Clearbrief vs Litem: how they compare in 2026

The models are the product, and the architecture choice is itself the differentiator. Citation detection across a document, the patented semantic scoring that measures how well a sentence is supported by the page it cites, and concept search across uploaded discovery are all machine learning and natural language processing. The vendor deliberately does not use generative models for the checking function, stating that classic AI means the checking tool cannot itself introduce hallucinations, while shipping separate generative features for drafting. Remove the models and nothing remains but a Word toolbar. Worth recording plainly since the index grades AI centrality: not being generative is not the same as not being AI, and the distinction is the point of this product.

The models are the product and there is no conventional substrate beneath them. The company describes itself as providing AI workflows and tools across the litigation lifecycle, starting in discovery, assisting through depositions in real time, and offering agentic features for further litigation phases, and every named product line is an inference function rather than a storage or workflow layer: analysis of transcripts and testimony, a facts capability over the case record, and an agent covering later phases. Real-time assistance during a live deposition is not something a database does. The comparator is the manual work of reading transcripts end to end, which is what the product replaces rather than accelerates, and the customer-reported outcome banked on this record is preparation time halved. Remove the models and nothing remains that a litigator would license. The agreement is consistent with that reading, defining third-party large language model providers as a term and disclaiming warranties specifically as to AI-generated content, which is a company contracting around model output rather than around software. Verified 2 September 2026.

Grounding is real, documented and mechanised, short of published measurement of the checker itself. The retrieval method is described concretely: citations are detected in the Word document, matched against caselaw databases through LexisNexis where the user holds a subscription and Fastcase and vLex where they do not, with proprietary Westlaw citations mapped across to those databases, and where a source cannot be displayed automatically the product still reports that the citation is verifiable and accepts a manual upload. Output grounds to sources the reader opens directly, with hyperlinks from every citation to the specific page and text. The failure mode is named rather than implied, and named unusually well: the vendor states that because the checking runs on classic AI rather than generative models there is no risk of the checker introducing hallucinations of its own, which is a disclosure about its own architecture that most vendors never make. What keeps this off an A is measurement of the product's own performance: searched the vendor site, the blog, the sandbox site and trade coverage on 29 Aug 2026 and located no published false positive or false negative rate for citation detection, no accuracy figure for the semantic score, no test set and no independent evaluation.

Accuracy is addressed candidly and only as a disclaimer, with no measurement anywhere. Clause 9.6 of the master services agreement is explicit rather than buried: no warranties are given as to the accuracy, completeness or reliability of AI-generated content, the customer assumes sole responsibility for it, and the customer must verify output before use. Saying so in the contract, in those words, is more honest than most records here manage and it is why this is not lower. What is absent is everything on the other side. No accuracy figure, benchmark or evaluation is published for any feature, and nothing describes how a summary or extracted fact is grounded in the underlying transcript, which matters unusually here because deposition testimony has a natural citation form in page and line numbers and the product's own blog writes about citing transcripts correctly. Nothing addresses hallucination by name. The five solution pages were not opened on this pass and may describe grounding; that is recorded as a limit on this reading rather than as an absence. Verified 2 September 2026.

A real published commitment with concrete review surfaces, and the product exists to serve oversight rather than to reduce it. The vendor states that source linked outputs keep humans in control and the evidence in view, and the Cite Check Report is explicitly built for the partner signing a filing, listing every citation with hyperlinks so each flagged issue can be reviewed in context. The founder frames the problem as partners being sanctioned for citation errors they did not personally make, which is a supervision argument. Not located as of 29 Aug 2026: the threshold at which a semantic score is treated as low, whether that threshold is configurable, and what the vendor commits to when its own checker misses something. That last matters here more than for most, because a verification tool that fails silently leaves a lawyer more confident than before they used it.

Oversight is allocated in the agreement rather than described in the product, and the allocation is unusually complete. Clause 9.5 places vetting, output review and privilege preservation on the customer, which names three distinct obligations rather than gesturing at supervision. Clause 9.6 requires verification of AI-generated content before use. Clause 3.3 goes further than either by restricting who may operate the system at all, requiring attorney users to be licensed and in good standing in the United States and providing that paralegal use must not constitute the practice of law, so the identity and competence of the reviewer is itself contractual. Taken together a firm knows who must check the output and what they are responsible for. What is missing is the product side. Nothing published states what runs unattended, where a review checkpoint sits in the interface, what confidence or uncertainty is surfaced, or how the agentic features described for later litigation phases differ in autonomy from the analysis features. Verified 2 September 2026.

User categories and testimonials stand in for deployment evidence. The vendor states the product is used by courts, Am Law firms, government agencies, arbitrators and legal teams handling high stakes matters, which is a notable set because courts and arbitrators are adopters rather than customers in the ordinary sense, and an attributed testimonial describes the LexisNexis backed hallucination detection in practice. Searched the vendor site, the blog, the sandbox site and trade coverage on 29 Aug 2026 and located no named organisation paired with figures and a date, no case study, and no assessable method. Categories of adopter are not the same as a named deployment.

Named firms at the top of the defence bar, with volume figures and one attributed outcome. Two customers are named on the vendor's own surfaces, Ogletree Deakins and Rutan & Tucker, both substantial defence firms whose adoption is meaningful evidence in a market where litigation data is closely held. Deployment scale is published as more than fifty Am Law 200 firms, more than 10,000 depositions handled monthly and more than 20,000 professionals using the platform. Three individuals speak on the record, including Kyre Stucklin, Director of Innovation at Rutan & Tucker, and a litigator reporting that preparation time was cut by half, which attaches a figure to a named person rather than leaving it floating. What holds this below the top band is method and dating: none of the volume figures carries a measurement basis or a period, and no case study sets out a before-and-after. One contractual fact belongs alongside any logo strip on this record: clause 12.1 permits Deposely to use customer names and logos, so a logo evidences a contract rather than an endorsement. Verified 2 September 2026.

Substantive commitments resting on an architectural choice, short of the full picture. The strongest element is Bring Your Own Storage, an enterprise option under which all uploaded documents remain inside the organisation's own cloud storage environment, which answers the confidentiality question by removing the vendor from custody rather than by promising to be careful with it. The vendor states the product is built for sensitive privileged legal work with enterprise grade security, privacy and access controls, and that work stays in the environment the team and IT already trust. A policy of not using client data for model training is reported, though located in third party review material rather than on a vendor page reached in this pass. Two gaps hold this off an A. Retention and deletion terms for customers not using Bring Your Own Storage were not located. Attorney client privilege and work product are invoked as a description of the work rather than addressed as a handling question, and no segregation model between users or matters was located.

Two limbs are answered in the agreement about as strongly as this corpus offers, and two are not answered at all. On training, clause 5.4 closes with an express statement that Deposely will not train artificial intelligence models using customer data as training materials, which names the activity and prohibits it in the operative contract rather than on a policy page. On retention, clause 6.4 requires immediate deletion of all proprietary information including customer materials promptly after termination, subject only to legal record-keeping, and clause 5.4 separately confines usage data to anonymised collection and aggregated-only disclosure. The privacy policy does not cut against either, because it expressly excludes user-uploaded content processed for business customers from its scope and routes that to the customer agreement, so its language about improving algorithms and AI models reaches prospect and visitor personal data rather than matter material. Privilege itself is named only in clause 9.5, which places its preservation on the customer rather than describing how the platform supports it. Nothing describes segregation between customers or between matters, and no model provider is identified. Verified 2 September 2026.

Substantive engagement with the professional duty the product exists to serve, short of the specific limbs an A requires. The vendor publishes material on how large firms operationalise responsible AI through citation checking policies, describes the norm courts expect that every sentence carries a verifiable citation in a standardised format, and frames the risk in terms of sanctions and reputational damage to the signing partner. That is engagement with candour to the court and with supervision of others' work, which is more than a disclaimer and more than most records here manage. The audience is professional throughout, spanning litigators, in house teams, courts, arbitrators and government agencies, with no consumer surface located. Short of an A because competence is not addressed as such, no statement on the advice line was located, and no jurisdiction limits are stated beyond references to Bluebook and local style rules.

Three clauses do the work and together they cover the ground this axis asks about, in the contract rather than in a footer. Clause 9.4 states that Deposely is solely a technology provider and provides no legal services or advice, which is the plain statement of what the product is and is not. Clause 9.5 allocates the professional obligations that follow, placing vetting, output review and preservation of privilege on the customer, so the firm is told what it retains rather than merely warned off reliance. Clause 3.3 is the limb that lifts this above the records that stop at a disclaimer: it restricts attorney users to those licensed and in good standing in the United States, and expressly provides that use by paralegals must not constitute the practice of law. That addresses who may operate the system and under what supervision, engages the unauthorised practice question directly rather than by implication, and carries a jurisdictional statement in the licensure requirement. What is not present is a named authority: no bar rule, ethics opinion or guidance instrument is cited, which is recorded on the corresponding signal rather than held against the grade here. Verified 2 September 2026.

A published position on responsible AI without a governance mechanism a buyer could audit. The vendor publishes on how firms should operationalise responsible AI through citation checking policies, and its architectural decision to keep generative models out of the verification path is a real governance choice with a stated rationale rather than a slogan. Searched the vendor site, the blog, the sandbox site and trade coverage on 29 Aug 2026 and located no published AI governance framework, no AI management certification such as ISO 42001, no named owner of model governance, no pre release testing regime, and nothing on uneven output across matter types, parties or populations. For a scoring product the last is a live question: a semantic score that behaves differently across drafting styles or document types would be invisible to a user and nothing published addresses it.

No governance material was located on any surface. Checked the security page in full, the privacy policy in full, the master services agreement by clause, the site navigation and footer and the trust centre entry point on 2 September 2026. There is no responsible AI page, no framework or set of principles, no individual or function named as accountable for model behaviour, no account of what is evaluated before a model or feature change ships, and no certification such as ISO 42001. The security page organises its content under software, governance and compliance, but every item beneath is an information security control, and this axis treats those separately. Nothing anywhere addresses uneven output, which has a specific form on this product: a system analysing spoken testimony will encounter accents, speech patterns, interpreters and varying transcription quality, and nothing states whether performance across those conditions is measured or disclosed. Verified 2 September 2026.

A general posture with one strong architectural element and little published detail. Bring Your Own Storage is real and material, letting an organisation keep uploaded documents in its own cloud, and the vendor states it works with customer security and IT teams through procurement and review. Beyond that the published material is general: enterprise grade security, privacy and access controls that IT and security teams can trust. Searched the vendor site, the blog, the sandbox site and trade coverage on 29 Aug 2026 and located no retention period, no deletion control, no encryption specifics, no named subprocessor list, and no incident or breach notification practice. Note also that published filings are hosted in the vendor's secure cloud for judges and clients to read, which is a second data location with no published terms attached to it.

The control set is specific where it counts and two elements are missing. Encryption is stated on both sides and with detail on one: all stored data is encrypted at rest using industry-standard algorithms, and transport requires TLS with HSTS, which names the mechanism rather than asserting protection. Access controls are described as restricting data to authorised users, with two-factor authentication available on accounts. Assurance is ongoing rather than one-off: regular penetration tests are conducted by third-party security experts, and the software supply chain is regularly scanned for vulnerabilities, which is a control few vendors in this corpus mention at all. Deletion is contractual and immediate, with clause 6.4 requiring destruction of customer materials promptly after termination subject to legal record-keeping. Against that, no incident or breach notification practice was located on any surface, and no subprocessor list exists for customer content, with the agreement acknowledging third-party large language model providers as a category without naming one. Retention during the term is not stated. Verified 2 September 2026.

Searched the vendor site navigation, the blog, the sandbox site and trade coverage on 29 Aug 2026. No published indemnity, liability cap, carve out, warranty on output or insurance position was located, and no customer terms of service was located as published on the property. Recorded as a pure absence on the surfaces reached. The shape of the gap is worth naming: this product is bought specifically to prevent sanctionable citation errors reaching a filing, and it publishes nothing about what happens if it misses one. A verification tool carries a different recourse question from a drafting tool, because its value is the assurance itself.

A complete allocation, published in full, with one provision that runs the customer's way more than most. Clause 10.1 caps liability at twelve months of fees, waives consequential damages in both directions rather than only the vendor's, and carves indemnification obligations out of the cap altogether, so the indemnity is not swallowed by the ceiling. Clause 8.1 is a mutual intellectual property infringement indemnity, which is genuinely unusual here: the great majority of agreements in this corpus run the indemnity provider-side only or, more often, customer-side only. Governing law is Washington with arbitration before JAMS in Bellevue. What keeps this out of the top band is the output side. Clause 9.6 expressly disclaims all warranties as to the accuracy, completeness or reliability of AI-generated content and assigns sole responsibility to the customer, so there is no warranty on output to weigh, and no insurance position is stated anywhere. A reseller channel operates on distinct terms which were not reviewed. Verified 2 September 2026.

CORRECTED 29 Aug 2026 during the trust portal sweep. The previous note stated that no named document management connector such as iManage or NetDocuments was located. That was wrong. The vendor names four document repository integrations in its own published material: Relativity, iManage, NetDocuments and Clio, describing them as letting a firm leverage existing secure infrastructure while adding the verification layer on top. For a litigation verification product that is the right connector set, spanning the two dominant law firm document management systems, the dominant ediscovery platform and the dominant small firm practice management system, and it is a materially stronger position than the note previously recorded. The rest of the row stands and is why this remains B rather than A. Delivery is a native Microsoft Word add in rather than a connector alongside Word, which is where drafting and signing happen. Research integrations are named individually with their commercial terms stated: LexisNexis where the customer holds a subscription, Fastcase and vLex where no subscription is required, with proprietary Westlaw citations mapped across so they remain checkable. Not located as of 29 Aug 2026: an integrations index page, and per integration documentation describing what each connector moves, in which direction, and what an administrator configures. Named connectors without that documentation is the B band.

Integration is a named product line and no system is named on any surface read. The platform navigation carries a dedicated integration offering alongside the deposition, facts, agent and agencies lines, so connectivity is sold as a distinct capability rather than mentioned in passing, and the presence of a separate offering for court reporting agencies implies a data path from the agencies producing transcripts into the firms consuming them. That page was not opened on this pass, which is recorded as a limit on this reading. What could not be established is anything concrete. No document or case management system is named, no transcript or court reporting platform, no repository, and no connector list, application programming interface or developer documentation was located. For a product whose input is transcripts produced by third parties and whose output belongs in a case file, how material arrives and where it goes afterwards is the integration question, and nothing published on the surfaces read answers it. Verified 2 September 2026.

Deployment model is stated clearly with a genuine customer controlled option, short of residency detail. Two models are published and the difference is meaningful: the standard cloud service, and an enterprise Bring Your Own Storage option under which uploaded documents remain in the organisation's own cloud storage environment, which the vendor presents as giving additional control over data residency and confidentiality. Customer held storage is a stronger answer than a region selector because the customer keeps custody. What is missing is the rest of the picture: searched the vendor site, the blog, the sandbox site and trade coverage on 29 Aug 2026 and located no hosting provider for the standard service, no named regions, no tenancy model, and no statement of where processing happens as distinct from where documents are stored, which matters because processing occurs even when storage is the customer's.

Nothing published addresses where the platform runs or how customers are separated within it. Checked the security page in full, the privacy policy in full, the master services agreement by clause, and the site navigation and footer on 2 September 2026. No cloud provider is named, no hosting region or country is stated, no residency option is offered, nothing distinguishes processing from storage, and no tenancy or isolation model is described. Cloud delivery is evident only by inference from a browser application at a login subdomain. The absence is consequential rather than formal on this product: the material processed is deposition testimony and case documents, which are among the most sensitive holdings a firm has, and a defence firm with client-imposed residency terms in its outside counsel guidelines has nothing published to assess. A Vanta-hosted trust centre exists and returned page metadata with no readable body, which is recorded as a retrieval limit rather than as an absence. Verified 2 September 2026.

CORRECTED 29 Aug 2026 during the trust portal sweep. Previously recorded with the certification claim attributed to third party review material and the note stating it was not located on a vendor page. That was wrong: the vendor states SOC 2 Type 2 certification on its own blog, describing it as demonstrating that it meets rigorous data protection standards and helping a lawyer fulfil the duty to protect client confidentiality through verified security controls. Source basis corrected from third party to vendor published accordingly. Also stated by the vendor in the same material: the Bring Your Own Storage option, framed explicitly as letting a firm control where client data is stored to meet jurisdictional requirements or firm retention policies, which is the vendor connecting that feature to professional obligations rather than to convenience. The grade nonetheless remains C for the reasons originally given, none of which the correction touches: no coverage period, no audit scope, no report date, no named auditing firm, no trust portal and no published request route were located, and the vendor's stated diligence path is to work with a customer's security and IT teams through procurement, which is a sales linked route rather than a self serve one. A certification asserted on a marketing page without scope, currency or an evidence route is the C band.

Two attestations are claimed with a trust surface behind them, and the evidence itself is not published. The security page states SOC 2 Type II certification and describes what the standard demonstrates, and separately states that the company has audited its practices for HIPAA compliance while candidly noting that Litem is not typically a business associate under those regulations, which is a more careful framing than the badge-wall approach and is worth crediting. Regular third-party penetration testing is described as ongoing rather than historical. A trust centre exists at the company's own subdomain and is signposted from the security page. What is absent is what a reviewer would test. No auditor or certifying body is named for the SOC 2 work, no report period, issue date or expiry is published, no scope statement identifies which services are covered, and no penetration test summary or date is given. The trust centre is Vanta-hosted, rendered page metadata with no body on 2 September 2026, and its access tier is therefore unestablished; that is a retrieval limit on this reading, not an absence. Verified 2 September 2026.

The architecture is described more clearly than most and the components are not named. Published and unusually specific: the verification path runs on classic machine learning and natural language processing rather than generative models, with a patented scoring system, so a buyer knows the checking function does not depend on a third party foundation model at all. Generative features exist separately for drafting. Searched the vendor site, the blog, the sandbox site and trade coverage on 29 Aug 2026 and located no named model or provider behind the generative features, no statement of where models run, no subprocessor list, and no commitment to notify customers when the supply chain changes. Knowing that the critical path avoids generative models is genuinely useful and is not the same as knowing what the rest of the stack is.

The category is acknowledged in the contract, one provider is named in a document that excludes the relevant data, and the two do not meet. Clause 5.3(f) of the master services agreement defines third-party large language model providers as a term, so a reader knows external models are in the path and that the vendor contemplated saying who they are. It names none. The privacy policy does name one, listing artificial intelligence and machine learning providers with Google Cloud AI given as the example among the service providers who receive data. But that policy states at the outset that it does not apply to files, documents and other uploaded content processed on behalf of business customers, routing that to the customer agreement, so the naming reaches prospect and visitor personal data rather than deposition transcripts and case files. A firm therefore cannot tell from anything published which providers see client material. No model or version is named, nothing states where inference runs or what a provider may retain, and no change notification commitment exists. Verified 2 September 2026.

Checked the vendor site navigation, the blog and the sandbox site on 29 Aug 2026. No pricing page was located, no rate is published, no unit of charge is stated and no tier structure appears on the surfaces reached. Third party review material states that pricing is custom quoted, which confirms the absence rather than filling it. Every commercial path located terminates in a demo or contact request. One published commercial fact worth recording, though it is not pricing: a recipient of a hyperlinked filing, including a judge or a client, does not need a Clearbrief subscription to open it.

No pricing information is published at any level. Checked the security page, the privacy policy, the master services agreement by clause and the full site navigation and footer on 2 September 2026. Under the rule that a site's own navigation establishes its page inventory, there is no pricing page to be unreachable: the platform menu carries the five solution lines plus security and contact, the resources menu carries careers, blog, help centre and onboarding, and the footer repeats those with the privacy policy and the master services agreement. Nothing names a plan, a tier, a unit of charge or a structure, and nothing states whether the product is licensed per seat, per matter, per deposition or per volume of transcript, which for a product handling more than 10,000 depositions a month is the first question a buyer would ask. Every call to action routes to a sales conversation through a scheduling link. No pricing row is written, because the rule that triggers one requires published structure and none exists. Verified 2 September 2026.

Segment coverage is described with substance and the practice focus is unusually well defined. The stated user set spans litigators, in house legal teams, courts, arbitrators and government agencies, and the inclusion of courts and arbitrators is distinctive: this is the only record on the index whose published adopters include the tribunal as well as the advocate. Practice focus is stated narrowly and consistently as fact intensive litigation documents, with named use cases covering briefs, tables of authorities, exhibit management, investigation reports for in house compliance teams, and real time trial support including cross examination outlines. Not located as of 29 Aug 2026: firm size segmentation, jurisdictional coverage beyond references to Bluebook and local style rules, and any statement of which document types or practice areas the product is not built for.

Two distinct buyer segments and a lifecycle description, with the boundary left open. The company states that it serves law firms and court reporting agencies, and backs the second with its own product line rather than a mention, which is a real segmentation choice: the agencies producing transcripts and the firms consuming them are different buyers with different needs. Coverage across a case is described in phases rather than as a single capability, starting in discovery, assisting through depositions in real time and extending into further litigation phases with agentic features, and the five named solution lines map onto that. Jurisdiction is stated indirectly but bindingly, since clause 3.3 requires attorney users to be licensed in the United States, which tells a non-US firm the answer. Named customers are large defence firms. What is missing is the rest: no practice area is named anywhere, no firm size band is given, nothing distinguishes what the plaintiff and defence sides get, and nothing states where coverage stops. Verified 2 September 2026.

An explicit policy of not using client data for model training is reported in third party review material summarising this vendor's security posture. Searched the vendor site, the blog and the sandbox site on 29 Aug 2026 and did not reach a vendor page stating that policy in its own words, so the value rests on a secondary source rather than on located vendor material, and the source basis records that. Two supporting facts from vendor material that are consistent with it without stating it: the Bring Your Own Storage option keeps uploaded documents in the customer's own cloud, and the verification path deliberately avoids generative models. Recorded at policy never on the strength of the reported policy, and rebuttable in either direction by a vendor page.

This is the cleanest instance of the value in the pull. Clause 5.4 of the master services agreement closes with the quoted undertaking, which names the activity and prohibits it in the operative contract, so it clears the naming test outright rather than by construction. Three things reinforce it. The same clause confines usage data to anonymised collection and aggregated-only disclosure, so the derived-data carve-out that usually swallows these prohibitions is itself bounded. The security page points at the contract rather than away from it, stating that zero client data is used for AI training and adding that it is in writing in the master services agreement, which is a vendor sending a reader to its own clause instead of asking for trust. And the privacy policy does not conflict, despite containing language about improving the company's algorithms and AI models: it states at the outset that it does not apply to files, documents or other uploaded content processed on behalf of business customers such as law firms, routing that to the customer agreement, so its improvement language reaches prospect and visitor personal data rather than matter material. Marketing, policy and contract agree.

Searched the vendor site, the blog, the sandbox site and trade coverage on 29 Aug 2026. No public material states how long uploaded documents, analyses or generated reports are retained, whether a customer controls the window, or whether deletion is available. The Bring Your Own Storage option changes the question for enterprise customers by keeping uploaded documents in their own cloud, so those customers set their own retention by controlling the store, but that option is not the default and nothing published addresses retention for customers without it. A second location is also unaddressed: filings published as hyperlinked versions for judges and clients are hosted in the vendor's cloud, and no retention terms for that store were located.

The post-termination obligation is contractual and immediate; nothing addresses the term. Clause 6.4 requires immediate deletion of all proprietary information, expressly including customer materials, promptly after termination, subject only to retention required for legal record-keeping. The period is expressed as immediacy rather than as a number of days, which is a definite obligation rather than a reasonableness standard and is why this is not recorded as vague, but a reader should note the distinction: there is no stated window in which to retrieve data before deletion, and no export mechanism is described. Clause 5.4 separately bounds what survives in derived form, confining usage data to anonymised collection and aggregated-only disclosure. What is absent is any retention position during the term: nothing states how long transcripts, uploaded case documents, generated analyses or prompts persist while the subscription runs, and no retention setting is described as available to the firm. The privacy policy's retention language covers personal data of prospects and visitors and is expressly scoped away from user-uploaded content.

Searched the vendor site, the blog, the sandbox site and trade coverage on 29 Aug 2026. No vendor material addresses segregation between users or matters, and no conflicts or ethical wall concept was located. The vendor refers to access controls that IT and security teams can trust without describing them. Bring Your Own Storage means an enterprise customer's documents sit in a store they already govern, so their own controls apply to the documents at rest, but nothing published states whether retrieval and concept search inside the product respect those controls per user at query time, which is the question this signal asks. No legal document management integration was located whose permissions could be inherited.

Checked the security page in full, the privacy policy in full, the master services agreement by clause, and the site navigation and footer on 2 September 2026. Nothing describes segregation between customers or between matters. The security page addresses access control only in general terms, stating that strong controls ensure only authorised users can access data, without describing what the boundary follows. The question is live on this product rather than formal. A litigation platform holds deposition testimony and case documents for opposing sides of the same dispute in principle, and the company sells separately to court reporting agencies, which sit between parties and produce transcripts for both, so a path exists by which material from one matter could meet another. Nothing published states what prevents it, whether ethical walls can be represented, or whether analysis performed on one matter can draw on another. Clause 9.5 places preservation of privilege on the customer, which allocates the duty without describing any mechanism supporting it.

Searched the vendor site, the blog, the sandbox site and trade coverage on 29 Aug 2026, and no published customer terms or privacy policy was reached. No clause committing to notify a customer of a government or law enforcement request for their data was located, and no transparency report was located. Noted for a reader: for enterprise customers using Bring Your Own Storage the practical exposure is reduced, since documents sit in a store the customer controls and a request would more naturally be directed to them, but the vendor does not make that argument in published material and it is not recorded as a value here.

The commitment is in the operative agreement and is notice before the fact rather than after it. Clause 6.3 permits disclosure under judicial or governmental order provided the receiving party gives the quoted notice, which ties the notice to the customer's ability to contest rather than merely informing them once material has gone. The obligation is expressed reciprocally, applying to the receiving party whichever party that is. Two limits belong on the record. No transparency report was located: the security page, the privacy policy, the blog and the site navigation were checked on 2 September 2026 and none publishes aggregate figures on demands received or complied with, which is what would lift this to the top value. And the clause carries no express undertaking to seek a protective order itself or to narrow the disclosure to what is strictly compelled, both of which appear in the strongest formulations in this corpus; here the burden of contesting sits with the customer, which is what the notice is for.

Sources are named individually with their access terms stated, which is unusual, and the rights basis behind them is not the vendor's to state. The product does not hold its own primary law corpus: it displays cited authority through named third party databases, LexisNexis where the customer holds a subscription and Fastcase and vLex where no subscription is required, and maps proprietary Westlaw citations across to those databases so they can still be checked. Naming which database serves which citation and whether the customer must pay for it is more concrete than most coverage statements on this index. What is not published is the licence or rights basis on which the vendor accesses those databases, nor any completeness or update cadence statement for the underlying caselaw, which sits with the database owners rather than here. Recorded at sources named with basis unstated.

Checked the security page in full, the privacy policy in full, the master services agreement by clause, and the site navigation and footer on 2 September 2026. Litem ships no corpus of its own and none is claimed. The material the platform works on is the customer's, being deposition transcripts, testimony and case documents supplied by the firm or by the court reporting agency acting for it, so there is no licensed database, published collection or third-party content behind any output and the provenance risks this signal tracks do not arise in their usual form. The honest record is that the question is not addressed rather than that a corpus was withheld. Nothing published describes the training data behind the underlying models either, which is a different question carried on the model supply chain axis.

The product performs two checks that are commonly confused with this one and is not documented as performing this one. What it does, published clearly: confirms a cited case exists and is not fabricated, and scores how well the cited page supports the proposition the sentence advances, flagging low semantic scores where the source may not support the assertion. Those address existence and support. What this signal asks is whether the authority is still good law, meaning subsequent history and treatment. Searched the vendor site, the blog, the sandbox site and trade coverage on 29 Aug 2026 and located no statement that the product checks whether a case has been overruled, distinguished or otherwise treated, and no commercial citator licence for treatment signals. Third party comparison material refers to citation validity tools in general terms without resolving it, and an ambiguous claim earns nothing. Recorded as not addressed with the distinction stated, because for this vendor in particular the difference between verifying a citation and validating the law is the whole question.

Checked the security page, the privacy policy, the master services agreement and the site navigation on 2 September 2026. The product analyses testimony rather than retrieving primary law, so a citator has nothing to operate on and the absence is not a criticism. The analogous question is nonetheless real and unaddressed. Deposition transcripts are corrected by errata, and testimony is superseded, clarified or struck as a case proceeds, so an analysis resting on a version of a transcript that has since been amended is wrong in the way a citator exists to catch. Nothing published describes whether the platform tracks transcript versions, surfaces errata, or revisits earlier analysis when the underlying record changes. The company's own blog writes about citing transcripts accurately and about the consequences of inaccurate citation, which makes the silence on the product side more noticeable.

Searched the vendor site, the blog, the sandbox site and trade coverage on 29 Aug 2026. This signal fits this product awkwardly and the fit is recorded rather than forced. The product does not answer legal questions from a corpus, it evaluates assertions a lawyer has already written, so the failure mode the signal targets, a confident answer to an unanswerable question, is largely outside its design. The nearest published behaviour runs the other way and is favourable: where a cited source cannot be displayed automatically because it is proprietary to another database, the product reports that the citation is verifiable rather than passing it or failing it silently, and invites a manual upload. That is a documented handling of an unresolvable case. It is not a statement about what the product does when it cannot ground an answer, so it does not satisfy this signal, and no such statement was located.

Checked the security page in full, the privacy policy in full, the master services agreement by clause, and the site navigation and footer on 2 September 2026. Nothing describes what the system does when it cannot produce a reliable result, and no confidence indicator is described as shown to the user for any output. The agreement resolves the question by allocation rather than by design: clause 9.6 states that no warranty is given as to accuracy, completeness or reliability and that the customer must verify before use, which tells a firm to check everything rather than telling it where checking matters most. That is the opposite of an uncertainty signal, since it distributes attention evenly instead of directing it. Nothing states what happens where audio quality is poor, where a speaker is unclear or interpreted, or where the system cannot locate support in the record for a proposition it has been asked about. The five solution pages were not opened on this pass.

No court order, opinion or disciplinary record naming this product has been located as of 29 Aug 2026. Instrument searched: the AI Hallucination Cases database maintained by Damien Charlotin, which tracks court decisions worldwide involving AI generated hallucinated content and records the AI tool implicated where it is known. Also checked published 2026 sanctions summaries and secondary sanctions trackers. The entries located name filers, and in some rows other products, rather than this one. This is a statement about the public record on the date shown and not a clearance. Worth recording for this vendor specifically: the product exists to detect fabricated citations before filing rather than to generate text, so the mechanism by which most products reach this database does not apply to it, and a hypothetical entry would more likely concern a citation it failed to catch than one it produced.

Searched the AI Hallucination Cases database maintained by Damien Charlotin, and reporting drawing on it, on 2 September 2026 on both the current and former product names, Litem and Deposely, and on the corporate name Deposely, Inc. No court order, opinion or disciplinary record naming the product was located. This is a statement about the public record rather than a finding about the product. One feature of this product class makes the search less reassuring than it looks and is recorded so a later reader does not over-read the value: the characteristic failure here is a misattributed or misquoted piece of testimony rather than a fabricated case citation, and a misquotation of the record surfaces as an evidentiary objection or a correction at deposition rather than as a sanctions order. The recent rename from Deposely to Litem also means any older record would sit under a name the tracker may not connect to the current one, which is why both were searched.

Public materials engage substantively with professional obligations in general terms without naming the guidance that states them. The vendor publishes on how large firms operationalise responsible AI through citation checking policies, describes the norm courts expect that every sentence carries a citation the court can look up and verify, refers to Bluebook and local style rules, and frames the risk as partners being sanctioned and suffering reputational damage for citation errors they did not personally make. That is engagement with candour to the court and with supervisory responsibility, and it is more substantive than the general references most records here carry. Searched the vendor site and blog on 29 Aug 2026 and located no engagement with a named ethics opinion, including ABA Formal Opinion 512, and no state bar guidance or standing court order on AI disclosure named specifically.

Professional obligations are engaged directly in the contract and no authority is named. Clause 3.3 requires that attorney users be licensed and in good standing in the United States and provides that use by paralegals must not constitute the practice of law, which is a substantive engagement with the unauthorised practice question rather than a gesture at professional standards, and it is more than most records in this corpus offer. Clause 9.5 adds preservation of privilege to the customer's obligations. But no rule, ethics opinion, bar association or guidance instrument is cited anywhere: neither the American Bar Association's opinion on generative artificial intelligence nor any state rule on competence, supervision or unauthorised practice is named, and no jurisdiction beyond the licensure requirement is identified. Checked the security page, the privacy policy, the master services agreement by clause and the site navigation on 2 September 2026.

Savings are claimed and the framing sidesteps the ethics problem rather than creating it, which is worth recording because no other vendor on this index has done so. The published claim is that the product significantly reduces time spent on manual non billable tasks such as fact checking and building tables of authorities. Time saved on work a firm was not billing for does not change the invoice, so the tension this signal exists to surface, marketing hours saved to the firm while staying silent on the client's side, largely does not arise on that framing. Searched the vendor site, the blog and trade coverage on 29 Aug 2026 and located no per matter record of AI assisted work intended for fee purposes and no guidance on billing, fee or client disclosure treatment. Recorded at savings claims only as the accurate value, with the non billable framing noted so a reader does not read it as the usual version of this gap.

A time-savings claim is published and attributed, and nothing addresses the bill. The clearest figure on the record comes from a named litigator reporting that preparation time was cut by half, and the product is sold throughout on the compression of deposition preparation and review, which is precisely the work that used to be billed by the hour. Nothing addresses what happens to the invoice when it is not. No per matter record of AI-assisted work is described, nothing states whether analysis performed by the platform is identified as such on a bill, and no guidance is offered on fee or disclosure treatment. The gap has particular weight in this lane, because deposition preparation is classically billed time on a matter where the client often receives a detailed narrative of what was done, and a firm whose preparation now takes half as long has something to explain or to absorb. Neither is engaged on any published surface.

Material exists and reaching it runs through a conversation. The vendor states it works with customer security and IT teams through procurement and review, which is an explicit offer to supply diligence material rather than a published pack, and third party review material reports SOC 2 Type 2 certification and a no training policy that a firm would want in writing. Bring Your Own Storage is a genuine answer to part of what a client AI clause asks, since client documents can remain in the firm's own cloud. Searched the vendor site, the blog and the sandbox site on 29 Aug 2026 and located no subprocessor list, no statement naming which model providers see client content, no published data processing agreement, and no client facing consent or notification pack. Recorded at on request on the strength of the stated procurement route.

No provider is named for customer content, which is what decides this. Clause 5.3(f) of the master services agreement defines third-party large language model providers as a term, so the agreement contemplates external models in the path and then names none. The privacy policy does name one, citing Google Cloud AI as an example among artificial intelligence and machine learning service providers, but that policy expressly excludes files, documents and other uploaded content processed for business customers from its scope, so the naming does not reach transcripts or case material and a firm cannot rely on it to answer a client. No subprocessor list exists in any form. What a firm can forward today is real but not AI-specific: the master services agreement is published in full and contains the training prohibition at clause 5.4, the deletion obligation at 6.4 and the disclosure notice at 6.3, and the security page states SOC 2 Type II and HIPAA positions. A Vanta trust centre exists and returned no readable body, recorded as a retrieval limit.

First positive value on this signal in the index, and the product is built for it. The Cite Check Report is a downloadable per document artifact listing every factual and legal citation the document contains, flagging missing sources, formatting errors and low semantic scores, with hyperlinks to each item so the signing attorney can review it in context, and the vendor describes it explicitly as an audit trail of citation review activity. Separately the product publishes hyperlinked versions of filings that a judge or client can open without holding a subscription, so the sources relied on travel with the document to the tribunal. Sources retrieved and human verification are both covered at document level. One limb of the positive value is not: model used is not recorded, and would be an odd fit here since the verification path deliberately runs on classic machine learning rather than a generative model. Recorded at exportable record because a genuine per document export exists, with that gap stated rather than hidden.

Checked the security page in full, the privacy policy in full, the master services agreement by clause, and the site navigation and footer on 2 September 2026. Nothing addresses disclosure of AI use to a court or tribunal, and no exportable record of which outputs a model produced, on what basis, or who reviewed them is described. The gap is pointed on this product because its outputs feed material that is filed and argued: an analysis of testimony informs a motion, a cross-examination outline or a summary judgment brief, and a court asking how a passage of testimony was located and characterised would be asking exactly this question. The agreement places verification on the customer at clause 9.6 and vetting and output review at 9.5, so the firm carries the duty, but nothing describes a record it could produce to show the duty was discharged. No audit trail of model use, review or approval is mentioned anywhere.