Tools
Lexis+ with Protégé Review: Workflows, Vault and the Legal Intelligence Engine
Lexis+ with Protégé replaced Lexis+ AI in February 2026 with 300+ workflows, Shepard's-grounded verification and an agentic engine. What it does, costs and still gets wrong.
On this page
- Why LexisNexis retired its own flagship
- The five layers, and when each arrived
- What the Legal Intelligence Engine actually does
- Verification: Shepard's Trust Markers and the 17% question
- Models, data and keys
- What it costs, and what LexisNexis will not tell you
- A ten-question trial protocol
- Where this leaves you
Lexis+ with Protégé is LexisNexis's replacement for Lexis+ AI, generally available in the US since 24 February 2026. It bundles 300+ configurable workflows, a no-code workflow builder, Shepard's-grounded citation checks, a 100,000-document Vault and, since 24 August 2026, a Legal Intelligence Engine that plans multi-step tasks for your approval. It is the most credible research-grounded agent platform from an incumbent, and it still needs a lawyer to read every citation.
Why LexisNexis retired its own flagship
The retirement of Lexis+ AI was not a routine version bump. In May 2024 Stanford's RegLab published "Hallucination-Free?", the only independent head-to-head test of legal research assistants to date: Lexis+ AI hallucinated on roughly 17% of 202 expert-scored queries and was accurate and grounded on 65%, against 33% and 41% for Westlaw's AI-Assisted Research (Stanford HAI). Lexis then withdrew from most tasks in the February 2025 Vals Legal AI Report and declined, with Thomson Reuters and vLex, to enter the October 2025 Vals research benchmark (LawSites).
Then came 3 February 2026. Anthropic's Claude legal plugin knocked 14% off RELX in a single session, the parent company's steepest one-day fall since 1988. Three weeks later LexisNexis launched Lexis+ with Protégé on 24 February and, within the same week, announced its own Anthropic integration (LawSites). The company has since set a public target of automating 15–20% of what a lawyer does by 2028. Read the product with that context: it is a defensive rebuild by a publisher whose data is its moat, executed at speed.
The five layers, and when each arrived
| Layer | What it is | Released | What to check in a trial |
|---|---|---|---|
| Workflows | 300+ configurable, multi-step tasks (research memo, clause comparison, deposition digest) plus a no-code builder | 24 Feb 2026 | Can a partner, not IT, build a five-step workflow in an afternoon |
| Protégé Work and Agentic Drafting | Drafting inside your documents, built on the Henchman technology LexisNexis acquired in 2024 (Lexis Create) | 7 May 2026 | Does it draft from your precedents or from generic Lexis forms |
| Vault | Upload up to 100,000 documents and run cross-document questions and tabular analysis | 7 May 2026 | Does every extracted answer carry a pinpoint citation you can open |
| Workrooms | Shared matter spaces with dual approvals, role-based permissions and audit trails | 7 May 2026 | Can a junior's output be blocked until a partner approves |
| Legal Intelligence Engine | An agentic harness that selects models, sources and skills and presents an executable plan before running | 24 Aug 2026 | Can you edit the plan, and does the log show which model did which step |
The May 2026 release also introduced customer-held encryption keys, which matters for firms whose bar rules or client terms require that the vendor cannot read matter data on its own initiative (LawSites).
What the Legal Intelligence Engine actually does
The 24 August 2026 announcement is the part that turns Protégé from a set of workflows into something closer to an AI agent. LexisNexis's own description is unusually plain:
"The Legal Intelligence Engine is an agentic harness." — Greg Dickason, CTO, LexisNexis, quoted by Artificial Lawyer (24 August 2026)
In practice the engine confirms its understanding of your request, chooses which of the underlying models (Anthropic, Google, OpenAI or Lexis's own) and which sources and skills to use, and then presents an executable plan for user approval before it runs (LexisNexis pressroom). That approval step is the single most important control in the product. Treat it the way you treat a trainee's research plan: if the plan says it will search only state appellate decisions when the issue is federal, stop it there rather than after it has written twelve pages. The same pattern is appearing across the market, from Harvey's agents that "flag moments requiring user confirmation" to the send gates in Anthropic's Claude for Legal plugins; Protégé's version is the one bolted directly to a citator.
Verification: Shepard's Trust Markers and the 17% question
Every Protégé research output is grounded in Shepard's and carries "Verify Trust Markers" that show whether a cited authority exists and how it has been treated. That solves the embarrassing class of error, the invented case, and it is a real improvement on a general chatbot. It does not solve the harder class. Damien Charlotin's database counted 854 misrepresented citations among its 2,041 decisions as of 14 September 2026, alongside 1,690 fabricated ones (Charlotin): real cases cited for propositions they do not support. Grounding retrieves the case; it does not guarantee the model has read it correctly. In the HAQQ benchmark of June 2026, 24% of 3,000 answers across leading models cited or misapplied law that did not support the claim.
The sanctions record makes the same point. Fletcher v. Experian (5th Cir., 18 February 2026) fined counsel $2,500 for 16 fabricated quotations traced to vLex and CoCounsel, and Lacey v. State Farm (C.D. Cal., 6 May 2025) cost Ellis George and K&L Gates $31,100 after a workflow that included CoCounsel and Westlaw Precision. Paid, grounded tools were in the chain in both. So the rule for Protégé is the rule for every research assistant: existence, support and treatment, checked by a human, as set out in the citation-checking protocol. LexisNexis's own general counsel asked in Fortune in May 2026, "Would you hire the lawyer who just got sanctioned for using AI?" (Fortune). The answer does not change because the tool has a citator.
Models, data and keys
Protégé is multi-model. LexisNexis routes tasks to models from Anthropic, Google and OpenAI and to its own models, and has promised custom legal models. That is a strength (you are not hostage to one lab's release schedule) and an opacity problem (you may not know which model drafted which paragraph). Three questions belong in any vendor diligence, alongside the ones in the AI vendor due diligence guide:
- Which sub-processors can see prompt content, and under which retention terms, for each model in the stack?
- Does the audit trail record the model and version used for each step of an engine plan?
- With customer-held keys enabled, what can LexisNexis support staff still access, and how is that logged?
The market context is favourable to Lexis's pitch. A LexisNexis survey of 543 legal professionals found 81% feel more comfortable with AI grounded in legal sources, up from 70% a year earlier (survey release). Vendor-run, but consistent with what firms tell each other. Lexis is also a connector inside Claude for Legal, so a firm can reach Shepard's from Anthropic's plugins rather than the Lexis interface, which is worth knowing if your lawyers already live in Claude.
What it costs, and what LexisNexis will not tell you
LexisNexis does not publish Protégé pricing as of September 2026. Thomson Reuters does: its CoCounsel and Westlaw bundles run from about $104 to $639 per user per month, with a solo Westlaw Advantage plus CoCounsel Essentials package at $639.20. Clio Work lists at $199 per user per month. Use those as anchors in negotiation, and see the legal AI pricing guide for the wider picture.
Adoption tells its own story. ILTA's 2026 Technology Survey (14 September 2026, 500+ firms) put Lexis+ AI at 26% of firms, behind Microsoft 365 Copilot (76%), CoCounsel (44%), Claude (44%) and Harvey (43%) (LawSites). Protégé's job is to close that gap with CoCounsel, and the head-to-head comparison is where most Lexis customers will end up.
A ten-question trial protocol
Demos show you the workflows that work. Run these on your own trial licence, with a lawyer who knows the answers, before you sign.
| # | Test | What a pass looks like |
|---|---|---|
| 1 | A research question in your jurisdiction where you know the leading case | It surfaces that case first, with a current Shepard's marker |
| 2 | A question with no authority on point | It says so, rather than assembling near-misses into an answer |
| 3 | A misgrounding trap: ask it to support a proposition slightly beyond what a real case holds | It qualifies or refuses; it does not stretch the holding |
| 4 | A multi-jurisdictional question | Watch for the 11–14 point accuracy drop Vals found across tools in 2025 |
| 5 | Build a five-step workflow in the no-code builder | A partner completes it without IT in under an hour |
| 6 | Load 200 documents into Vault and ask a cross-document question | Every cell or answer links to a pinpoint you can open |
| 7 | Route a junior's draft through a Workroom with dual approval | The draft cannot leave the Workroom until a partner approves |
| 8 | Ask the engine for a research memo and read the plan | You can edit the plan before it runs, and the log names each model used |
| 9 | Export the audit trail for a matter | You receive a file your risk team can keep with the matter |
| 10 | Request BYOK setup, retention periods and EU residency in writing | Answers arrive as contract terms, not slideware |
Compare the results with the same ten questions in CoCounsel Legal and, if your firm has it, a general model with a research connector. The benchmarks guide explains why vendor scores and your own results will differ.
Where this leaves you
If you already pay for Lexis+, Protégé is the version you should be on, and the Workrooms approval model is a sensible fit for firms that want juniors using agents under supervision. If you are choosing between Lexis and Thomson Reuters for the first time, the decision will turn on which content set you trust for your jurisdictions and which vendor gives you audit terms in writing. Either way, keep the human step: as the AI legal research overview sets out, no grounded tool has passed a public zero-hallucination test, and the courts require zero. Module 2 of the AI Lab for Lawyers course runs exactly this kind of head-to-head between research tools on live questions, so you can see the misgrounding trap fail in real time before it fails in a brief.
Frequently asked questions
Is Lexis+ with Protégé the same as Lexis+ AI?
No. LexisNexis retired Lexis+ AI and launched Lexis+ with Protégé in the US on 24 February 2026. The new product is an end-to-end workflow platform with 300 or more configurable workflows, a no-code builder, Shepard's-grounded verification and models from Anthropic, Google and OpenAI. Later 2026 releases added Vault, Workrooms and the Legal Intelligence Engine.
How accurate is Lexis Protégé for legal research?
LexisNexis has not submitted Protégé to an independent public benchmark as of September 2026. The last independent test, Stanford RegLab in 2024, found the older Lexis+ AI hallucinated about 17% of the time and was fully accurate on 65% of queries. Grounding in Shepard's reduces fabricated cases, but every citation still needs a lawyer to read the authority.
What does the Legal Intelligence Engine do?
Announced on 24 August 2026, it is what LexisNexis's CTO calls an agentic harness. It confirms what you are asking, chooses which models, sources and skills to use, and presents an executable plan for your approval before it runs multi-step work such as research, drafting and cross-document analysis inside Vault or a Workroom.
How much does Lexis+ with Protégé cost?
LexisNexis does not publish Protégé pricing as of September 2026; quotes depend on your existing Lexis+ subscription, seat count and modules. For comparison, Thomson Reuters publishes CoCounsel and Westlaw bundles from about $104 to $639 per user per month, and Clio Work lists at $199 per user per month. Ask for the audit-trail and BYOK terms in writing.
Put this into practice
AI Lab for Lawyers: a live, hands-on course for practising lawyers
4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.
Related reading
Comparisons
CoCounsel vs Lexis+ Protégé: The Research Incumbents' AI Compared
CoCounsel vs Lexis+ Protégé in 2026: KeyCite against Shepard's, a Claude Agent SDK rebuild against a multi-model harness, prices, jurisdictions, when to switch.
10 min read
Tools
CoCounsel Legal (Thomson Reuters) Review: Westlaw-Grounded Agents
CoCounsel Legal reviewed as of September 2026. The Claude-built rebuild, Brief Builder, Tabular Analysis, Deep Research Verify, $104 to $639 pricing and grounding.
8 min read
Guides
AI Hallucinations in Legal Practice: Why They Happen and How to Catch Them
AI hallucinations in legal practice: why models invent cases, what 2,041 court decisions and the HAQQ benchmark show, and a verification routine that holds.
9 min read
Playbooks
How to Verify AI Legal Research: A Citation-Checking Protocol
Verify AI legal research with a five-check protocol: open every authority, match the quote, confirm holding and status, log it. Template, time budgets, tools.
11 min read
Use Cases
AI Legal Research in 2026: What Works, What Hallucinates
AI legal research now beats a lawyer baseline on accuracy yet misapplies law in 24% of answers. Which tools are grounded, where errors hide, a routine that holds.
11 min read
Guides
Legal AI Benchmarks Explained: Vals, BigLaw Bench, LAB, HAQQ and LegalBench
Legal AI benchmarks explained: what Vals, BigLaw Bench, Harvey LAB, HAQQ and LegalBench measure, the September 2026 leaders, and how to run your own evaluation.
10 min read