A
AI.Law
AI.Law is a litigation platform for solo and mid-size firms that combines drafting tools with a full practice-management system, sold as roughly sixty named tools under one login. The drafting side is the core: complete complaints with AI-suggested claims, answers with affirmative defenses, interrogatories and requests for production and admission, case-aware discovery responses and objections, deficiency letters, motions to compel, dismiss and for summary judgment, opposition and reply memoranda, and court-compliant page formatting with captions and line numbers.
Around it sit analysis tools that read the file: data extraction into timelines, parties, costs and liens, deposition analysis that cross-references transcripts against written discovery, an inconsistency finder across depositions, insurance coverage reports, a dismissal risk analyser, and a three-dimensional evidence map linking claims, defences and gaps to the underlying documents. A plaintiff-side group produces thirty-two-section demand letters, medical chronology and causation reports, and AI-generated medical illustrations for trial.
Legal research runs against Court Listener caselaw across every US jurisdiction, and a brief analyser checks citations and surfaces counter-authorities. The company publishes a written Citation Accuracy Guarantee: every claim carries a clickable citation to the source document and page, and if a citation link is wrong it credits the account. Underneath all of it is a practice-management layer offered free and without AI, covering client intake, conflict checks, matter and document management, calendaring and deadlines, a client portal, e-signature, time entry and IOLTA trust accounting.
Firms administer the AI through fifteen built-in permission roles and controls that switch AI categories on or off firm-wide or matter by matter. AI.Law was founded by the litigation attorney Troy Doucet and holds US Patent No. 12,461,932.
Capability grades
All 15 axes, graded from public sources on the date shown. 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.
The models are the engine of a core capability layered on a product that would function without them, and this is the cleanest instance of that shape in the corpus, because the vendor sells the underlying product separately for nothing. The Free Forever tier is described in the vendor's own words as the full practice setup with no AI tools: unlimited matters and seats, CRM, lead pipeline, intake, conflict checks, calendar, billing, IOLTA trust accounting, document storage, e-signature and workflow automation, at zero cost and with no time limit.
That is a working practice-management system standing on its own. The AI is then sold on top, by seat and by role, at three price points, and the paid tiers are organised around what the models do: understand, draft, win. The drafting and analysis catalogue is large and genuinely model-driven, from complaints and discovery to deposition analysis and medical chronologies, so this is not a thin layer. But the test for A is whether anything remains when the models are removed, and here the vendor answers that question itself by publishing the answer as a product. Verified 12 September 2026.
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.
Grounding is real, documented and backed by a commercial remedy, short of any published measurement, which is B at the top of the band. The grounding architecture is described rather than asserted. Output is confined to the file: the vendor states that the AI can only cite what is in the matter's files, that every claim carries a clickable citation to the source document and page, and that every cited entry links back to the source page or transcript line.
Caselaw is grounded in a named external source, Court Listener, with the vendor stating that every case citation is verified against it and that its research tool produces no invented cites. A separate brief analyser checks citations and surfaces counter-authorities. On top of that sits a published Citation Accuracy Guarantee at its own URL, which sets out a three-step mechanism, the AI drafts with a citation on every claim, the user clicks through to the exact source, and if a citation is wrong the vendor credits the account. **That remedy is the first of its kind located in this corpus** and it is why this row sits where it does.
What holds it off A is measurement. No accuracy figure, test set, error rate, sample or third-party evaluation is published anywhere, and the guarantee is a commitment about linkage rather than a demonstration of correctness: it addresses whether a citation points at a real source, not whether the proposition it supports is right. Verified 12 September 2026.
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.
The oversight position is published and includes a real customer-side control, short of any threshold at which the system acts alone. Two published features do genuine work here. AI Controls lets a firm toggle AI categories on or off firm-wide or matter by matter, which is an administrative switch over what the models may touch and is rarer in this corpus than it should be; and Permissions and Roles supplies fifteen built-in roles with matter visibility, so who can invoke which tool is a configurable question rather than an open one.
The verification loop is published as a workflow rather than a disclaimer: the Citation Accuracy Guarantee makes the user's click-through the second of its three steps, so checking the source is described as part of using the product. The tier structure carries an implicit supervision model that the vendor states openly, the Associate seat being for associates, of-counsel attorneys and senior paralegals drafting under supervision.
What is missing for A is the boundary. Nothing published states a confidence threshold, describes what the system does when the file will not support an answer, sets out an escalation route, or distinguishes work the AI completes from work it proposes. On a platform that drafts complaints and discovery responses, that boundary is the thing a reviewing partner would most want written down. Verified 12 September 2026.
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.
Named customers without figures, which is the B band's stated shape. Three firms are named on first-party surfaces with attributed quotations rather than anonymous praise: Liu Law Firm on searching key information across large volumes of legal documents; Akiva Shappiro of Akiva Shapiro Law, PLLC, quoted as a solo litigator on value for money and reduced drafting time; and Sean Akari of Akari Law Firm on accuracy and speed.
Attribution to a named individual at a named firm is what separates this from the band below. What is absent is measurement attached to any of them. The figures the vendor does publish float free of the customers: that a Paralegal seat pays for itself by reducing around twenty hours of paralegal work a month, that an Associate seat pays for itself with one drafted motion or one set of discovery responses, and, in a third-party interview, that deposition summarisation reduced billable time in an insurance defence case by up to twenty per cent.
None states a basis. Two things are recorded and deliberately not credited: the founder's own law firm has handled over 2,200 matters and resolved more than $500 million in claims, which is the vendor's provenance rather than a customer outcome; and a case-studies page is published and was not opened, so it is named here as the route by which this row would move. Verified 12 September 2026.
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.
Confidentiality is asserted in general terms on the surfaces that can be read, and the instrument that would settle it could not be rendered, which is C and the note must be read with that limit in view. What is readable is real but partial: fifteen built-in permission roles with matter visibility, automatic conflict screening, AI controls that can be switched off firm-wide or per matter, a scoping statement that the AI can only cite what is in the matter's own files, and an assertion of a completed SOC 2 Type II examination.
Together those describe a product built with a firm's confidentiality obligations in mind. What is not readable is the part that would carry the commitments. The Terms of Service and Privacy Policy are published, ungated and linked in the footer of every page, and both render client-side; neither the direct fetch nor a search of the index recovered their bodies on the date shown. **That is a limit on this index's reader under R20, not a finding about the vendor, and this row is graded accordingly rather than at D.** So no first-party statement was located on training on client data, retention, deletion, segregation between matters at the model layer, or what any model provider may retain, and nothing addresses privilege or work product at all.
One tension is recorded rather than resolved: Cross-Matter Intelligence and Firm Intel run retrieval across every matter in the firm, which is exactly the capability a matter-level wall would need to constrain. Verified 12 September 2026.
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.
A real published position on professional responsibility, short of full treatment. The strongest element is a named engagement rather than a disclaimer: the vendor markets the platform as ABA Opinion 512 Guided, putting the profession's own generative AI ethics opinion on the face of the product. The Citation Accuracy Guarantee frames itself the same way, opening on the fact that courts are sanctioning attorneys for AI-hallucinated citations and positioning the guarantee as the answer to that exposure.
The audience is unambiguous throughout, the product being sold only to litigation attorneys and law firms and priced by professional role, and the Associate tier is described as being for those drafting under supervision, which is a supervision statement in commercial clothing. What is missing for A is the rest of the treatment. Nothing published states that output is not legal advice, that the vendor is not a law firm, or that a lawyer must exercise independent judgement before filing; nothing sets out how a supervising attorney discharges competence and supervision duties over machine-drafted pleadings; and no jurisdictional limit appears despite caselaw coverage being claimed across every US jurisdiction.
The Terms of Service would ordinarily carry that language and could not be rendered by this index. Verified 12 September 2026.
AI Governance and Bias Disclosure
Published governance over model behaviour: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
Responsible AI is asserted through product controls rather than through a governance programme, which is C. What exists is configuration, and it is genuine: AI Controls to switch AI categories on or off firm-wide or per matter, fifteen permission roles governing who may invoke what, and a stated alignment with ABA Formal Opinion 512. A completed SOC 2 Type II examination is asserted, which is a controls attestation rather than an AI governance one and is graded on the certifications row.
None of what the higher bands require is published. No responsible AI or AI governance page exists, no framework is named, nobody inside the company is identified as accountable for AI decisions, nothing describes what is evaluated before a model or a tool ships, and no testing regime of any kind is set out. Bias is addressed nowhere, which is worth naming on a platform whose outputs include dismissal risk scoring, settlement correspondence and medical causation analysis, all of which can vary systematically by claimant, venue or injury type.
The Citation Accuracy Guarantee is the one place where the vendor commits to a standard and accepts a consequence for missing it, and it is confined to citation linkage. Verified 12 September 2026.
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.
Access is documented and the rest of the stewardship question is not readable, which places this at C rather than lower. The access half is published in product terms and is specific: fifteen built-in roles with matter-level visibility, automatic conflict screening, AI controls that can be disabled firm-wide or matter by matter, firm-wide document search scoped by permission, and a completed SOC 2 Type II examination asserted on the vendor's own pages.
The platform also holds trust money through IOLTA accounting, which raises the stakes on access control and is recorded as context. The other half of the axis, what happens to documents and prompts after processing, is not established: no retention period, deletion right, subprocessor list, encryption statement, breach notification commitment or incident practice was located on any readable surface. **The Privacy Policy that would ordinarily carry all of it is published and ungated and renders client-side; this index could not read its body on the date shown, and R20 means that limit is not converted into a finding against the vendor.** The row would move on a reading in either direction.
Named here so a future reader knows exactly what was and was not seen: the marketing, features, pricing and citation-guarantee pages were read in full; the Terms and Privacy pages returned their frames only. Verified 12 September 2026.
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.
A real published position on what the vendor stands behind, short of the full picture, and it rests on something no other record in this corpus has. The Citation Accuracy Guarantee is published at its own URL and is written as a commitment with a consequence: every hyperlinked citation is verified against the source document, and **if a citation link is wrong the vendor credits the account.** The page says so twice and describes it as a written guarantee.
That is a specific, invocable remedy attached to a defined failure mode of the output itself, which is the question this axis asks, and it is offered to every customer without negotiation. Its limits are equally clear and belong in the reading. The remedy is an account credit, not an indemnity, and its scope is citation linkage, so it does not reach a wrongly drafted claim, a missed affirmative defence, a mistaken limitation calculation or an incorrect medical causation analysis.
Nothing published addresses indemnity, insurance, or a warranty covering the substance of any document the platform drafts. The Terms of Service, whose meta description records that it covers subscription, billing, arbitration, class action waiver and governing law, could not be rendered by this index, so the general cap and any exclusions are unestablished and would move this row in either direction. Verified 12 September 2026.
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.
Real integrations, named and published as product features, short of the depth an implementer could work from. The named connections cover the systems a small litigation firm actually runs on: Google and Outlook calendar sync for matter events and deadlines, email integration inside client and contact management, and Stripe for invoicing and payments alongside IOLTA trust accounting. Two further connections are unusual for a vendor at this size and are published as plan features rather than as roadmap: a RESTful API described as reaching any AI.
Law data point, and MCP connections for attaching external AI agents to the platform, with white-glove onboarding for both included at the Partner tier. That the vendor exposes the Model Context Protocol at all puts it ahead of most of this lane. What holds it off A is documentation. No developer portal, API reference, authentication description, field mapping or sync-direction statement was located, and nothing describes what a firm must configure to make any connection work.
The platform's own answer to integration is largely to absorb the adjacent systems rather than connect to them, since it ships its own CRM, document management, calendar and billing. Verified 12 September 2026.
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.
Cloud delivery is implied and neither the tenancy model nor the region is stated, which is C in its exact terms. The product is plainly a hosted multi-tenant web application: it is reached through a browser sign-in, provisioned instantly on self-serve signup with no credit card on the free tier, priced per seat per month, and the pages themselves load a hosted workspace. Nothing beyond that is published. No cloud provider is named, no region or data centre is identified, no residency option is offered or refused, no single-tenant or private deployment is mentioned, and nothing distinguishes where documents are stored from where model inference happens, which matters because the platform routes work to ten large language models it does not name.
There is no on-premises or self-hosted option described anywhere, which for a product holding client files and IOLTA trust records is itself a fact a buyer would want stated rather than inferred. The Privacy Policy is the document that would ordinarily carry a processing-location statement, and it renders client-side and could not be read by this index on the date shown, so a reading could move this row. Verified 12 September 2026.
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.
A certification claim appears on the site with no scope, no date and no report available, which is C. The claim itself is specific enough to be meaningful and is made on the vendor's own pages: a completed SOC 2 Type II examination, asserted in the same hero strip as the patent number and the ABA Opinion 512 alignment. That is the right standard for this product class and it is a Type II rather than the weaker Type I, so the assertion is not empty.
Everything a buyer would need to test it is missing. No auditor is named, no observation period or report date is given, no scope statement identifies which trust services criteria were in scope or which systems the examination covered, no certificate or attestation letter is published, and there is no trust centre, security page or documented route to request the report. No penetration testing regime, vulnerability disclosure programme or encryption statement was located anywhere.
Recorded and expressly not credited because it is not first-party publication: a fuller scope description, naming security, availability, processing integrity and confidentiality, appears in vendor-authored company boilerplate on a third-party business directory. If that text were on the vendor's own site this row would move. Verified 12 September 2026.
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 vendor refers to models without identifying what sits underneath, which is C. The one published fact about the supply chain is a count: the AI Associate is described as letting a user chat with ten large language models, with every answer citing source pages. Ten is a number, not a disclosure. No provider is named, no model or version is identified, nothing states which model performs which task or which one drafts a complaint as against summarising a deposition, no default is stated, and nothing commits to notifying a customer when the set changes.
There is no subprocessor list to fall back on, and the Privacy Policy that might name processors could not be rendered by this index. What is disclosed, and credited on other rows rather than here, is the data side: Court Listener is named as the caselaw source and the vendor states that generated text can only cite what is in the matter's files. Those describe where the grounding comes from, not what generates the text.
The gap is the sharper for the choice being exposed to the user: a firm is invited to select among ten models for work on privileged client files without being told whose they are. Verified 12 September 2026.
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.
A buyer can learn what this costs without entering a sales process, which is the A band, and this is the fullest commercial disclosure located in pull 8. Rates are published outright: Paralegal seats at $149 per seat per month, Associate at $299, Partner at $699, an Owners' Suite add-on at a flat $999 per month for the whole firm, and a Free Forever tier at $0 with unlimited matters and seats. The unit is stated and so is its logic, seats being bought by professional role and mixable across one firm, with each user on their own tier.
The term structure is published, annual billing saving ten per cent, and so is the exit, cancellation at any time with access to the end of the billing period. What each tier contains is set out feature by feature, including which tools are free forever and which are Partner-only. The vendor even publishes a worked example, a six-person firm with one Partner, one Associate, two Paralegal and two free seats at about $1,296 per month, which is the kind of arithmetic most vendors leave the buyer to guess at.
Purchase is self-serve, with paid checkout in product and no credit card required to start. On R15: the band's implementation limb does not bite, because there is no implementation charge to disclose on a product that provisions on signup, and white-glove API and MCP onboarding is included at the Partner tier. Verified 12 September 2026.
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.
Coverage is described with real substance and the boundaries are left open, which is B. The segment is named rather than implied: the platform is stated to be built for solo and mid-size litigation firms, and the pricing structure describes the roles it expects to find in one, with seats defined for paralegals and admin staff, associates and of-counsel attorneys, and partners and solo trial attorneys running their own cases.
Jurisdictional reach is claimed specifically for research, caselaw covering every US jurisdiction through Court Listener. Practice coverage is legible from the tool catalogue rather than asserted at large: general civil litigation on both sides through complaints, answers, discovery and motions, a substantial plaintiff personal-injury group in demand letters, medical chronology and causation reports and medical illustrations, insurance coverage analysis, and contract drafting at the edge.
Client-facing reach is described too, intake in English and Spanish and dialect-aware translation. What is left open holds it off A. Nothing states which practice areas the platform is not suitable for, no limit is placed on the jurisdictions whose procedural rules its drafting and formatting tools handle despite court-compliant captions being a named feature, and nothing distinguishes what a paralegal seat may do on a matter from what a partner seat may do beyond the tool list. Verified 12 September 2026.
5 public documents
The public pages on file for AI.Law, with the recorded signals each one supports and the date it was last read. Open any of them and check the reading against the record.
-
ai.law/features3 signals
Ethical Walls and Matter Segregation, Good Law Verification, Outside Counsel Guideline Readiness
Read Sep 12, 2026
-
ai.law/pricing3 signals
Primary Law Corpus Provenance, Billing and Fee Posture, Court Disclosure Support
Read Sep 12, 2026
-
ai.law/terms3 signals
Client Data in Training, Prompt and Output Retention, Third Party Request and Subpoena Notice
Read Sep 12, 2026
-
ai.law/citation-guarantee1 signal
Refusal and Uncertainty Behaviour
Read Sep 12, 2026
-
ai.law/features/ai-case-strategy1 signal
Bar Guidance Alignment
Read Sep 12, 2026
Free Forever at $0; AI seats from $149 per seat per monthUSD, as published, never converted
- AI.Law publishes its prices in full, which is rare in this market, and you can sign up and pay without ever speaking to a salesperson.
- The practice-management side is genuinely free and stays free: unlimited matters, unlimited seats, CRM, calendar, billing, IOLTA trust accounting, document storage and e-signature, with no credit card and no time limit. What you do not get on it is any AI.
- AI is bought per person, by role. A Paralegal seat is $149 a month and is aimed at document work. An Associate seat is $299 and adds research, drafting and discovery. A Partner seat is $699 and adds deposition and trial preparation, injury-case tools and firmwide cross-matter intelligence.
- You mix tiers within one firm and pay only for the AI each person actually uses. The vendor's own example: a six-person firm with one Partner, one Associate, two Paralegals and two free seats runs at about $1,296 a month.
- Annual billing takes ten per cent off, and you can cancel at any time and keep access to the end of the period you have paid for.
Fully published per-seat pricing with figures, tiers, inclusions and terms, all read on the vendor's own pricing page with no sales gate. Three paid seat tiers, priced per seat per month and named for the role each is aimed at: **Paralegal $149**, described as the understand tier for paralegals, admin staff and attorneys whose AI work is document-driven, covering AI document management with auto-routing, complex information extraction, AI timelines and party lists, AI chat with any file or matter, case health monitoring and the Evidence War Room. **Associate $299**, the draft tier for associates, of-counsel attorneys and senior paralegals drafting under supervision, adding AI research, claim and defence identification, drafting of pleadings, motions and discovery, filing analysis and response drafting, and case workflow management, inclusive of everything in Paralegal. **Partner $699**, the win tier for partners, solo trial attorneys and senior associates running their own cases, adding advanced injury-case features, settlement correspondence packages, deposition preparation and multi-deposition transcript analysis, AI-drafted action plans and trial preparation including voir dire, inclusive of everything in Associate, plus Firm Intel at no extra cost, being firmwide cross-matter AI with portfolio analytics, opposing counsel pattern analysis, phone support and white-glove API and MCP onboarding. **Owners' Suite add-on at a flat $999 per month** for the whole firm, cross-matter AI on top of paid Partner seats and expressly not a replacement for per-seat plans. **Free Forever at $0**, unlimited matters and seats, full CRM, calendar, billing, IOLTA trust accounting, document storage, lead pipeline, e-signatures and workflow automation, with no AI tools, no credit card and no time limit.
Structural terms published alongside: tiers are mixable so each user picks their own; annual billing saves ten per cent; the Evidence War Room is included on all three paid tiers; cancellation is available at any time with access through the end of the billing period; and a worked example puts a six-person firm with one Partner, one Associate, two Paralegal and two free seats at about $1,296 per month. Purchase is self-serve with paid checkout in product. No implementation, onboarding, migration or charge is published or referred to anywhere.
Confidentiality and data terms: No Business Associate Agreement is offered or referred to on any surface located, and none of the published material addresses HIPAA. That is worth stating plainly rather than leaving to inference, because the platform ingests medical records as a core function: it produces medical analysis reports covering chronology, causation and costs, generates medical illustrations as trial exhibits, extracts liens, and produces thirty-two-section demand letters built on that material. A plaintiff firm handling personal injury work will be putting protected health information into this system. The Privacy Policy that might address it is published and renders client-side, and this index could not read its body on 12 September 2026; that is a limit on the reader under R20 and not a finding about the vendor, and this field would be revised on a reading.
Note: Every figure, tier name, inclusion and term read directly from the vendor's own pricing page on 12 September 2026, which publishes them outright with no gate. The page carries the seat rates, the add-on, the free tier, the annual discount, the mix-and-match rule, a worked six-person-firm example, the per-tier feature comparison and the cancellation term. entryPriceUsd is recorded as 0 because the Free Forever tier is a genuine zero-cost offering with unlimited matters and seats, no credit card and no time limit, and R10 distinguishes zero, meaning free, from null, meaning no figure published; entryPriceDisplay carries the lowest paid AI rate alongside it so a reader is not misled into thinking the AI is free. **Reported as a judgement call for sampling: recording entryPriceUsd as 149, the entry point for any AI capability, is equally defensible, and the Genie AI record is the closest existing precedent for a floor figure with the free tier described in prose.** Commercial Transparency is graded A on this evidence.
Legal Signals
What each signal meansA signal records what public sources say on the date shown. It is not a grade and it is not a recommendation. Where a signal reads Not addressed, it means the index did not locate the material in public sources on that date, which is a statement about disclosure rather than about the product.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
A published agreement or policy exists and none of it addresses the question either way, or the document that would answer it could not be read and the summary names the retrieval limit. The summary states which shape the silence takes: an improvement right granted that never names training, or no improvement right granted at all.
A customer agreement exists and is published, and the document that would answer this question could not be read, so the row records the retrieval limit rather than a position. The vendor publishes a Terms of Service and a Privacy Policy, ungated, linked in the footer of every page. Both render client-side: a direct fetch returned the page frame, navigation and footer with a loading placeholder in place of the body, and a search of the index recovered only their metadata.
That metadata does establish the character of the agreement, the Terms being for the AI.Law Litigation Operating System and covering subscription, billing, arbitration, class action waiver and governing law. Under R20 this is a limit on this index's reader and not a finding about the vendor. Nothing on any readable surface states a training position either way. The nearest thing is architectural rather than contractual: the vendor states that the AI can only cite what is in the matter's files, which describes grounding rather than a limit on training.
Recorded and expressly not credited because it is not first-party publication: vendor-authored company boilerplate on a third-party business directory states that clients retain full ownership of their data and that it is never used to train AI models. If that sentence appears in the published Terms or Privacy Policy, this row moves to a contractual or policy value on a reading.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
No located public material states how long prompts and outputs are retained.
No located public material states how long prompts, generated documents or uploaded case files are retained. The published Privacy Policy is the document that would ordinarily carry it, and it renders client-side; neither a direct fetch nor a search of the index recovered its body on the date shown, which under R20 is a limit on this index's reader rather than a finding about the vendor, and the row would move on a reading.
What is readable describes storage as a feature rather than a period: document management with upload, organisation and search; a firm-wide Workdrive; discovery compilation indexed by Bates number; and cross-matter retrieval across every matter in the firm. Those tell a buyer that material is held durably and searchably, which is the product working as intended, and say nothing about what happens at the end of a matter or the end of a subscription.
Nothing published addresses deletion on request, export on termination, or whether prompts and generated drafts are retained separately from the underlying files. The surfaces read in full on the date shown were the features catalogue, the pricing page including its FAQ, and the Citation Accuracy Guarantee.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
The product maintains its own permission model, documented, requiring the firm to keep it aligned.
The product maintains its own documented permission model that the firm must administer, which is this value, and the documentation is more specific than most records in this corpus manage. Three published features carry it: Permissions and Roles, offering fifteen built-in roles with matter visibility as a stated dimension; Conflict Checks, described as automatic conflict screening and included on every tier including the free one; and AI Controls, which lets a firm toggle AI categories on or off firm-wide or matter by matter, so the reach of the models is itself a permission the firm sets.
The alignment burden sits with the buyer, which is what separates this value from the tier above: nothing enforces an external document management system's access model, and no default wall configuration is published. One tension belongs in the reading and is recorded rather than resolved. Cross-Matter Intelligence is described as retrieval across every matter, Document Reconnaissance as full-text search across the firm, and Firm Intel as firmwide cross-matter AI including opposing counsel analysis drawn from the firm's case history.
Those are precisely the capabilities a matter-level wall would have to constrain, and nothing published states how the role model interacts with them.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
No located term or policy addresses third party requests for customer data.
No located public material addresses what happens when a third party demands customer data. Nothing on any readable surface refers to subpoenas, warrants, court orders, law enforcement or government requests, and no commitment or reservation about notifying the customer appears. No transparency report exists. The two documents that would ordinarily carry the position, the Terms of Service and the Privacy Policy, are published and ungated but render client-side, and neither a direct fetch nor a search of the index recovered their bodies on the date shown.
Under R20 that is a limit on this index's reader and not a finding about the vendor, and this row would move on a reading. The question has particular weight on this product and is worth stating rather than passing over: the platform holds a firm's complete case files, deposition transcripts, medical records, discovery productions and IOLTA trust accounting records, so it is a single custodian holding material from every matter a firm runs, and whether the firm hears about a demand before its client's file is produced is exactly the kind of term a litigation practice would want settled in advance.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
Sources are identified without stating the licence or rights basis.
The primary law source is named and no licensing basis is stated for it, which is this value, and the naming is more useful than most because the source is externally verifiable. The vendor states that legal citations are verified via Court Listener with clickable links, that its research tool draws on Court Listener caselaw with no invented cites, and that coverage includes every US jurisdiction. Court Listener is the Free Law Project's public database, so a reader can go and see what is in it, which is more than a proprietary corpus claim offers.
The second corpus is the customer's own file, and the vendor is explicit that generated text is confined to it: the AI can only cite what is in the matter's files, and every cited entry links back to a source page or transcript line. What is not stated anywhere is the rights position. No licence, terms of use or attribution arrangement with the Free Law Project is described, no update cadence or data vintage is published, no coverage table sets out which courts and which date ranges are included, and nothing addresses what the ten unnamed language models behind the platform were themselves trained on.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
No located public material addresses whether authority is checked for subsequent history.
No located public material addresses whether authority is checked for subsequent history, and the distinction matters here because the vendor does publish an adjacent check that a reader could mistake for one. What is published is existence and linkage verification: every case citation is verified against Court Listener, every hyperlinked citation is verified against the source document, the research tool is described as producing no invented cites, and a Brief Analyzer offers citation lookup and surfaces counter-authorities.
Those confirm that a cited case is real and reachable, and in the case of counter-authorities that contrary law exists on the point. None of them is a treatment check. Nothing published states that the platform reports whether a decision has been reversed, vacated, overruled, superseded, distinguished or criticised, and no citator relationship, treatment taxonomy or method is described. The gap is worth naming precisely because the vendor has built its positioning on citation safety and publishes a written accuracy guarantee: a citation can be perfectly real, perfectly linked and no longer good law, and the guarantee as published addresses the first two.
Refusal and Uncertainty Behaviour
What does the product do when the answer is not in the corpus?
No located public material addresses what the product does when it cannot ground an answer.
No located public material addresses what the product does when it cannot ground an answer. What is published is a constraint on grounding rather than a description of behaviour at the edge of it: the vendor states that the AI can only cite what is in the matter's files, that its research produces no invented cites, and that every claim carries a clickable citation to a source page. Those describe what the system will not do.
Nothing describes what it does instead when the file does not support the proposition being drafted, which is the question this signal asks. No abstention path, no declining behaviour, no unsupported-assertion marking, no confidence or grounding score exposed to the reviewer, and no evaluation demonstrating any of it was located. The absence sits oddly against the rest of the record, because the vendor has thought harder about hallucination than most in this corpus and backs its citation position with a written guarantee and an account-credit remedy; the missing piece is the case where there is nothing to cite.
Surfaces read in full on the date shown: the features catalogue, the pricing page including its FAQ, and the Citation Accuracy Guarantee.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
No court order, opinion or disciplinary record addressing fabricated or hallucinated legal citations produced by this product has been located as of the date shown. This is a statement about the public record on that one subject, not a finding about the product, and this signal is not a litigation history.
Searched on 12 September 2026, on the product name, the corporate name and the founder's name as a qualifier, against published trackers of decisions on AI-generated fabricated citations including coverage of the Damien Charlotin AI Hallucination Cases database and two independent sanctions trackers, for any court order, opinion or disciplinary record addressing fabricated or hallucinated legal citations produced by this product.
None located. This is a statement about the public record on that one subject as of the date shown, and under R119 this signal records fabricated citations and nothing else, so it is not a litigation history. A note for a future reader, because it makes this particular search harder than most: the vendor's name is close to a generic description of the category, so a bare name search returns the general literature on AI hallucinations in filings rather than anything about this product, and every query needs the corporate or founder qualifier to mean anything.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
Public materials engage with at least one named ethics opinion.
The vendor engages a named ethics opinion in its public materials, which is this value, and it is the only record in pull 8 to do so. The platform is marketed as ABA Opinion 512 Guided, placing the American Bar Association's 2024 formal opinion on generative AI in legal practice on the face of the product alongside its patent number and its SOC 2 claim. The surrounding material is consistent with that framing rather than decorative: the Citation Accuracy Guarantee opens on the fact that courts are sanctioning attorneys for AI-hallucinated citations and offers itself as the answer, and the seat structure names supervision explicitly, the Associate tier being for attorneys and senior paralegals drafting under supervision.
What stops this reaching the top value is that nothing maps the claim onto anything. No page explains what being guided by Opinion 512 means in practice, which of the opinion's duties, competence, confidentiality, communication, supervision and fees, the product addresses, or how. No state bar guidance is named, and no jurisdiction-by-jurisdiction treatment appears. Provenance note: this phrase was read in the search index's rendering of the vendor's own AI Case Strategy page rather than on a page fetched in full, the page bodies being client-side rendered.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
Public materials claim time savings without addressing billing or disclosure, and the product sits inside a fee relationship between a lawyer and a client where those savings would change the bill.
Time savings are claimed and nothing addresses billing or disclosure, on a product sold squarely inside a lawyer-client fee relationship. The savings claims are the commercial argument itself and are stated tier by tier: a Paralegal seat pays for itself by reducing around twenty hours of paralegal work a month, an Associate seat by one drafted motion or one set of discovery responses, a Partner seat by one medical chronology, complaint draft or deposition summary.
The founder has separately put deposition summarisation at up to a twenty per cent reduction in billable time on an insurance defence matter. The buyer is a law firm that bills clients for exactly this work. What makes the silence conspicuous is that the vendor ships the billing system too: time tracking and invoicing, IOLTA trust accounting, AI UTBMS billing described as producing UTBMS codes and polished narratives, and billing validation before invoicing.
So the platform both compresses the hours and writes up the entries, and nothing published states whether an entry identifies work as AI-assisted, whether a client is told, or how a firm should price work that now takes minutes. Recorded under R21 and R24: an AI-written billing narrative is a record of the lawyer's work, not a record of what the AI did, so it does not reach the audit-record value.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
No located public material supports a client side disclosure obligation.
No located public material supports a client-side disclosure obligation. A firm asked by a client's outside counsel guidelines to identify the AI it uses, name the providers that see the client's files, and produce the vendor's data terms could answer none of those from anything published here. No subprocessor list exists on any surface. No model provider is named anywhere: the only published fact about the supply chain is that the AI Associate offers a choice of ten large language models, and the ten are not identified.
No data processing addendum, security exhibit, client-notification pack or forwardable disclosure document was located, and the SOC 2 Type II examination is asserted without a report, a scope statement or a route to request either. The value is not on-request, because nothing indicates such material exists behind a sales conversation: there is no security contact, no trust centre and no request mechanism published. The two documents that might carry some of it, the Terms of Service and the Privacy Policy, are published and render client-side, and could not be read by this index on the date shown; under R20 that limit is recorded rather than held against the vendor, and this row would move on a reading.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
Some elements of the record are available, short of a document level export.
Several elements of a record are available and no document-level export is described, which is this value, and the elements here are stronger than most. Source traceability is built into the output rather than offered as an afterthought: every claim carries a clickable citation to the source document and page, every cited entry links back to a source page or transcript line, case citations are verified against Court Listener, and a brief analyser will check citations and surface counter-authorities in a document already drafted.
The human verification step is written into the published workflow, the Citation Accuracy Guarantee making the user's click-through to the exact source its second step, with an account credit if a link proves wrong. Between them those cover two of the three things a certifying lawyer needs: what the assertion rests on, and that a person checked it. The third is missing and is the reason this is not the top value. Nothing identifies which model produced a passage, and no model is named anywhere in any event; nothing marks machine-drafted text against human-edited text; and no export, certification form, disclosure template or court-facing guidance was located, despite the vendor building its positioning on the risk of judicial sanction.