D
DiliTrust
DiliTrust sells a single integrated suite to in-house legal departments and corporate boards, built around five modules rather than a single application. The Board Portal handles the full meeting cycle from agenda preparation and document distribution through votes to minutes. Contract Management covers the contract lifecycle. Entity Management gives a consolidated view of legal entities, mandates and delegations of authority.
Matter Management tracks disputes and litigation with their associated risk and cost. The Documentation Library, also sold as a Dataroom, provides secure storage with access control and audit capability. Running through all five is Lini, the company's artificial intelligence engine, named for Legal Intelligence and given its own identity in late 2025, under which every intelligent feature in the suite now sits. Lini ships seven distinct capabilities: Ask Lini, a natural-language assistant that summarises, searches, translates and extracts across documents; Automated Data Extraction, which uses extractive AI and optical character recognition to pull terms, dates, obligations and clauses from contracts into structured summary sheets; Document Summarization across financial, regulatory and contractual material; Minutes Generation, which drafts board minutes from the agenda and related documents; Audio Transcription, which converts meeting recordings into text feeding those minutes; Risk Detector, which flags risky clauses, applies the firm's own compliance rules and proposes validated alternatives; and QuickView, an AI-generated snapshot of a matter's status.
The company positions Lini as sovereign, stating that its models are designed and trained exclusively in-house on synthetic data, publicly available datasets and its own knowledge base, and that processing happens in data centres located in the regions where its clients operate. It also allows organisations to connect their own large language models through its API. DiliTrust SAS is headquartered at Paris La Defense in France, with offices in the United States, Canada, Dubai, Spain, Italy, Germany, Mexico, Colombia and Peru, and it acquired the Delaware enterprise legal management company doeLEGAL in September 2024.
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 real capabilities layered on a suite that would function without them, which is the B band, and the gap between the vendor's framing and the product decides it. The framing is maximal: DiliTrust describes itself as the only AI-native, fully integrated platform, and every intelligent feature across the suite now sits under the Lini umbrella. The substance behind that is genuine and unusually broad for this lane: seven named capabilities shipping across all five modules, covering natural-language assistance, extractive data capture with optical character recognition, document summarisation, board minute drafting, audio transcription, contract risk detection with playbook application, and matter snapshots.
What the record cannot support is A. The five modules are a board portal, a contract system, an entity register, a matter tracker and a dataroom, and each is a system of record whose core job is holding and organising corporate legal information. Strip Lini out and a firm still has its board pack, its entity structure, its contract repository and its litigation tracker; what it loses is speed. The vendor's own account confirms the layering rather than contradicting it, describing Lini as powering every dimension of legal work rather than constituting the product, and dating its emergence as a named engine to late 2025 against a suite that long predates it. 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.
Reliability is asserted without measurement and no grounding method is described, which is the C band, and R15 governs the weight. This product cites no legal authority. Lini reads the customer's own contracts, board packs and matter files, so the limbs on primary sources, openable citations and citator status do not bite and the record is not penalised for them. What does bite is that several outputs are consequential and the vendor publishes confidence rather than evidence.
Risk Detector is described as identifying risky clauses, applying the firm's internal compliance rules and suggesting compliant and reliable edit suggestions; Minutes Generation drafts board minutes, which are a corporate record with legal effect; Document Summarization is offered across regulatory and financial material. For none of these is an accuracy figure, error rate, test set or evaluation published, and no hallucination disclosure of any kind was located on the surfaces read.
Two things sit adjacent and are recorded rather than credited. The extraction feature is described as extractive, which is a real architectural constraint tending against fabrication because the output is pulled from the source document rather than generated, but the vendor never makes that argument or evidences it. And the AI Code of Conduct commits to transparency in the sense of explaining how decisions are made, which is a governance statement graded on that row rather than an accuracy one. 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.
A written commitment that the models support rather than replace human decision-making, published in a governance instrument, short of the full control structure. The commitment is explicit and sits under an Ethical AI heading in the published AI Code of Conduct: AI at DiliTrust is used to enhance human decision-making, not replace it, and the vendor states a firm stance that critical decisions always involve human oversight.
That is a position a buyer can hold the vendor to in writing, and it is repeated in the product framing, which describes Lini as existing not to replace people but to empower them. The feature descriptions are consistent rather than contradicting it, each stopping at proposing: Risk Detector suggests validated alternatives, QuickView produces a snapshot, Minutes Generation produces a draft, and the extraction features fill summary sheets a user reviews.
Real review surfaces exist in the platform and are documented on the security page: a comprehensive audit trail tracking user interactions, data changes and system events in real time, and granular access control defining who may do what. R124(2) sets the ceiling and it is applied here. Nothing attaches a boundary to a named mode or tier and states what that tier's output may not be used for, which is what a categorical constraint must do to substitute for a numeric threshold.
No confidence signal, no stopping condition and no error-handling path is published, and the phrase critical decisions is left undefined, so which decisions the vendor considers critical is the customer's judgement rather than a published boundary. 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 in quantity and at scale, with no figure attached to any of them, which is the B band. The naming is a logo wall on the vendor's own security page rather than a set of case studies, and it is substantial: Accor, Atos, BNP Paribas, LVMH, Vivendi, Renault, Sodexo, Carrefour, Danone, Veolia, Lavazza, Illy, Geox, Nexi, Webuild, Mahou San Miguel, DIA, Invex and STM. Those are identifiable enterprises across banking, luxury, energy, food, construction and transport, and the concentration of large French and Italian groups is consistent with the vendor's European positioning.
Scale is claimed alongside, the company stating support for more than 2,500 businesses across over 60 countries in its own acquisition announcement. What is missing is measurement tied to any of it. No named customer carries a figure, nothing is dated to a deployment, and no result is published with a basis. The vendor's outcome material takes a different and weaker form, five return-on-investment calculators covering the suite and each module, which invite a prospect to generate their own projection rather than reporting what any customer achieved; a projection a buyer produces about itself is not deployment evidence and is not credited.
A customer stories library is published and was not opened; under R25 it corroborates rather than carries a grade already resting on the named-customer wall, and it is what would move this row if the stories carry dated figures with a stated basis. 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.
Substantive published commitments on confidentiality and training use, failing the limb R33 makes decisive. What is published is stronger than most records at this grade and rests on a specific architectural claim rather than an adjective. The vendor states a zero-access principle in terms: once a customer entrusts it with data, that data is exclusively the customer's and the vendor's own team does not have access to it.
Around that sit encryption of data at rest and in transit, a complex data separation solution, granular access control and permissions, document watermarking for traceability, a comprehensive audit trail, mandatory and continuing security training for all employees, and a named Data Protection Officer function reachable by customers. On training use the position is clear enough to grade, the AI Code of Conduct enumerating the training corpus as synthetic data, publicly available datasets and the vendor's own knowledge base, and stating that personal data is processed without any third-party sharing.
The limb that fails is privilege and work product, and it fails completely. Neither privilege, professional secrecy, nor the confidentiality duties of in-house counsel is addressed anywhere on the surfaces read, on a platform whose Matter Management module holds live disputes and whose Board Portal holds board deliberations. Two further gaps are recorded: no retention period and no deletion commitment for customer data was located, because the only published privacy instrument governs website visitors rather than the product. 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.
Nothing published on professional responsibility was located, and R15 requires naming which limbs bite before the grade is read as heavier than it is. The advice-line limb applies only lightly. The buyer is a corporate legal department or a board secretariat, not a firm selling advice, and the outputs are internal governance artefacts rather than advice to a client. The audience is unambiguous, addressed through five named departmental pages covering board members, general secretaries, legal operations, contract managers and corporate lawyers, so nobody could mistake who this is for.
What is entirely absent is any statement connecting the AI's output to the professional obligations of the lawyers relying on it. There is no statement that output is not legal advice, no disclaimer of any kind on any surface read, and no published agreement in which such a position might otherwise sit. Two places where it would bite are named rather than passed over. Minutes Generation drafts the minutes of a board meeting, which are a corporate record with evidential and statutory consequences and which a company secretary signs.
Risk Detector proposes alternative contract clauses and applies compliance rules, which is drafting judgement being exercised by a model on documents a lawyer will approve. Nothing published addresses the supervision either requires, and the vendor's only adjacent statement, that critical decisions always involve human oversight, is an autonomy commitment graded elsewhere and does not identify what the professional obligation is. 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.
A published governance framework with real substance, short of testing results and a named accountable owner, which is the B band, and this is among the more complete frameworks in the corpus. The instrument is a standalone AI Code of Conduct, dated, downloadable as a PDF the vendor expressly invites customers to circulate, and organised into seven numbered sections. Its content is specific rather than aspirational. It sets out how the models are developed and what they are trained on.
It states the security standards they operate under and the jurisdictional position of the data centres. It commits to GDPR alignment and privacy by design. Section four does something no other record in this corpus does: it walks through the runtime data path step by step, from user request, through encryption before data leaves the device, transfer to a local data centre, decryption with a unique key, processing, insight extraction and return to the interface.
Section six carries three ethical commitments including a bias commitment that claims a mechanism rather than an intention, the vendor stating that it rigorously tests and monitors its models to identify and address potential biases. Alignment with the European AI Act is claimed. What A asks for is still missing. No testing result, evaluation or finding is published, so the bias claim is a described practice with no output.
Nobody is identified as accountable for AI: the Data Protection Officer covers data protection and the framework is endorsed by the chief executive, but neither is an AI accountability designation. 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.
Substantive published policy covering most of the ground, missing the data lifecycle, which is the B band. Access and protection are documented in detail across two surfaces. The platform carries two-factor authentication and single sign-on, encryption of static and transiting data, granular role and permission control, document watermarking with a unique stamp per document for traceability, secure document sharing with tracked activity, and a comprehensive audit trail capturing user interactions and system events in real time for incident detection.
Organisationally the vendor publishes a Data Protection Officer function, mandatory and continuing security training, continuous monitoring and updating of controls, and physically secured server sites with video surveillance and permanent on-site staff. The AI-specific handling is described end to end in the AI Code of Conduct, including encryption before data leaves the user's device and decryption at the destination with a unique key.
The zero-access principle is asserted at the strongest level, the vendor stating its own team cannot reach customer information. What is absent is what happens to data over time. No retention period is published for customer content, prompts or outputs, no deletion or return commitment on termination was located, no subprocessor register exists for the platform, and no incident notification commitment to customers was found.
The cause is structural and is recorded rather than treated as silence: the only published privacy instrument is scoped to website visitors and the contact form, and no customer agreement is published at all. 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.
Nothing published on who bears the loss when the system is wrong was located, which is the D band. No warranty, indemnity, liability cap, exclusion, service level, service credit or insurance position appears anywhere on the surfaces read. The cause is that no customer agreement is published. The complete legal inventory in this vendor's footer is a privacy policy and a legal notice, the latter being the statutory publisher identification a French site must carry rather than an agreement; there is no master subscription agreement, no terms of service, no data processing addendum and no order form template.
An Information System Security Policy is named on the security page and is expressly gated, the vendor asking prospective readers to contact it to access what it calls a confidential document, so even the security position behind the certifications is not readable in advance. This is an enterprise vendor selling to companies of the scale of BNP Paribas and LVMH, so negotiated contracts certainly exist; the point the axis measures is that a buyer cannot read any of it before entering a sales process.
The exposure is concrete rather than theoretical on this product, and it is worth naming: the AI drafts board minutes, proposes contract clause alternatives, applies compliance rules and summarises regulatory documents, and every one of those errors lands in a corporate record. 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 individually and organised by class, with the target module and the benefit stated for each, short of the configuration detail an implementer needs. The catalogue is published on a dedicated page and split into three declared categories. Native integrations, described as plug and play and fully embedded, cover Salesforce, HubSpot and Microsoft Dynamics 365 into contract management, with contract generation from opportunities and deals and automated approvals; a Microsoft Word add-in and a Google Docs add-on for drafting against centralised templates and clause libraries; and Microsoft Teams, an Outlook add-in and a Gmail add-in feeding the Board Portal and Matter Management with agendas, meeting invitations, emails and attachments.
Partner integrations cover governance and registry work, naming Impal'Act for entity structure and governance events, PF Registres for registry filings, and eWitness for certified timestamping and witness verification. Third-party integrations cover electronic signature at unusual depth, naming DocuSign, Adobe Sign, SignaturIt, Dropbox Sign, BoxSign, Yousign, UniverSign and Connective, and distinguishing between global providers and eIDAS-compliant qualified signature providers for European work.
Each entry states which DiliTrust module it connects to and which roles it serves, and direction of flow is described for several. What holds it off A is configuration: no public API reference or developer documentation was located, the page closing on a general reference to plug-and-play and API integrations, and nothing states what a firm must set up or what objects synchronise. 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.
Residency is addressed as a positioning commitment rather than a passing mention, and stops short of the specificity A requires. What is published: server locations across Europe, North America, Africa and the Middle East, physically secured with video surveillance and permanent on-site staff; a commitment in the AI Code of Conduct that data centres are located locally in the regions where clients operate, framed as guaranteeing compliance with data sovereignty laws and protecting against unauthorised cross-border transfers; a statement that the AI operates entirely within a secure encrypted environment hosted on local infrastructure; and, unusually, an express jurisdictional claim that the solutions are not subject to the United States CLOUD Act, qualified as except for US clients.
The chief executive puts the same point in the buyer's own language, that data is stored in the client's country and under the same jurisdiction, eliminating the risk of interference from foreign legislation. For a European buyer weighing sovereignty that is a real disclosure and it is the strongest part of this record after the certifications. What holds it off A is that the detail stops at continents and a principle.
No specific data centre location or country list is published, so a buyer cannot confirm where its own tenant would sit without asking. No tenancy model is described: complex data separation is asserted in the AI Code of Conduct and separation controls appear on the security page, but nothing states whether the deployment is multi-tenant or isolated, and no single-tenant or on-premises option is offered or refused. 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.
Certifications are current and the evidence is reachable without a sales conversation, which is the A band, and this is the first A on this axis in pull 8 after nineteen records without one. The distinguishing fact is that the certificates themselves are published rather than described. A dedicated security documentation page links the ISO 27001 certificate and the ISO 27701 certificate as downloadable documents, both dated 23 February 2026, and a Spanish ENS certificate dated 25 September 2024 under the Esquema Nacional de Seguridad.
Alongside them sit two further ungated documents, a consolidated Security Sheet setting out the implemented measures and the AI Code of Conduct, the latter published as a PDF the vendor invites readers to circulate. SOC 2 Type 2 compliance is claimed and its report is offered on request. So a buyer can establish, before speaking to anyone, which standards are held, which entity holds them, when they were issued and where the AI-specific commitments sit.
The page also sets out the control environment behind them, covering the Data Protection Officer function, physical server security, the zero-access principle, security training, transparent communication and continuous monitoring. Three limits are recorded and none displaces the grade. The certificate PDFs were not opened, so this grade rests on the published access flow and the standards and dates the page states rather than on the certificates' contents.
No auditor is named for the SOC 2. And the Information System Security Policy is expressly gated, the vendor describing it as confidential and directing readers to contact it. 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 supply chain is partly disclosed and the shape is unusual, because the vendor's answer to who supplies the models is itself. The AI Code of Conduct states that the models are designed and trained exclusively by the vendor's own teams, giving it full ownership and control, and the Lini material draws the contrast explicitly, saying that unlike others relying on open solutions such as ChatGPT the AI is built entirely in-house.
Where inference runs is answered: entirely within a secure encrypted environment on the vendor's local infrastructure, in data centres in the regions where clients operate. Customer control over provider choice is published as a feature, the interoperability section stating that organisations can connect their own large language models through secure API connectivity while using the vendor's infrastructure. That is more than most records disclose.
What holds it off A is that no model is named and the in-house claim is qualified by the vendor's own words. The same document states that the vendor pre-finetunes and finetunes its legal-specific language model using the most advanced open-source and commercially permissible models available, which means the foundation is third-party open-source work rather than built from nothing, and neither those base models nor the resulting legal model is identified by name or version.
R34 requires the models to be named as a limb separate from identifying their providers, and that limb fails. No change notification commitment was located, the vendor stating only that its systems are continuously updated. 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.
No pricing information is published at any level, including the unit of charge, which is the D band, and this is a page-inventory finding rather than a retrieval limit. The full navigation was read and there is no pricing page: the menu runs Home, Product, DiliTrust's AI, ROI, Solutions, Resources, Support and Request a demo, and the single conversion path across every page is a demo request or a conversation with the sales team.
Nothing states whether the suite is licensed per user, per module, per entity or per company, no minimum or band appears, and none of the five modules carries a figure. Under R10's closing discipline an estate that only invites a sales conversation is an absence belonging in this note alone, so no VendorPricing row is written for this record. What sits where pricing would be is worth recording because it is a deliberate substitution rather than an oversight: the vendor publishes five return-on-investment calculators, one for the full suite and one for each of contract lifecycle management, board management, matter management and entity management, inviting a prospect to model its own savings.
A buyer can therefore generate a projection of what the product might save without being able to learn what it costs, which is an unusual asymmetry and one the note states plainly. No agreement is published either, so the commercial terms are equally unreadable. 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 documented with real substance across buyers, roles and sectors, short of the stated limits the A band asks for. Who this is for is addressed on surfaces of its own rather than asserted: five departmental pages cover board members, general secretaries, legal operations, contract managers and corporate lawyers, so each buyer inside the organisation has a page written to it. The segment is unambiguous throughout, in-house legal departments and corporate boards, and the vendor is explicit that it exists to serve the general counsel's function rather than law firms.
Sector coverage is enumerated with four named industry pages for pharmaceuticals, private equity, real estate and energy, and a compliance solution area with a dedicated page for the European DORA regime, which is a specific regulatory workload rather than a generic claim. Geographic reach is evidenced rather than asserted, with contact estates in ten countries and the site published in English, French, Spanish, Italian, German and Canadian French.
Functional coverage maps to the five modules and is stated for each. R15 applies to the practice-area limb: this is corporate governance and legal operations infrastructure that does not vary by practice area in the way a research or litigation product does, and the record is neither credited nor penalised for it. What holds it off A is that limits are absent. Nothing identifies an organisation size floor, no segment is named as out of scope, government use is neither claimed nor excluded, and nothing states which modules are available in which of the ten territories. Verified 12 September 2026.
5 public documents
The public pages on file for DiliTrust, 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.
-
dilitrust.com/ai-code-of-conduct6 signals
Client Data in Training, Third Party Request and Subpoena Notice, Primary Law Corpus Provenance and 3 more
Read Sep 12, 2026
-
dilitrust.com/security-documentation2 signals
Ethical Walls and Matter Segregation, Court Disclosure Support
Read Sep 12, 2026
-
Billing and Fee Posture
Read Sep 12, 2026
-
dilitrust.com/dilitrusts-ai-lini1 signal
Good Law Verification
Read Sep 12, 2026
-
dilitrust.com/privacy-policy1 signal
Prompt and Output Retention
Read Sep 12, 2026
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 public policy or trust page states no training on customer content, with no matching term located in the published agreement.
A public policy states the position on training and no agreement exists in which to look for a matching term, which is this value, though the statement is constructive rather than express and the note says so rather than overstating it. The AI Code of Conduct enumerates the training corpus exhaustively under a Strict Data Policy heading: by leveraging synthetic data, publicly available datasets and its own sovereign knowledge base, the vendor has developed an internally-built private dataset, a synthetic legal dataset tailored to legal tasks.
Customer content appears nowhere in that enumeration. Reinforcing it, the security page states a zero-access principle in terms, that once a customer entrusts data to the vendor it is exclusively the customer's and the vendor's own team does not have access to it, and the code states that personal data is processed without any third-party sharing. Read together a buyer would reasonably conclude that customer content is not training material.
What the vendor never does is say so in those words, and that is why this is recorded as constructive: there is no sentence reading that customer data is not used to train the models, only a complete account of what is used instead. R43(1) was run and is why a contractual value is unavailable rather than declined: the estate publishes a privacy policy and a French legal notice and no customer agreement of any kind, so there is no instrument in which a training term could sit.
One counterweight is recorded: the privacy policy states that client employee or user data may be used to improve the performance and quality of the Service, which on the R28 naming test is not a training right because it names neither training nor machine learning.
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, outputs or AI working records are kept, and this is an established absence with a documented cause rather than an untested one. The only published instrument that could carry a period is the privacy policy, last updated 27 January 2025, and it is scoped by its own opening words to data collected when a visitor uses the website and submits the contact form. Its retention clause is correspondingly generic, committing only to retain personal data for no longer than necessary for the purposes collected and to erase it within a reasonable timeframe on request, and directing anyone wanting detail to email the Data Protection Officer.
No customer agreement and no data processing addendum is published, so nothing governs the product's data. The AI Code of Conduct describes the runtime path in unusual detail, from encryption before data leaves the device through processing to delivery of results, and stops at delivery: it says nothing about what persists afterwards, how long a prompt or a generated summary is held, or whether AI working records are retained separately from the documents they were derived from.
The gap is worth stating concretely because of what this AI touches: audio recordings of board meetings, generated minutes, contract risk assessments and matter snapshots are all outputs a company would want a retention position on, and none is published.
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 permission model and documents it feature by feature, which is this value. The security documentation sets out granular access control and permissions as a named capability, describing the ability to define roles, permissions and access levels for each user so that only authorised individuals reach specific data and functionality. Around it sit several mechanisms that serve the same purpose: two-factor authentication and single sign-on for identity, a comprehensive audit trail tracking user interactions and data changes in real time so that access can be reconstructed, secure document sharing with per-document access permissions and tracked activity, and document watermarking that stamps each document uniquely to deter and trace unauthorised distribution.
The AI Code of Conduct adds a complex data separation solution at the infrastructure level. Taken together that is a documented model the customer administers, which is what separates this value from a bare claim. Two limits are recorded. Nothing published describes a conflicts or ethical wall function by name, or matter-level walls of the kind an in-house team running contentious matters against the same counterparties would ask about, and the Matter Management module holds exactly that material.
And nothing states whether Lini's retrieval respects the asking user's permissions: the AI is described as operating across the suite, and no published statement confirms that an assistant answering a question cannot surface a document the user could not otherwise open.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
Published terms or policy address disclosure to authorities or in response to legal process, and no commitment or reservation regarding customer notice is located anywhere. The vendor has told the customer that data can leave and has said nothing about whether the customer hears of it.
Compelled disclosure is addressed substantively and customer notice is absent everywhere, which is this value, and this record is an unusually clear instance because the vendor volunteers the subject rather than being silent on it. The AI Code of Conduct claims exemption from the United States CLOUD Act, qualified as except for US clients, and frames it as safeguarding client data from foreign government access; the same document presents locally sited data centres as protecting against unauthorised cross-border transfers; and the chief executive states on the security page that data stored in the client's own country and jurisdiction eliminates the risk of interference with foreign legislation such as the CLOUD Act.
That is a vendor making resistance to government access a selling point. The privacy policy addresses the other side of the same question for the data it governs, stating that the vendor will refuse government and law enforcement requests it believes unfounded, too broad or unrelated to their stated purpose, while reserving the right to cooperate where it believes disclosure necessary and appropriate to comply with legal process.
What appears nowhere, in either place, is any commitment or reservation about telling the customer. Nothing says the customer would be notified of a demand, given an opportunity to object or seek a protective order, or informed afterwards. No transparency report was located. The asymmetry is the finding: extensive published assurance about which governments cannot reach the data, and silence on what happens when one lawfully can.
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 material behind the models is described by category and no individual source is named, which is this value, and the shape needs stating because it is not the usual one. This product answers from the customer's own documents rather than from a legal corpus, so the signal's usual object, a body of primary law with a licensing position, does not exist here. What the vendor does publish, and what is graded, is the provenance of its training material, and it is more than most disclose.
The AI Code of Conduct states that the models are trained on synthetic data, publicly available datasets and the vendor's own sovereign knowledge base, combined into an internally-built private synthetic legal dataset covering multiple legal use cases. It goes further on the foundation, stating that the vendor pre-finetunes and finetunes its legal-specific language model using the most advanced open-source and commercially permissible models available.
That phrase is a licensing posture and is recorded as one: it says the base models were licensed for commercial use. What is not published is any individual source. No dataset, no publisher, no jurisdiction and no code or corpus is named, the sovereign knowledge base is not described, and no licence is stated for the publicly available datasets, which is the category where a provenance question would ordinarily bite hardest.
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 on this product the question does not arise in its usual form. Nothing in the suite cites law. The seven Lini capabilities work on the customer's own material: transcribing board audio, drafting minutes from an agenda and its papers, extracting terms and dates from contracts, summarising documents the customer uploaded, detecting risky clauses against the customer's own compliance rules, and snapshotting a matter's status.
None produces a proposition about the state of the law whose treatment a lawyer would verify, so R15 governs and the limb is recorded as inapplicable rather than failed. One adjacency is named so it is not mistaken for the thing: Risk Detector applies the firm's playbooks and suggests compliant alternatives, which is currency against an internal policy rather than currency of a legal authority, and it is graded on the citation accuracy row.
A second is worth recording because it is the closest real analogue on this estate. The Entity Management module synchronises with official registries through a named partner integration and the vendor presents that as reducing the risk of outdated or inconsistent records, which is a currency mechanism for corporate registry data rather than for case law. The surfaces read on the date shown were the Lini feature page, the AI Code of Conduct, the security documentation and the integrations page.
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 describes what the system does when it cannot produce a reliable answer. The surfaces where it would appear were read in full on the date shown: the Lini feature page with all seven capability descriptions, the AI Code of Conduct including its section on how the AI works, the security documentation and the integrations page. None states that any feature declines a request, flags low confidence, surfaces uncertainty in a draft, or puts an ambiguous input back to the user.
The published account of the runtime path is unusually detailed and runs from user request to results delivered without a branch: there is no described case in which the system returns nothing or returns a caveat. Two things sit nearby and neither is credited. The code commits to transparency in the sense of making AI processes understandable and providing clear documentation and insight into how decisions are made, which is an explainability commitment about the system in general rather than a behaviour at the point of answering.
And the commitment that critical decisions always involve human oversight places the uncertainty burden on the user rather than describing a system behaviour, and is graded on the autonomy row. The gap has a specific edge on this product: Minutes Generation drafts the record of what a board decided, and nothing published indicates whether the system signals where the source material was thin or the discussion ambiguous.
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 company name and on the AI engine name, against published trackers and coverage of decisions on AI-generated fabricated citations, including coverage of the Damien Charlotin AI Hallucination Cases database and reporting on the 2025 and 2026 sanctions decisions across United States federal and state courts and other jurisdictions. None located. Under R119 this signal records fabricated citations and nothing else, so it is not a litigation history and no other proceeding involving the vendor would appear here.
One point of context is recorded because it bears on how the absence should be read rather than on the vendor's conduct: the exposure this signal tracks arises where a product generates legal authority for filing, and nothing in this suite does. Its outputs are board minutes, contract extractions, clause suggestions, document summaries and matter snapshots, all drawn from the customer's own material. The analogous failure here would be a fabricated term in an extraction or a misstated resolution in generated minutes, which no tracker records and which would surface, if at all, as a corporate governance dispute rather than as a sanctions order.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
No located public material engages with bar or ethics guidance.
No located public material engages with bar or ethics guidance, in general terms or otherwise. No bar opinion is named anywhere on the estate, no professional conduct rule of any jurisdiction is cited, and nothing maps a product or an AI feature to the obligations of the lawyers using it. Nor is professional responsibility engaged generically: there is no requirement that the customer use the platform consistently with its professional obligations, which is the clause that would ordinarily sit in a customer agreement, and no customer agreement is published.
The absence is worth distinguishing from a lack of regulatory engagement, because the vendor engages regulation heavily and that makes the gap sharper rather than softer. The AI Code of Conduct claims alignment with the European AI Act, commits to GDPR compliance and privacy by design, and the estate carries a dedicated page on the DORA regime. Those are regulatory instruments binding on the customer as an enterprise.
What is missing is the separate body of rules binding on the customer as a lawyer. The point bites on a specific output: minutes of a board meeting drafted by a model and signed by a company secretary sit inside the corporate secretarial duties that professional bodies in several of this vendor's ten territories regulate, and nothing published addresses that.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
The product does not touch a fee between a lawyer and a client. It operates before an engagement exists, or it is bought by a team that bills no client for the work. Savings claims aimed at the buyer’s own cost are recorded in the summary and do not make the row a savings claim, because no client bill is in the loop.
The product sits outside a lawyer-to-client fee relationship, which is this value, and the reasoning is worth setting out because the vendor makes savings claims that could be mistaken for the thing this signal measures. The buyer here is a corporate legal department or a board secretariat, and the lawyers using the product are employees of the company they advise. There is no bill to a client, so the question this signal asks, what happens to the bill when the work takes an hour instead of six, has no addressee: time the AI saves accrues to the company's own operating cost, not to a fee a client pays.
The savings claims are extensive and are recorded rather than credited against the wrong question, taking the form of five published return-on-investment calculators, one for the suite and one for each of contract lifecycle management, board management, matter management and entity management, which invite a prospect to model its own time and money saved. Two boundary facts are recorded. The Matter Management module tracks the cost of disputes, which is outside counsel spend viewed from the payer's side, and nothing published connects it to AI-assisted work or to disclosure.
And the vendor acquired an enterprise legal management business in 2024, a product class that sits closer to the fee relationship; nothing located indicates that capability has reached this suite, and if it does the value should be revisited.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
A subprocessor and model provider list plus client facing disclosure material is published or available without an agreement in place.
A forwardable pack exists, is ungated, and answers the model-provider question, which is this value, though it is reached by an unusual route and the gap in it is named. The client-facing artifact is the AI Code of Conduct, published as a PDF the vendor expressly invites readers to download and share with colleagues, sitting alongside an ungated Security Sheet consolidating the implemented measures and downloadable ISO 27001 and ISO 27701 certificates.
That is material drafted to be handed on, which is what R29 means by a consent or notification pack. The model-provider question is answered emphatically rather than by a list: the vendor states that its models are designed and trained exclusively by its own teams, that processing happens on its local infrastructure, and that personal data is processed without any third-party sharing, so a firm asked which third party sees its content can answer that on the vendor's account there is none.
The interoperability section adds that any third-party model in play would be one the customer itself connected. The gap is the platform-level register. No sub-processor list is published for hosting, monitoring, storage or support, so while the AI supply chain is accounted for, the infrastructure around it is not, and a client demanding a full sub-processor schedule would have to ask. Recorded as the artifact that would complete the pack.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
No located public material addresses court disclosure or verification certification.
No located public material addresses disclosure of AI involvement, and the note records two published features that come close without being it. The first is the audit trail, described as maintaining visibility and traceability of user activities and system events, tracking interactions and data changes in real time to detect and investigate incidents, maintain compliance and ensure data integrity. It is a genuine record of who did what and when, and it is not credited here because nothing states that it identifies which passages a model produced, distinguishes machine-drafted from human-edited text, or survives export in a form anyone outside the organisation could rely on.
The second is watermarking, which stamps each document uniquely for traceability against unauthorised sharing; that is a provenance mechanism aimed at leakage rather than at authorship. Nothing published provides a disclosure template, a certification, a model identifier attached to output, or an export designed for a tribunal or a regulator. The gap has a concrete edge on this product and it is not the courtroom. Minutes generated by a model become the formal record of a board's decisions, relied on by auditors, regulators and, in a dispute, by a court; if the question later arose whether a passage recording a resolution was drafted by a person or a model, nothing published would let the company establish which.