarter, but the categories and their risk profiles are stable. A general chatbot carries fabrication and confidentiality risk whether it's this brand or that; a grounded legal-research tool carries reduced-but-real hallucination risk regardless of name. Evaluate by what a tool does, whether it's grounded in real law, and how it handles your data — not by brand.

The mindset: the legal AI stack splits into general chatbots (most-used, highest misuse risk — ungrounded and often unprotected) and legal-specific platforms (grounded in real law, enterprise-secure, but still hallucinating a meaningful minority). Match the tool to the task: never treat a general chatbot as a research tool or feed it confidential data, use grounded legal-specific tools for authoritative work, and verify output from all of them. Knowing which tier you're in — and its risk profile — is the foundation for using legal AI safely.