Harvey vs Litera: how they compare in 2026

Agents, Vault review tables, Knowledge research, Contract Intelligence review, Intake triage and Horizon Scanning all produce their output with language models. No document management or workflow system sits underneath that a firm would buy on its own. Vault stores matter documents, and its purpose is extraction, comparison and reporting across them.

The AI drives a core capability on a product that stands without it, and the vendor says so. The Lito site describes an agentic layer that routes high stakes work to Litera's deterministic comparison, document analysis and metadata removal engines, which return the same result every time, and flexible work to skills built on frontier models. It states that generative AI can be switched off entirely while the rules based engines keep working. Draft Base, the entry package sold from the store, is document comparison, and Lito was added to existing subscriptions at no charge in October 2025 rather than sold as the product. The published agreements treat the generative features as additional terms layered on the SaaS or on premises software. Remove the models and a thirty year drafting, comparison and metadata business remains.

In the February 2025 Vals Legal AI Report, a third party benchmark set against a lawyer baseline, Harvey Assistant was tested on six tasks. Its results ranged from 65.0 to 94.8 percent, and it beat the lawyer baseline on five. Document question answering reached 94.8 percent, 24.7 points above the baseline, with responses under a minute. Answers on scanned and badly formatted court transcripts reached 77.8 percent, 24.1 points above it. That study measured task accuracy rather than citation validity, and Harvey did not enter the later Vals study of citation authority. BigLaw Bench, Harvey's own benchmark, publishes its task categories and judging criteria on a public repository, with hallucination and source results by model. It defines a hallucination as a factual claim that a source of truth can disprove and tracks reasoning errors apart. Harvey reports that its systems hallucinate less often than the foundation models on legal tasks while giving longer answers. Every agent claim links to the source passage behind it, and the full BigLaw Bench dataset is available on request. Through Ask LexisNexis, cases are validated with Shepard's Citations and the Shepard's Signal is shown beside them. Harvey's own United States case law source, more than 9 million CourtListener opinions, runs on Harvey Graph, which maps who cites whom and how each case was treated, so an agent checks treatment before it answers. Harvey says precedent flags and cite checking for that source will follow.

Grounding is claimed, and the vendor itself says accuracy is not measured. The Lito site says every recommendation shows its reasoning and its source, and that the agent is grounded in the firm's own templates, clauses, playbooks, precedents and matter data. Web search over public material is supplied by You.com, per the subprocessor register. No retrieval method is described, and no accuracy figure or test set is published. Section 3(c) of the Generative AI Terms, Version April 2026, says outputs may be inaccurate or misleading and are not tested, verified, endorsed or guaranteed by Litera to be complete or current. It makes the customer solely responsible for reviewing and verifying any output before use. Lito drafts and reviews against firm documents and does not cite primary legal authority, so there is no citator and no authority link to check.

Agents show their plan before they run, so a lawyer can adjust the scope and approve the work before Harvey begins. While an agent works, Harvey brings the lawyer back in when a judgment call is needed, logs every step and cites a source for every claim. Legal Work Orchestration, launched in October 2026, sets defined review gates inside a matter. Each task carries its scope, inputs and reviewer, and when an agent finishes a review Harvey notifies the designated partner. The task completes only once that partner approves it, and the next step then begins. Intake applies routing rules the legal team defines to send each request to an approved agent, to Contract Intelligence or to a lawyer. In the Connector Library, admins choose which tools are enabled, who may use them and which require human confirmation before they run. Reports and reviews can run on a schedule, and admins govern access through roles and workspace permissions. What happens to an agent's work when a reviewer rejects it, and what Harvey commits to when that work is wrong, are not stated.

A written commitment that the lawyer keeps judgment, with real controls, short of the full structure. The Lito site states that the lawyer keeps judgment either way, that firms can switch generative AI off entirely while the rules based engines keep working, and that MCP connectors are governed centrally and enabled firm wide. The Generative AI Terms, Version April 2026, place sole responsibility for reviewing and verifying output on the customer (3(c)). They prohibit deploying autonomous agents, recursive prompt chains or usage automation outside intended user flows (4(c)), which is a published boundary on what the system may run alone. Not published is the threshold at which Lito acts without a lawyer in the loop when it chains deterministic engines at workflow milestones, what review surface a chained action presents, or what happens after an output is wrong beyond the customer's duty to verify.

Customer stories name the firm and give figures. Reed Smith, with about 1,800 lawyers in 34 offices, reports 700 Harvey users a day, 1,400 a month and AI certification for every associate. Repsol's legal division started a pilot in January 2024 with 50 lawyers in 12 jurisdictions and reports cutting the time on one process by more than 90%. CMS reports 95% adoption, and The Adecco Group's general counsel says more than 80% of its legal professionals report real gains in quality and speed. In October 2026 CVS Health expanded Harvey across its corporate affairs functions after a pilot, with the in house legal team using it for document review, research, drafting and knowledge retrieval. Cozen O'Connor and Spencer West chose Harvey in November 2026. Other named customers include A&O Shearman, Gleiss Lutz, Cuatrecasas, Macfarlanes, KKR, PwC, Deutsche Telekom, HubSpot, Bridgewater and Carvana. The figures are usage counts and time savings as each customer reports them, with no stated method for how they were measured.

Named customers without figures, and figures without named customers. The Lito site carries attributed testimonials from Brian Corbett, partner at Poyner Spruill LLP, and Joshua Tan, partner at Delta Law Corporation. It also quotes CCA Law Firm and unnamed Am Law firm associates; none carries a measured result. The figures on the same page are headline counters with no basis stated: 99 percent of the Am Law 100, more than 15,000 firms, more than a million daily users and a 90 percent efficiency gain. The home page adds 74 percent of the Fortune 100 for corporate legal. The October 2025 press release states that Litera One was used by roughly 40 percent of the customer base and that early access firms shaped Lito, with no named firm and no figure. No dated, named deployment with a measured change and a method is published.

Neither Harvey nor its subprocessors may train any AI model on customer content or data, and model providers work under zero data retention. A firm can ask for a bespoke model trained on its own data and used only for that firm. Harvey syncs and enforces the firm's existing ethical wall policies and never creates, modifies or deletes walls, so the walls provider stays the system of record. Workspaces are logically separated, access follows roles, and data is deleted within 30 days of termination. If the law or a court compels disclosure, Harvey gives advance written notice where it is legally permitted. Attorney client privilege and work product are not addressed as such.

Substantive published commitments on training and the model providers, without segregation or privilege. Section 2(c) of the Generative AI Terms, Version April 2026, says Litera will not, and will not permit its third party providers to, develop, train or fine tune any generative or foundational model with Customer Data, End User Prompts or Outputs unless agreed separately in writing. Section 5.1 of the Master Terms, Version November 2025, makes Customer Data the customer's Confidential Information, and section 6.2 leaves ownership with the customer with a license limited to providing the Services. Section 3.2 of the SaaS Terms commits to prompt deletion after termination. The Litera One / Lito subprocessor register states each model provider's position. Anthropic retains inputs and outputs for up to thirty days for trust and safety, and Amazon Bedrock does not pass inputs to third party model providers. Nothing published addresses segregation between users, matters or clients inside a firm's tenant, or privilege or work product handling directly. The trust center's Firm AI Search FAQ and security FAQ sit behind a Get access request.

Output is not legal advice and may contain errors, misstatements or gaps, and the Platform Agreement calls the service a research tool. Harvey sells to lawyers, legal departments and professional services firms, and its agent pages leave the judgment on every task with the lawyer. Access credentials belong to one named user and may not be shared. Nothing addresses competence and supervision duties under professional rules, or the jurisdictions where the product may be used.

A real published position on advice versus tooling, without jurisdiction limits or the supervision side. Section 3(d) of the Generative AI Terms, Version April 2026, says the generative features are not intended as professional advice, cannot replace advice from a qualified professional and form no such relationship, naming the attorney client relationship as the example. Section 3(c) places sole responsibility for reviewing and verifying output on the customer. The intended audience is plain: every segment page addresses lawyers, law firms of every size and corporate legal departments, and the Master Terms restrict use to the customer's internal business purposes, with no consumer surface. No jurisdiction limit is named, and nothing treats how the product supports a lawyer's competence and supervision duties beyond the marketing line that the lawyer keeps judgment. Litera also advertises a CLE program.

ISO/IEC 42001:2023 certification covers Harvey's AI management system, and the Statement of Applicability published with it sets out the controls in scope. Its AIUC-1 certification, an AI security and safety standard, was conducted by Schellman and, in Harvey's account, validates its adversarial testing and AI security program. The trust center adds IRAP attestation, an EU AI Act item and AI Governance, AI Monitoring and AI Overview documents, with an AI Acceptable Use Policy among its policies. An AI Policy dated 26 December 2025 commits to testing the accuracy, reliability and suitability of the AI in the service and to change management for continuing legal compliance. No accountable owner is named, and nothing is published about uneven output across matter types or populations.

A position on training data and acceptable use is published; a governance framework, an owner and a testing regime are not. The Generative AI Terms, Version April 2026, commit that no generative or foundational model is trained on customer content and prohibit a listed set of abusive uses (2(b)). The Lito site names a security and governance layer made up of ISO 27001, SOC 2 Type 2, SOC 3, GDPR, NIS 2 and DORA, which are security and regulatory attestations rather than AI governance. The SafeBase trust center lists an item titled AI Training Data and Bias under an AI section, alongside a Firm AI Search FAQ, both behind a Get access request whose tier is not stated. The title alone says nothing about the contents. Nothing published names who inside Litera is accountable for AI output, what is tested before a skill ships, or any finding about uneven output.

Customers choose what to upload, set their own retention policies and can delete data at any time. Within 30 days of termination, Harvey deletes remaining customer data and content unless told otherwise, and the DPA extends return or deletion to every subprocessor. Access runs on role based controls, logical workspace separation, SAML SSO, audit logs and IP allow listing. The Subprocessor List, updated 21 September 2026, names each provider with its purpose and location, and subscribers receive at least 30 days' notice of an addition with a right to object. Under the Security Addendum, Harvey notifies the customer of a security incident without undue delay and within 48 hours. It then reports the incident's nature, the investigation and the mitigation, and keeps relevant logs for at least a year.

Retention, deletion, access control, subprocessors and incident practice are all published in binding documents a buyer can read before signing. The Security Addendum 2023, incorporated by section 5.3 of the Master Terms, commits to notify the customer of a security incident without undue delay and within seventy two hours of determining that Customer Data is affected. It describes least privilege access with multi factor authentication and quarterly access reviews, AES-256 encryption at rest and TLS 1.2 or above in transit, and annual third party penetration testing. It also provides for deletion of remaining Customer Data after termination. Section 3.2 of the SaaS Terms, Version March 2025, commits to prompt deletion after termination, with backup copies removed on their normal cycle. The subprocessor register is published for each product with hosting locations, and the Litera One / Lito entry names Microsoft Azure, Amazon Bedrock, Anthropic and You.com with what each processes and retains, plus a registration link for change notices. No period is stated for prompts and outputs held on the platform during the subscription, only the provider's thirty day window and deletion after termination.

The Platform Agreement, last updated 9 January 2026, sets out what Harvey stands behind. Section 8 indemnifies the customer against third party claims that the service infringes intellectual property. If a claim arises, Harvey may substitute the service, secure a license to continue, or terminate and refund prepaid fees. The indemnity excludes materials Harvey did not provide and output from input the customer knew was likely to infringe. Harvey warrants that the service conforms to its documentation, is delivered in a professional manner and complies with applicable law. It also commits to carry industry standard insurance with insurers rated A- or better by A.M. Best. Total liability is capped at the greater of 12 months of fees or $250,000. Payment, indemnity and claims the law will not let a contract limit, such as gross negligence, sit outside that cap. Data breach and confidentiality claims carry a separate cap of twice 12 months of fees or $500,000. After a security incident, the Security Addendum commits Harvey to an independent forensic investigation at its own cost. The agreement states that output is AI generated and may contain errors.

What the vendor stands behind is published and specific, including insurance, and on wrong output it stands behind nothing. Section 7.4 of the Master Terms, Version November 2025, gives a defense and indemnity for third party intellectual property claims in WIPO Copyright Treaty countries, with five named exclusions and modify, license, or terminate and refund remedies. Section 7.8 caps Litera's liability for direct damages at fees paid under the ordering document in the preceding twelve months, with the indemnity outside the cap. Section 7.7 excludes consequential loss for both parties except under the indemnities, and section 7.3 disclaims accuracy and results. Section 2 of the SaaS Terms warrants substantial conformance with the documentation, with repair or refund as the remedy. The Security Addendum commits Litera to keep cyber and technology errors and omissions insurance with an A.M. Best rating of at least A- and financial size of at least VII. Section 3(a) of the Generative AI Terms, Version April 2026, disclaims any liabilities, warranties or representations on output, notwithstanding the Master Terms. So the recourse for a wrong AI output, readable before signing, is the customer's own verification.

Word, Outlook, Teams and SharePoint all run Harvey inside them, and Vault syncs documents from iManage, SharePoint and Google Drive. The iManage connection is a direct OAuth integration with an embedded web extension, and NetDocuments access follows the firm's existing permissions. Since October 2026, Harvey also applies the ethical wall policies kept in iManage Security Policy Manager to Threads, Vaults, Review Tables and Spaces, with Security Policy Manager remaining the system of record. Intake takes requests from email, Jira, Slack, ServiceNow, iManage, Google Drive and SharePoint and sends updates back to the source channel. Command Center syncs employee profiles through SCIM. The Connector Library, generally available since October 2026, connects Harvey through MCP to tools such as Ironclad, Docusign, Intapp, Aderant, Everlaw, LawVu, Box, Gmail, Slack and Jira, and customers can add their own MCP connectors. Admins choose which connectors are enabled, who may use them and which tools need human confirmation, and each connector passes Harvey's security review before launch. LexisNexis content reaches research through Ask LexisNexis.

Real integrations are named and documented, without saying what moves in which direction. The Lito site states that Litera is delivered inside Microsoft Word and Outlook, Google Workspace, a web app, Apple iOS and a Windows and Mac desktop client, and names iManage and NetDocuments as connected document management systems. It lists documents, emails, timecards, financial data, HR data and legal research as sources Lito reads from and writes back to, and states that MCP connectors are governed centrally by the firm. The October 2025 press release adds that Lito connects with Kira and NetDocuments. Section 3.2 of the Master Terms records that the software is meant to be used with third party office software and document management systems the customer procures. Nothing published gives an implementer's description of what syncs, in which direction, and what a firm must configure. An integrations page is published at litera.com/integration.

Microsoft Azure hosts Harvey's cloud environment, with each customer's data logically separated and access set by role. Customers can keep data in region in the EU or Switzerland, the United States or Australia, and sign in through a separate US, EU or AU login. Customers with localization needs can also have processing in the EU and Switzerland or in Australia, and that commitment extends to Harvey's subprocessors. The Subprocessor List gives each model provider's processing locations. Where Anthropic or Mistral models are not offered in a chosen region, Harvey reaches them through hosting platforms such as AWS Bedrock and Google Vertex. Federal customers can use Harvey within Knox Systems' FedRAMP Class D certified cloud environment. What changes between plans or tiers is not stated.

Deployment options, regions and where processing happens as distinct from storage are all published. Section 2.1 of the Master Terms, Version November 2025, defines two deployment types, SaaS and on premises, and the Lito site adds a Windows and Mac desktop client for firms whose most sensitive work cannot move to the cloud. The Litera One / Lito subprocessor register says the application is hosted on Microsoft Azure in US, Canadian or European regions. Amazon Bedrock inference runs in the same regions as the platform, and inputs are not passed to third party model providers. Anthropic inference is hosted in the United States, with inputs and outputs retained up to thirty days, so a customer hosted in Europe can read that generative processing may leave the storage region. Section 2(d) of the Generative AI Terms commits to hosting models in the same jurisdiction as the SaaS software where commercially and technically feasible, and otherwise in a jurisdiction with comparable standards. Whether the service is single or multi tenant is not stated, and nothing published says the residency options differ by Draft package.

Schellman audits Harvey for SOC 2 Type II and ISO 27001 every year, and the 2026 audit and certification renewal are complete. The trust center also lists ISO 27701, ISO/IEC 42001, AIUC-1, IRAP, TX-RAMP Level 2, GDPR, CCPA and an EU AI Act item. It publishes Statements of Applicability for ISO 27001, 27701 and 42001. In October 2026 Harvey announced FedRAMP Class D certification, operating within Knox Systems' FedRAMP Class D certified cloud environment for federal agencies and contractors. Schellman, NCC Group and Bishop Fox perform external audits, and third party penetration tests run every year. Reports, the CAIQ v4.0.3, SIG Core and SIG Lite questionnaires and the data flow and network diagrams are released through a self serve access request, with bulk download.

Certification is real and stated on a working trust center, without scope, dates or a report reachable without a request. The SafeBase portal at trust.litera.com lists SOC 2 Type 2, SOC 3, ISO/IEC 27001:2022 with a statement of applicability, GDPR, DORA and NIS 2. It lists the reports, a penetration test report, a security whitepaper, cyber insurance, a CAIQ and SIG Lite behind a Get access flow that offers to start a security review. The flow does not say whether access is self serve on an email or click through, or runs through a sales conversation. The Security Addendum states that independent auditor reports are available to the customer on request, that Litera's ISO 27001 certification covers its corporate offices, and that cloud providers are audited under SOC 2 Type II and ISO 27001. No auditor, audit period or scope statement is published. The trust center overview describes the Litera Platform and Lito, so it covers the product.

The Subprocessor List names ten AI providers with their processing locations, among them Microsoft, OpenAI, Google Cloud, Amazon Web Services, Anthropic and Mistral. ElevenLabs supplies voice services, DeepL machine translation, and Baseten and Fireworks.ai further model hosting. Claude and Gemini models run through Amazon Bedrock and Google Vertex AI, as Harvey's subprocessor FAQ explains. Model providers work under zero data retention, with no human review and no training on customer data. Harvey Tenet, in research preview since August 2026, is Harvey's own open weight model, post trained from Moonshot's Kimi K3 with Fireworks for long legal tasks. Harvey says it was trained on synthetic, public legal and expert data, with no customer data. Where Tenet runs and which customers can use it are not stated. Under the DPA, customers who subscribe to the list receive notice at least 30 days before a new subprocessor is added and may object. Which model serves which feature is not stated.

Providers are named, where they run is stated and a change notice channel exists; the models themselves are not named, and three surfaces give three different provider lists. The Litera One / Lito subprocessor register, the binding disclosure, names Anthropic for model inference, hosted in the United States, possibly through Amazon Bedrock or Microsoft Azure Foundry. It names Amazon Bedrock for inference in the platform's own region and You.com for web search, and offers a registration link for notice of subprocessor changes. The Lito site describes routing across OpenAI, Anthropic and Gemini models. The Generative AI Terms, Version April 2026, refer to the terms of Azure OpenAI as terms Litera must follow and to GPT-4 as an example in the fair usage clause. OpenAI and Google appear on the marketing page and in the agreement and not on the Lito register, so the register alone does not show that OpenAI models may process prompts. No specific model version is named beyond that one example.

Licensing is per seat. The help center FAQ, updated November 6, 2026, puts the average cost at several hundred dollars per user per month, and the price depends on the number of seats, the products deployed and the contract terms. No rate card, plan or tier is published, and buying starts with a demo and a quote from sales. Ask LexisNexis, the research add on, is licensed separately from the base service. Fees are set in each customer's order form and invoiced under the Platform Agreement. If an order form is terminated early under the agreement, Harvey refunds a pro rata share of prepaid unused fees. Undisputed late amounts carry a finance charge at the Federal Funds Rate plus 1.5% a year.

Real pricing is published for part of the range, with the unit and structure stated. The Litera Store at litera.com/store offers Draft Base, Draft Pro, Clean (Metadact) and pdfDocs, each with a purchase button leading to a product page on store.litera.com, where the store price is shown. The packages page states that firms with ten or more lawyers request a demo and talk to sales. The Master Terms publish the unit: each license is for a single named individual (3.1), fees and currency are set in the order form, terms renew automatically for twelve months unless thirty days' notice is given (8.2), and overages on end users, lawyers or documents are invoiced (4.2). Lito is included with Draft Base, Pro and Advanced and with Kira at no extra charge, per the site and the October 2025 press release, and a store promotion offered twenty percent off during ILTACON. Enterprise pricing goes through sales.

Large law firms, mid sized firms, in house legal departments and professional services firms are the stated buyers. Since October 2026, federal agencies and contractors can also use Harvey within a FedRAMP Class D environment. Industry pages cover healthcare and life sciences, technology, energy and utilities, media, entertainment and telecom, and the public sector. More than 200,000 lawyers at over 3,500 organizations in more than 85 countries use Harvey, including 80% of Am Law 100 firms and five of the Fortune 10, and its in house page counts more than 1,000 enterprises. Solutions pages address litigation, transactional work and innovation teams, and research spans legal, regulatory and tax questions across jurisdictions. Firm sizes or practice areas the product is not built for are not stated.

The site has separate solution pages for small, large and global law firms and corporate legal departments. Role pages cover lawyers, knowledge management, marketing and business development, CIO and IT, finance, and talent and HR, and the Lito site walks through the associate, partner, transactional lawyer, knowledge manager, COO and marketing seats. Practice coverage is stated by skill rather than by practice area. The October 2025 press release names a launch library of eleven skills, with planned additions for capital markets, M&A, business development, litigation and compliance, and the Lito site names contract review and redlining, deal management, precedent surfacing and pitch drafting. Store purchases are limited to firms under ten lawyers, with larger firms routed to sales. No practice area, jurisdiction or language is named as out of scope. The home page claims 99 percent of the Am Law 100 and 74 percent of the Fortune 100 without a basis.