Harvey vs Magnar: how they compare in 2026
Jurisdiction decides most of this choice. Harvey is built for large firms and legal departments, and more than 3,500 organizations in over 85 countries use it, among them more than 80 of the AmLaw 100. Its agents, Vault, Knowledge and Spaces run inside Word, Outlook and iManage and draw context from NetDocuments, SharePoint and LexisNexis. Magnar is built for Latin American civil law. Its agent works from each country's official sources. In Chile those run from the Supreme Court and the Constitutional Court to the CMF and the SII, and the agent reads case files of up to 10,000 pages. It also acts in court and registry websites while the lawyer watches, and Monitor reports new rulings and regulations every hour. Harvey publishes no price, while Magnar sells Plus at $50 a month and team seats priced in UF. Harvey suits an international organization that needs one platform across offices and practice groups. Magnar suits a firm or legal department in Chile, Peru, Colombia or Argentina that wants local law cited to its official source.
At a glance
All 15 axes, side by side
The same grid applied to every vendor in the index, graded from public sources. Hover a grade to see what the letter means on that axis.
AI Centrality
How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.
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.
Everything turns on the agent. A lawyer states an objective, Magnar breaks it into stages of reading, research and drafting, and the lawyer adjusts or approves the plan before it runs. It then reads the full case file, retrieves legislation and case law with citations, works in the websites it needs and leaves a structured document with its sources. Case analysis takes files of up to 10,000 pages and answers questions, builds timelines and drafts from them, while tabular review pulls defined fields from hundreds of documents into one table. Sidechat works on the document that is open, Compare explains the changes between two versions, and Monitor runs an agent every hour over official sources and keeps only the findings that meet a lawyer's condition. Vaults and electronic signature are conventional modules sold inside the same subscription, and every document stored in Vaults is open to the assistant.
Citation Accuracy and Hallucination Disclosure
Whether the vendor publishes measured accuracy on citations and assertions, grounds output to primary sources, and says plainly what its system does when it does not know. Legal has a documented public record of fabricated citations reaching filed briefs, so an untested claim of accuracy is not evidence.
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.
Research answers keep each conclusion tied to the decision, regulation or official document behind it, and the original text opens in one click. The Chilean database lists its sources by name, from the Supreme Court, the Courts of Appeals and the Constitutional Court to the CMF, the SII, the Comptroller General, the Labor Directorate and the environmental tribunals. Magnar counts more than 70 million laws and court decisions across the region, and the coverage of sources, courts and periods varies by country. Clause 3.4 of the terms says the AI is probabilistic and can produce errors and hallucinations. No accuracy rate, test set or failure analysis is published.
Autonomy and Oversight Model
What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.
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. Intake applies routing rules the legal team defines to send each request to an approved agent, to Contract Intelligence or to a lawyer. Reports and reviews can run on a schedule, and admins govern access through roles and workspace permissions. The point at which an agent stops on its own, and what Harvey commits to when an agent's work is wrong, are not stated.
Approval points are built into each mode. A plan appears before any run, and the lawyer can adjust or approve it. Copilot proposes an action on the contract, case or vault on screen and carries it out only once the lawyer approves. In Word, edits arrive as tracked changes. When the agent works in a court or registry site, the lawyer watches the browser live and can take over at any point, and portal sign ins need the lawyer's own credentials. Scheduled tasks run with the app closed and alert only when something changes. The terms make the client, as a legal professional, responsible for reviewing and validating every output before it goes into a filing, a contract or advice. Filing a brief through a court's electronic filing system is among the tasks shown. Whether that step needs its own approval, a confidence threshold at which work stops, and recovery after an error are not described.
Operational and Outcome Evidence
Named, dated evidence that the product works in production at real firms or legal departments. Case studies with figures and identified customers count. Unattributed testimonials and launch announcements do not.
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. Other named customers include A&O Shearman, Gleiss Lutz, Cuatrecasas, Macfarlanes, KKR, PwC and Deutsche Telekom. The figures are usage counts and time savings as each customer reports them, with no stated method for how they were measured.
About 30,000 users and 350 law firms worked with Magnar in September 2026, a little over a year after it started, and more than 23,000 lawyers in Chile use it by its count. Partners at three firms are quoted by name and on video. Roberto Lewin of Barros Silva Varela & Vigil speaks about the effect on associates and on hiring. Diego Rodríguez of A/C/R Legal talks about the changing role of lawyers, and Germán Barrios of Bafur Abogados calls AI a vital tool for firms. None of the quotes carries a figure, a date or a measured result, and no case study with outcomes is published. The research product's promise of hundreds of hours saved each month comes with no method or customer behind it.
Privilege and Confidentiality Posture
How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.
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.
Training is settled in writing. Clause 6.6 licenses customer content to Magnar only to process it for the service and states that Magnar will not use it to train its models or share it with other customers. Under the privacy policy, content is processed transiently and each organization's data is visible only to its own users. Magnar's agreements with model providers such as OpenAI, Google and Anthropic bar training, and any provider retention is limited to abuse monitoring for at most 30 days. Stored history stays until the user deletes it and is erased within 30 days of account closure. Privilege and work product are not addressed by name, and separation inside a firm runs on Magnar's own folder permissions rather than matter walls.
UPL and Professional Responsibility Posture
Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point. Where the advice line is not the duty a product raises, the axis is read through the nearest professional duty it does raise: judicial conduct rules and the reviewing duty for products sold only to courts, and the duty to bill for time actually spent for products that draft time entries.
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.
Clause 3.3 of the terms describes Magnar as a technology assistant, not a law firm, and rules out legal representation and legal advice. Clause 3.4 states that the AI is probabilistic and can produce errors. It makes the client, as a legal professional, solely responsible for reviewing and validating any output before using it in a legal document, a filing or advice to anyone. The product is sold to lawyers, firms and legal departments, with a free plan and student pricing. Supervision of staff, competence duties and the jurisdictions where output may be relied on are not addressed.
AI Governance and Bias Disclosure
Published governance over model behavior: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
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.
ISO 42001, the AI management system standard, is in progress. Magnar describes it as covering transparency, fairness and accountability with controls for bias and explainability, and its trust center lists the certification as under way. The same trust center publishes more than 40 policies and controls, from secure development and change management to breach notification and personnel screening. No AI policy, release testing, accountable owner or finding about how output varies across countries, courts or practice areas is published.
AI Safety and Data Stewardship
Retention, deletion, access control, and what happens to prompts and documents after they are processed. Whether the vendor states its subprocessors and its incident practice, or leaves the buyer to assume.
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 is set out by data type. User content is processed transiently. Stored history and case files remain until the user deletes them, and everything is erased securely within 30 days of account closure unless a legal hold applies. Model providers may keep content for abuse monitoring for no more than 30 days. Data is encrypted with AES 256 at rest and TLS 1.2 or higher in transit, administrative staff work on least privilege access with multifactor authentication, and penetration tests run periodically. Subprocessors appear in the trust center with entity, country, service, data types and storage location, a list last updated on 25 September 2026. Among them are Google's Gemini API, AWS hosting and Amazon Bedrock, MongoDB Atlas, GitHub, ElevenLabs and Google Workspace. As a provider registered with Chile's National Cybersecurity Agency, Magnar notifies an affected user within three hours of a confirmed breach.
AI Liability and Recourse
What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.
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.
Under the terms of 6 April 2026, governed by Chilean law, total liability is capped at the fees the client paid in the 12 months before the event, and lost profit, lost opportunity and indirect damages are excluded. Magnar warrants that it will provide the service professionally and use commercially reasonable efforts to keep it available. Otherwise the platform is offered as is, with no promise that legal analysis is accurate, complete or binding. Against third party claims that the platform infringes intellectual property, Magnar defends the client, provided the client gives prompt notice and hands over the defense. No carve outs from the cap, no insurance and no remedy for wrong output are stated.
Practice Systems Integration Depth
How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.
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. 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. Aderant and Ansarada integrations are documented in the trust center, and customers can build their own integrations, including through an MCP connector library. LexisNexis content reaches research through Ask LexisNexis.
A Word add in analyzes the open document and proposes edits as tracked changes, and Magnar also works in Outlook. Google Drive connects on paid plans, while OneDrive, NetDocuments and iManage are listed as coming soon. Magnar Drive mirrors Vaults in Finder or File Explorer. A lawyer edits a file in Word, Magnar records the new version on save, permissions stay in Magnar, and files download only when needed. Client datarooms let a client sign in by email to the folders the firm shares. Court, registry and apostille sites are reached through the agent's own browser, and Google Workspace sign on and Gmail access are optional. The Enterprise plan adds custom integrations, with no description of what they move.
Deployment Model and Data Residency
Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.
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. 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. What changes between plans or tiers is not stated.
Lawyers reach the product in a browser or through desktop and mobile apps, as cloud software. Each organization's data is isolated from other Magnar users. Hosting, the database and model inference run in the United States on AWS, MongoDB Atlas and Google, and the privacy policy covers processing outside Chile through adequacy findings or standard contractual clauses. No regional hosting choice, private deployment or difference between tiers in where data sits is offered.
Security Certifications and Trust Center
Independent attestation a buyer can pull without a sales call: SOC 2, ISO 27001, penetration test summaries, a trust center with current reports and named scope rather than a badge image.
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, GDPR, CCPA and an EU AI Act item. It publishes Statements of Applicability for ISO 27001, 27701 and 42001. 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.
ISO 27001:2022 certification covers the servers, databases, internal processes and handling of customer information, and a SOC 2 Type 2 report covers security, availability and confidentiality controls over a period. Compliance status, controls and more than 40 policies sit in a trust center, with documents released through a request access step. The auditor and the report period are not shown. ISO 42001 is in progress.
Model Supply Chain Disclosure
Which models sit underneath, whose they are, where they run, and whether the vendor commits to telling customers when that changes. A legal buyer inherits every dependency it cannot see.
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. 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.
The subprocessor list names Google's Gemini API and AWS with Amazon Bedrock for model inference, both in the United States. The privacy policy names OpenAI, Google and Anthropic as examples of model providers used through enterprise APIs under zero retention agreements. Users choose between General, Light and Pro modes, and the specific models behind each are not named. Customers can subscribe to the trust center for notice of subprocessor changes, the list last changing on 25 September 2026, though no notice period for a change of model is committed.
Commercial Transparency
Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.
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. No rate, plan, tier or unit of charge appears on Harvey's site, and buying starts with a contact sales form.
Prices are published for individuals and teams. Free gives 5 credits a month with no card, and Plus costs $50 a month with 30 credits a week, three modes, a document library, Google Drive and one monitor. Team seats are priced per user each month in UF, the Chilean unit of account, plus VAT, with unlimited users and a two week free trial. Base is 0.5 UF with 10 weekly credits, Plus 1.5 UF with 40 and Pro 5.25 UF with 130. Credits pay for questions and file processing. Extra use on team plans is billed at whichever costs less, extra credits or the next plan up, and price rises need 30 days' notice. Enterprise is quoted, and there are no setup fees.
Firm and Practice Coverage
Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.
Large law firms, mid sized firms, in house legal departments and professional services firms are the stated buyers. Industry pages cover healthcare and life sciences, technology, energy and utilities, and media, entertainment and telecom. 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. 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.
Lawyers, law firms and corporate legal departments use the product in Chile, Peru, Colombia, Argentina, Uruguay, Ecuador and Costa Rica, and Brazil and Mexico are opening with local teams and offices. Visitors from other countries are told it is not yet available there. Research depth follows each country's sources. In Chile those span civil, criminal and labor courts, tax, financial market, competition, environmental, consumer and social security bodies, and the Inter American Court of Human Rights. Availability of sources, courts and periods varies by jurisdiction. Students get a validated discount on individual plans, and teams of any size can mix seat levels. Which practice areas fall outside the product is not stated.
The 12 legal signals, side by side
Recorded rather than graded. These are the questions a practitioner has to answer before a tool touches a client matter, and the answers are taken from public material only.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
Under the Platform Agreement, Harvey will not train any AI model on customer content or customer data, and its subprocessors will not either. The term covers both uploaded documents, which Harvey calls customer data, and the prompts and answers it calls customer content. A firm that asks for it can have a bespoke model trained on its own data and used only for that firm. Harvey may use usage data, such as feature and session statistics, to develop and improve the service, and usage data excludes customer data and content.
Clause 6.6 of the terms licenses customer content to Magnar only to process it for the service and states that Magnar will not use it to train its AI models or share it with other customers. The privacy policy repeats that content does not train Magnar's foundation models unless the user instructs otherwise. Magnar's agreements with OpenAI, Google and Anthropic bar those providers from training on it, and Google Workspace data stays outside model building under Google's limited use rules.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
Customers decide what to upload, how long it is kept and whether it can be shared inside the organization, and they can delete data at any time. No default retention period is published. Model providers work under zero data retention, and subprocessors other than cloud storage providers may not retain or log content for human review. After termination, Harvey deletes remaining customer data and content within 30 days unless the customer instructs otherwise.
Content is processed transiently. If the user keeps chat history or case files, they remain until the user deletes them, and on account closure everything is erased within 30 days unless a legal or court order requires keeping it. Clients are expected to download their content before a subscription ends. Model providers may hold content for security and abuse monitoring for no more than 30 days. No zero retention setting is described.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
NetDocuments permissions carry through, so a user sees only the cabinets, matters, folders and files they can already open, and Harvey does not change those permissions. The iManage connection is a direct OAuth integration that respects iManage permissions and ethical walls. Harvey also syncs the firm's own ethical walls provider and blocks walled users from reaching or sharing restricted content.
Each organization's data is reachable only by its own users. Inside the firm, Vaults organize documents by organization, client and project with granular permissions, and Magnar Drive keeps those permissions when files open in Word. Client datarooms show an outside party only the folders the firm allows, watermarking each page and logging each action. Permissions live in Magnar rather than following a firm's existing document system.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
If law or a court order compels Harvey to disclose a customer's confidential information, the Platform Agreement commits it to advance written notice where legally permitted and to help seek confidential treatment. The DPA makes the same commitment for law enforcement requests for personal data. Where the law forbids notice, Harvey seeks a waiver and tells the customer once the prohibition ends. Its Law Enforcement Requests policy sends agencies to the customer first, challenges requests its lawyers consider unlawful and answers foreign requests only through channels such as a Mutual Legal Assistance Treaty. No transparency report is published.
The privacy policy lists compliance with tax, judicial or administrative orders from a competent authority among the purposes for processing data. Neither it nor the terms commit to telling the customer about such a request or to resisting it, and no transparency report is published. Breach reports go to Chile's National Cybersecurity Agency under Law 21.663, with notice to affected users within three hours.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
Knowledge connects questions to legal databases, curated public sources and Harvey's data partners, more than 500 legal sources across jurisdictions. The Subprocessor List names RELX, doing business as LexisNexis, for Ask LexisNexis, and Parallel Web Systems and you.com for web browsing and knowledge sources. Harvey's AI Policy commits it to obtaining the rights and licenses its data sets require. No source list by jurisdiction, license basis per source or update schedule is published.
Each country has its own source list. In Chile the database names the Supreme Court, the Courts of Appeals and the Constitutional Court. It adds civil, criminal and labor courts, the environmental and competition tribunals, and regulators such as the CMF, the SII and the Superintendency of Social Security. The Comptroller General, bills before Congress and the Inter American Court of Human Rights are also listed. Monitor checks selected sources every hour for new publications. The rights basis for each source and the update schedule of the research database are not stated.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
Knowledge answers from more than 500 legal sources and draws LexisNexis content through Ask LexisNexis. Harvey does not describe a treatment signal or a later history check on the authorities it returns.
Answers cite the ruling, regulation or official document behind each conclusion, and the original opens in one click. Magnar describes tracking how each law is applied and comparing lines of precedent across courts. No flag on a decision that has been reversed or overruled, and no check of subsequent history on a cited authority, is described.
Refusal and Uncertainty Behavior
What does the product do when the answer is not in the corpus?
BigLaw Bench publishes hallucination rates and explains how Harvey defines and detects hallucinated claims, and agents cite a source for every claim. No setting or behavior is described for declining to answer when Harvey cannot ground a response.
Errors and hallucinations are possible, the terms warn, because the AI is probabilistic, and every research answer keeps its conclusions linked to their sources. Monitor separates the documents it analyzed from those that meet the lawyer's condition, so a run with no match reports none. What the agent does when the sources do not support an answer, whether it declines or flags low confidence, is not described.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
The AI Hallucination Cases database, maintained by Damien Charlotin, records court decisions worldwide that address hallucinated AI content and names the tool involved where known. It lists one matter naming Harvey, Hill v. Foundation Media LLC (S.D.N.Y., No. 25-cv-5947). On 22 September 2026 the court directed defense counsel to say whether AI was used on a brief with two questioned quotations and a citation, and a week later it took no further action.
Harvey is named only in counsel's own letter to the court, which says the brief was prepared with Harvey. No order makes any finding about Harvey or its output. A Galicia High Court order of 7 July 2026 lists Harvey AI among verified legal AI tools while setting a €1,800 fine, again with no finding about the product. Harvey's agreement states that output is AI generated and may contain errors.
The AI Hallucination Cases database maintained by Damien Charlotin, which records court decisions worldwide that address hallucinated AI content and the tool involved where known, has no entry naming Magnar. Magnar's terms tell users the AI can hallucinate and require them to verify output before using it in a filing.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
Harvey's terms call the service a research tool whose output is not legal advice, and the agent pages leave the judgment on each task with the lawyer. No named ethics opinion appears, such as ABA Formal Opinion 512 or a state bar opinion.
Professional responsibility for any output sits with the client as a legal professional, who must review and validate it before using it in a filing or advice. Magnar's own position is that it supplies the technology while judgment and legal decisions stay with the lawyer. No ethics opinion or guidance from a bar association in the countries Magnar serves is discussed, and no mapping of the product to professional rules is published.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
ROI calculators for law firms and in house teams sit on Harvey's site, and customer stories report time saved, such as Repsol's. Command Center reports adoption by user, practice group and role. Nothing records AI assisted work per matter for billing, or addresses how a firm bills for that work or discloses it to clients.
Law firms and legal departments buy Magnar, and its research product promises to save hundreds of hours a month in finding and analyzing case law. Organization admins see how much each user consumes, which tracks usage by person rather than by matter. No guidance on billing clients for AI assisted work or disclosing it is published.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
The trust center offers a Data Processing Addendum, a Business Associate Addendum and a Data Subject Requests item. It adds the CAIQ v4.0.3, SIG Core and SIG Lite questionnaires, data flow and network diagrams, a HIPAA report and a Security Welcome Packet. Sensitive items are released through a self serve request with bulk download. Statements of Applicability for ISO 27001, 27701 and 42001 show what each certification covers.
The public Subprocessor List names every provider with its purpose and location, and subscribers hear of additions 30 days ahead.
The trust center publishes a subprocessor list giving each provider's legal entity, country, service, data types and storage location, with Google's Gemini API and Amazon Bedrock named for model inference, and customers can subscribe to change notices. Policies and audit documents are released through a request access step. No client facing disclosure pack is published.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
Audit logs are a standard enterprise control, agents log every step, and every claim links to the source passage behind it. Command Center reports usage by user and feature. No export brings together the model used, the sources retrieved and the human review for one document.
Every step of the agent's plan stays visible, and each conclusion cites the source behind it. Vaults keep every version with a line by line comparison and record who uploaded what and when, and datarooms log each action. A brief filed through a court system comes back with its filing receipt. No export ties a filed document to the model used, the sources retrieved and the lawyer who verified it, and no court disclosure template is published.
The questions both sides leave open
Derived from the records above rather than written, so it cannot favor either vendor. Take these into both conversations and ask each side the same question.
- Good Law Verification
- Refusal and Uncertainty Behavior
Which one fits
Choose Harvey if
- You work across many offices and legal systems. Harvey operates in more than 85 countries and counts more than 80 of the AmLaw 100 among its users.
- Your documents sit in iManage or NetDocuments. Harvey runs inside iManage, Word and Outlook and draws context from NetDocuments, SharePoint, Google Drive and LexisNexis.
- You need data held in a specific region. Harvey keeps data in the EU or Switzerland, the United States or Australia and enforces the ethical walls a firm already maintains.
Choose Magnar if
- Your matters run under Chilean, Peruvian, Colombian or Argentine law. Magnar researches each country's official courts, regulators and gazettes, and every citation opens the original text.
- You want the agent to act on court and registry websites. It downloads filings and certificates and can file a brief while you watch the browser live, taking over when a sign in is needed.
- You want to start small with a visible price. A free plan, Plus at $50 a month and team seats from 0.5 UF per user let a firm try Magnar before a contract.
In summary
Harvey
Harvey sells a legal AI platform to large law firms, in house teams and professional service networks in more than 85 countries. Agents carry work through to a result, Vault analyzes large document sets in bulk, Knowledge answers legal, regulatory and tax questions, and Spaces opens matters to other organizations. According to the AI Legal Index, Harvey holds SOC 2 Type II, ISO 27001, ISO 27701 and ISO 42001. It offers in region storage in the EU or Switzerland, the United States and Australia, and publishes no price.
Magnar
Magnar SpA, founded in Santiago in 2025, sells an AI agent for lawyers in seven Latin American countries, with Brazil and Mexico next. The agent plans legal work for the lawyer's approval, then researches official legislation and case law, analyzes case files and drafts. According to the AI Legal Index, Magnar's terms bar training on customer content, and its trust center names Google's Gemini API and Amazon Bedrock among its subprocessors. Individual and team prices are published.
Questions buyers ask
Does Harvey or Magnar cover Latin American law?
Harvey draws on more than 500 regional knowledge sources and is used in more than 85 countries. Magnar is built for Latin America and publishes a source list for each country it serves. In Chile that list covers the Supreme Court, the Courts of Appeals, the Constitutional Court, regulators such as the CMF and the SII, and bills before Congress. From the AI Legal Index, based on each vendor's own published materials as of October 13, 2026. No vendor pays for placement.
Can either one act on court websites?
Magnar's agent works in court, registry and apostille sites through a browser the lawyer watches live. It downloads new filings, fetches certificates and files briefs, and hands control back when a portal needs the lawyer's password. Harvey's agents run work end to end across its own platform and the systems it connects to, such as iManage and SharePoint. From the AI Legal Index, based on each vendor's own published materials as of October 13, 2026. No vendor pays for placement.
How do Harvey and Magnar handle client data?
Harvey states that its Platform Agreement bars training underlying models on inputs, outputs or uploaded documents, and customers set retention and can delete data at any time. Magnar's terms bar training on customer content, content is processed transiently, and its model providers may hold data only for abuse monitoring, for up to 30 days. From the AI Legal Index, based on each vendor's own published materials as of October 13, 2026. No vendor pays for placement.
What does each one cost?
Harvey quotes after a demo. Magnar publishes its prices: a free plan with 5 credits a month, Plus for individuals, and team seats in UF at 0.5, 1.5 or 5.25 per user each month with weekly credits. From the AI Legal Index, based on each vendor's own published materials as of October 13, 2026. No vendor pays for placement.
Neither vendor paid for inclusion, placement or a grade, and neither reviewed this page before it published. Everything on it comes from public material on the dates shown. How the index grades.
Harvey's latest raise, announced in 2026, was $550M at a $15.5B valuation. Magnar SpA closed an $8M Series A in September 2026 after Carey, a Chilean law firm, invested in January, and a third party research profile lists Harvey among Magnar's competitors. Neither vendor reviewed this page.