Legalyze vs Parambil: how they compare in 2026

The purest AI product in the category, and the vendor makes the claim in the strongest available form by stating that there is no human in the loop and that this is what keeps costs low. Where EvenUp sells an expert reviewed tier and Supio verifies bill summations with human experts on high value cases, Legalyze positions the absence of human review as the commercial proposition. Everything the product delivers is model output: chronology construction from uploaded records at ten pages per minute, event classification by visit type, page level source resolution, plain English question answering over the case file, and detection of missing records, treatment gaps, prior injuries and inconsistencies. Handwritten, scanned and electronic records are all handled, which is extraction work across three quite different input problems. Remove the models and there is no product. Seventh consecutive A on this axis in this category, and the pattern is now unbroken across all seven plaintiff side vendors.

There is no content asset and no conventional product underneath. Parambil licenses nothing and publishes nothing of its own; the material it works on is the customer's medical records. Every deliverable it sells is a model output: the chronology, the de-duplicated Provider Guide, the billing analysis, the missing-document detection, the agents that investigate a case end to end and draft complaints and discovery, and the natural-language query layer over the record. Remove the models and what remains is an upload folder. The company's own framing matches, describing the platform as automating the ingestion, synthesis and verification of unstructured medical records, and its comparator is not another software product but manual review by paralegals, legal nurse consultants and retired physicians. That is the clearest form this axis takes: the AI is not accelerating an existing product, it is the whole of what is bought. Verified 2 September 2026.

Page level grounding applied consistently across both output types, with published throughput and no published accuracy. Every medical event can be opened in one click with a direct link to the exact page in the original record, and Case Chat answers are each cited to the exact location in the source documents, so the citation mechanism covers the conversational surface as well as the generated chronology, which several peers do not. Independent material identifies why that matters in this category, noting source linking to the original page is the detail that counts when opposing counsel challenges a timeline. Held at B because the published numbers are speed rather than correctness: ten pages per minute, and a thousand page file reduced from fifteen to twenty hours to roughly two. Those are benchmarks of throughput. No accuracy figure, no extraction recall, no error rate on handwritten records where extraction is hardest, no evaluation and no test corpus were located. A vendor that publishes its processing rate to the page and nothing about correctness has measured the easy quantity.

Grounding is real, specific and structurally verifiable. Every chronology entry is hyperlinked to its source page, in-text citations link each summary detail to the original document for instant validation, and the platform flags documents referenced in the record but missing from the file, which is a form of telling the user where the evidence base is incomplete rather than filling the gap. An accuracy figure is published and it is unusual: a 44 per cent increase in correct information against traditional methods, alongside 95 per cent faster and more than five million records reviewed. What is not published is the method behind it. No test set, comparator definition, sample, evaluation protocol or date accompanies the 44 per cent claim, and nothing states who conducted the measurement or against what baseline, so a buyer cannot check the single number that would most justify the purchase. Nothing addresses hallucination in the drafting layer, where the Drafting Agent produces complaints and discovery requests rather than retrieving from the record. Verified 2 September 2026.

The oversight model is disclosed by being disclaimed, which is unusual and is credited as disclosure even though what it discloses is an absence. The vendor states plainly that there is no human in the loop, framing it as the reason its costs are lower than competitors that employ reviewers. A buyer therefore knows exactly what they are getting, which is more than a vendor that leaves the question open. What is absent is any account of the verification that replaces it: nothing states whether output is checked by any automated means before delivery, whether a low confidence extraction is marked, what the system does when a page is illegible, or what review the vendor expects the attorney to perform. The product does surface missing records, treatment gaps and inconsistencies, which is checking the records rather than checking the model. Held at C: the model is stated, its bounds are not, and the entire verification burden sits with the firm without the vendor saying so.

This is a high-autonomy product with almost no published oversight model, and the gap between the two is the finding. The agents are described as executing multi-step workflows and investigating cases without continuous prompting and without waiting for prompts, re-evaluating conclusions as new records arrive, with a Case Agent working end to end and a Drafting Agent generating complaints, demands, interrogatories and discovery requests. Against that, the published verification affordances are real but passive: every assertion carries a hyperlink to its source page, and incomplete evidence is flagged. Nothing states where a human is required, what an agent completes before a person sees it, what happens when the record is ambiguous or the agent cannot support a conclusion, or what a firm should check before a generated complaint is filed. The website terms of use are the only agreement published and they address site access rather than the platform, so no review obligation exists in any document. For a product drafting pleadings from medical evidence unprompted, the oversight model is the thing a buyer most needs and is the thing least described. Verified 2 September 2026.

Published benchmarks are specific and no customer is identifiable. Stated: ten pages per minute processing, a thousand page file moving from fifteen to twenty hours of manual review to roughly two, and thousands of records reviewed in hours. Independent material attributes those figures to Legalyze's published benchmarks, which is a vendor source relayed rather than an independent measurement. Two further items carry weight without naming anyone: the vendor states it has passed extensive security reviews at large law firms, which is an assertion about procurement outcomes at unnamed firms, and independent material describes Legalyze as among the most widely adopted tools for personal injury medical chronology work alongside Supio and DigitalOwl. Partner listings with Smokeball and CASEpeer are corroborated on those vendors' own surfaces. Held at C because no law firm is named, no case study exists, the one user quote located is anonymous, and no funding, analyst placement or award was found.

The reference base is strong and specific. Fourteen customer logos appear on the home page, including Kline & Specter, Jenner & Block, Cory Watson, Meyers & Flowers and Kershaw Talley Barlow, which is a serious roster in mass tort and medical malpractice rather than a set of unknown names. Four testimonials carry a name, a title and a firm, among them Jay Edelson, founder of Edelson PC, and Benjamin Present of Kline & Specter, whose comment is substantive rather than promotional in that it claims the platform surfaces what a paralegal, legal nurse consultant and physician expert would miss. Four figures are published: more than five million medical records reviewed, over 20,000 hours saved, 95 per cent faster than traditional methods, and a 44 per cent increase in correct information. What holds this below the top band is that not one figure carries a method, a date or a named customer. No case study attaches an outcome to any of the fourteen firms, the hours-saved and accuracy figures are aggregate and unsourced, and nothing is dated, so a buyer can see who uses it and what is claimed but cannot connect the two. Verified 2 September 2026.

Infrastructure control is stated with one useful specific and no privilege position exists. Published: all data in the system is encrypted, and data is stored on servers under Legalyze's own control, which is a statement about custody rather than a generic secure cloud reference and rules out the arrangement where a vendor's data sits on infrastructure it does not govern. Business associate agreements are stated to be in place with the AI providers, which addresses the downstream chain for protected health information. Against that: no business associate agreement is described as being offered to customers, which is the instrument a law firm actually needs to execute, and independent guidance in this category names a signed BAA before any protected health information is processed as the first baseline check. No SOC 2 or independent attestation was located, and no treatment of attorney client privilege or work product appears. Held at C on that footing, below the peers at B which carry a customer facing BAA, an attestation, or both.

One limb is answered well and the rest are not answered at all, on a product handling the most sensitive category of case material. The model supply chain is named openly, with Anthropic identified for the agent layer and Perplexity for live web search, which is more candour than most records in this corpus offer. HIPAA support and a business associate agreement on request are published, which is the right regime for medical records and is a real commitment. Beyond that the posture is thin. Nothing addresses whether customer content is used for training, in either direction. Nothing describes segregation between customers or matters; the only access statement is that granular controls govern user roles and system-wide data visibility, which is permissioning inside an account rather than a boundary between them. Privilege and work product are not mentioned anywhere, and the attorney work product doctrine is the doctrine that actually protects a litigation chronology. Retention is vague, and the website terms state that user content may be deleted at any time without prior notice with no backup obligation. One specific exposure is undescribed: a live web search feature on a platform holding protected health information, with nothing published about what leaves the system when it runs. Verified 2 September 2026.

Not located. The no human in the loop positioning makes the question sharper than for peers rather than softer: the vendor has removed the intermediate professional review its competitors sell, so the attorney is the only reviewer between generated output and reliance, and nothing published addresses that responsibility. No statement that output is not medical or legal advice, no positioning on the supervising attorney's verification duty over a machine built chronology, and no engagement with any bar guidance. Checked the home page, the medical chronologies page, the competitor comparison pages, the blog material and the partner listings on 29 Aug 2026.

Nothing published addresses the professional responsibility questions this product raises. Checked the home page, solutions page, practice areas page, security page, terms of use in full and privacy policy in full on 2 September 2026. There is no statement that output is not legal advice, nothing about who may use the platform or under what supervision, no reference to competence or verification duties, and no jurisdictional statement. The closest thing is a positioning claim rather than a professional one: Parambil describes itself as a neutral, third-party software vendor and an independent technology platform delivering fact-based insights to all parties, which addresses partisanship rather than the practice of law. The absence carries more weight here than on a research tool. The Drafting Agent generates complaints, demands, interrogatories and discovery requests, and the chronologies are described as courtroom-ready and as surfacing missed diagnoses and causation links, which is analysis close to expert opinion. A platform producing filings and causation analysis for plaintiff and defense firms, publishing nothing at all on the advice line, sits at the bottom of this axis. Verified 2 September 2026.

Nothing published about how the models are governed, evaluated or monitored. No AI policy, no model card, no bias or fairness testing, no evaluation methodology, no accuracy monitoring, no drift statement, no named governance body and no ISO 42001. The vendor publishes competitor comparison content ranking platforms on performance and cost efficiency, so it engages the evaluation question commercially and not for its own system. The category risk applies with a specific edge here: the product classifies handwritten and scanned records where documentation quality varies systematically by provider and setting, and no evaluation of performance differences across record types or populations is published. Seventh consecutive D on this axis in this category, and the pattern is now unbroken across all seven records. Checked the home page, the medical chronologies page, the blog material and the partner listings on 29 Aug 2026.

One sentence exists and nothing stands behind it. The security page states that Parambil leverages an advisory board of leading experts from large institutions and academia to help build trusted AI systems. That names a structure without naming a member, a remit, a cadence, an output or a single decision it governs. Nothing else was located: no responsible AI page, no governance framework or principles, no individual or function accountable for model behaviour, no account of what is tested before a model or agent change ships, and no certification such as ISO 42001. Nothing anywhere addresses uneven performance, which matters unusually here because the inputs are medical records and the outputs feed damages and causation analysis, so differences in transcription and extraction quality across handwriting, facility, record format or patient population would translate directly into case value. Checked the home page, solutions page, practice areas page, security page, terms of use and privacy policy on 2 September 2026. Verified 2 September 2026.

The most complete stewardship position in this category, and the only one on the index that extends a commitment down the supply chain. Three published statements do the work. First, Legalyze never uses customer data to train its AI models, stated without qualification and not scoped to third party, public or shared models, which distinguishes it from the carve outs recorded on LinkSquares, Eve and Exterro. Second, business associate agreements are in place with the AI providers, which contractually covers protected health information reaching the downstream models rather than leaving that boundary undescribed. Third, all data is encrypted and stored on servers under the vendor's own control. Naming the downstream contractual instrument is what lifts this above its peers: every other record in this category answers what the vendor does with the data and leaves what its model providers do unanswered. Held at B rather than A because it remains policy rather than a customer facing term, no retention period is published, no deletion right is described, and the providers themselves are not named.

Two elements are published clearly and three are not. Access control and protection are stated: multi-factor authentication enforced across all user and admin accounts, granular permissions governing user roles and data visibility, and encryption of all sensitive data at rest and in transit. HIPAA support with a business associate agreement on request is published, which is the governing regime for this data class and is a genuine commitment. What is missing is the rest. No subprocessor is named anywhere: the security page refers to verified enterprise-grade cloud infrastructure partners and to regularly auditing vendors without identifying one, and the privacy policy lists categories such as hosting, cloud services and IT providers rather than entities. No incident or breach notification practice was located on any surface. Retention is vague in the privacy policy and adverse in the terms of use, which state that user content may be deleted at any time without prior notice and that the company has no obligation to back it up. The privacy policy itself, dated October 2023 and covering the Platform by its own terms, never once mentions medical records, protected health information or HIPAA, so the published document does not reach the data the product exists to process. Verified 2 September 2026.

No published position located. Nothing was found on liability for AI output, warranty, service levels or remedy. The exposure is concentrated rather than diffuse on this record because of the no human in the loop model: an omitted provider, a misread handwritten entry or a missed treatment date reaches the attorney with no intermediate check, and a chronology built on an extraction error will read as complete. The product's gap detection surfaces missing records and treatment gaps, which mitigates one failure mode and does not address what happens when the extraction itself is wrong. No published vendor position addresses any of it. Checked the home page, the medical chronologies page, the comparison pages and the site navigation on 29 Aug 2026.

No agreement governing the platform is published, so nothing published answers who bears the loss when the output is wrong. The only liability text located is the website terms of use, version 1.0 dated 26 March 2024, and its scope is stated on its face: it governs use of the website, and section 2.1 licenses the site for the reader's own personal, noncommercial use, which is not a description of a platform bought by law firms. Under the rule that credit follows scope, that document cannot be read as the platform's allocation of risk. Recorded for completeness rather than credited: it caps the company's total liability at fifty US dollars for any cause whatsoever, disclaims all warranties including accuracy, gives no indemnity to the customer while requiring the user to indemnify the company, and compels individual JAMS arbitration with class and jury waivers subject to a thirty day opt-out. No master services agreement, subscription agreement or business associate agreement is published, and the BAA is available only on request. A firm handing over client medical records therefore cannot see, before contacting sales, any cap, indemnity, warranty or insurance position that would apply to it. Verified 2 September 2026.

Two case management integrations named and both corroborated outside the vendor's own site. Smokeball is published with a dedicated integration generating chronologies directly from documents already in Smokeball, and it is listed in Smokeball's own partner marketplace rather than asserted only by Legalyze, which is independent confirmation the relationship exists. CASEpeer has a dedicated integration page describing upload from CASEpeer into Legalyze, with a one week trial offered to CASEpeer firms. Both are the case management systems plaintiff firms actually run on. Held at B rather than A because the depth is shallow and one directional: both integrations are described as getting documents in, nothing describes finished chronologies returning to the matter file, no API is offered, no authentication or permission detail is published, and only two systems are covered against Tavrn's four plus API access. Checked the Smokeball partner listing, the CASEpeer integration page and the site navigation on 29 Aug 2026.

No integration into the systems legal work already lives in was located, and none is claimed. Checked the home page and its full feature list, the solutions page, practice areas page, security page, terms of use and privacy policy on 2 September 2026. No case management or document management system is named, no connector or integration page exists, and no API or developer documentation was located. What the platform offers instead is export: chronologies, notes, billing ledgers and data tables download to Word or PDF, as single pages or complete case packages. That is a file leaving the system rather than two systems staying in step. The gap is conspicuous in this lane, where the neighbouring records are the case management systems a plaintiff firm already runs, so a firm using Parambil moves records in and documents out by hand at both ends. Verified 2 September 2026.

Custody is stated and geography is not. The vendor states that data is stored securely and encrypted on servers under Legalyze's own control, which answers who holds the infrastructure and says nothing about where it sits. No country, region or data residency commitment is published, no hosting provider is named, and no single tenant or dedicated instance option is described. The statement about server control is credited on the Privilege axis where it belongs and does not substitute for a residency answer here. Compare Supio at B with data centres named in three countries, and ProPlaintiff at C for stating United States storage. Checked the home page, the medical chronologies page, the comparison pages and the site navigation on 29 Aug 2026.

Cloud delivery is evident and neither limb this axis asks for is properly answered. The only residency statement located is in the privacy policy, which tells the reader that personal information will be transferred from their location to the company's facilities and servers in the United States. That is a country, stated once, in a document about personal information rather than about the medical records the platform processes, and it is the whole of what is published. No cloud provider is named, the security page referring only to verified enterprise-grade cloud infrastructure partners. No region or residency option is offered, nothing distinguishes processing location from storage location, and no tenancy or isolation model is described, so nothing states whether the platform is multi-tenant or how one firm's case files are separated from another's. For a product ingesting protected health information under business associate agreements, where data location is a compliance question rather than a preference, this is thinner than the product class warrants. Verified 2 September 2026.

A real artifact with a real request route, and no certification of any kind. The vendor states it has undergone an extensive third party security penetration test and that the report is available upon request, that it periodically undergoes external security testing, and that it has passed extensive security reviews at large law firms. Under the three tier test the artifact is a self serve request flow rather than absent, which is why this sits above the D recorded on peers with nothing at all. But the artifact is a penetration test report, which examines technical vulnerabilities at a point in time, and it is not an attestation over a control environment across a period. No SOC 2 of either type, no ISO 27001, no named auditor, no examination scope or date, and no trust centre were located. Independent guidance in this category names a current SOC 2 Type II report as a baseline check, and this vendor does not meet it. Held at C: better evidenced than nothing, and materially short of the peers at B and A which carry named attestations.

Two seals are displayed and neither is accompanied by a statement that would let a buyer test it. A SOC 2 badge and a HIPAA badge appear as images at the top of the security page. No SOC 2 type is stated, so it is not established whether this is Type I or Type II, and no auditor, certificate number, report date, coverage period, scope or trust services criteria appear anywhere. No route to obtain a report is offered, whether self-serve, on request or under non-disclosure, and there is no trust centre or portal. The HIPAA seal sits beside a carefully hedged sentence, that the company supports HIPAA compliance and signs business associate agreements upon client request, which is accurate framing since HIPAA is a regime rather than a certification, but it means the badge asserts more than the text does. No penetration testing or vulnerability assessment programme is described. What is published is a graphic; what this axis asks for is evidence. Verified 2 September 2026.

The existence of the chain is confirmed and contractually covered, and no party in it is named. The statement that business associate agreements are in place with the AI providers does two things at once: it establishes that third party model providers process customer data, which most vendors leave to inference, and it states that the relationship is governed by the instrument HIPAA requires for protected health information. That is more supply chain disclosure than any other record in this category makes, and it is the reason this grades at C rather than D. What is missing is identity: no provider, model family or version is named, no subprocessor list is published, and nothing states how many providers are involved or what each processes. A firm can tell its client that the downstream models are under BAA and cannot tell them who the models belong to. Compare Onspring at B, the only record on this index naming its provider outright.

Two providers are named openly in product copy, which is more than most records in this corpus manage and is done without hedging. The agent layer is stated to work through a partnership with Anthropic, and the live web search through a partnership with Perplexity, each named at the point where the capability is described rather than buried in a policy. That tells a buyer whose technology handles the medical record and whose handles the outbound literature search, which is the more useful half of this axis. The rest is absent. No model or version is identified for either provider, nothing states where inference runs or under what terms, nothing addresses what either provider may retain, and no commitment exists to notify customers when the provider set or the models change. The infrastructure layer is unnamed entirely, described only as verified enterprise-grade cloud infrastructure partners. The Perplexity relationship carries a disclosure question the published material does not reach, since a live outbound search from a platform holding protected health information raises the question of what is contained in the query. Verified 2 September 2026.

The vendor claims transparent pricing, the pages located route to a demo request, and independent material reports a specific published structure, which is a combination worth recording precisely. Vendor comparison pages state that transparent and competitive pricing keeps costs low, and the call to action beside that statement is Request a Demo, with a free trial offered separately. Independent material states that Legalyze publishes per month pricing tied to page volume starting around $500 per month for 2,000 pages, and describes this in contrast to peers using custom quotes, which suggests a published pricing page exists that was not reached in this pass. Graded C on that basis: the pricing model is discoverable at structure level, being per month against page volume, and no figure was located in vendor material. Recorded as a correction candidate in the upward direction, since locating a published pricing page would move this materially. A per page structure is also the most transparent unit in this category for a contingency firm, because case cost scales with record volume rather than being a flat unexplained fee.

Nothing about cost is published. No pricing page exists, no tier or plan is named, no unit of charge is identified, and no structure is described, whether per user, per case, per page, per record or by subscription. The only commercial route on the site is Book a Demo, which appears in the header, the footer and at the end of every section. Nothing states whether implementation, onboarding or training is charged separately, and no trial or free tier is offered. A buyer cannot learn from anything published whether this is priced per matter or per seat, let alone what it costs. No pricing row is written, because the rule that triggers one requires published structure and none exists here. Checked the home page, solutions page, practice areas page, security page, about page, terms of use and privacy policy on 2 September 2026. Verified 2 September 2026.

Practice scope is stated and input coverage is the most specific in the category, which for a record review product is the coverage question that matters. Practice areas named are personal injury, medical malpractice and workers compensation. Input coverage is enumerated rather than gestured at: Word documents, PDFs, images and spreadsheets accepted, with handwritten, scanned and electronic health records all stated to be processed. Naming handwritten records explicitly is the first instance on this index and it is the hardest extraction case in medical review, where provider notes are often the only record of a symptom. Held at B rather than A because nothing is characterised beyond the list: no statement of whether accuracy or throughput holds evenly across handwritten and electronic records, no jurisdictional scope, and no indication of practice depth in workers compensation as distinct from personal injury, which involves different record profiles and different statutory frameworks.

Practice coverage is enumerated properly rather than gestured at. Six areas each have their own section with tailored treatment: medical malpractice, birth injury, nursing home negligence, mass torts and product liability, motor vehicle accidents, and admiralty and maritime injury, with the maritime template described as the personal injury framework adapted for Jones Act and vessel incident work. That is a real statement of where the product is built to operate. Both sides of the bar are named explicitly and repeatedly, with the platform described as trusted by plaintiff and defense firms and as an independent platform serving all parties, which is unusual in this lane and is borne out by a roster spanning mass tort plaintiff firms and a large defense firm. What is missing is the boundary. No firm size band is stated, nothing addresses solo or small firm suitability against mass tort portfolios running to thousands of claimants, no jurisdiction is named although United States practice is implied throughout, and nothing states what the product does not support. Verified 2 September 2026.

Policy never, and the only unqualified one in this category. The quoted commitment carries no scope limiter: it is not bounded to third party models as on LinkSquares, to shared models as on Eve, or to a named product layer as on Exterro. It is repeated in the vendor's own product material as our AI models do not train on your data. It is reinforced by a second statement that business associate agreements are in place with the AI providers, which extends contractual coverage of protected health information to the downstream model layer rather than leaving that boundary undescribed, and no other record on this index makes that extension. Held at policy never rather than contractual never because both statements appear in product and partner marketing rather than in terms, a data processing agreement or any document a customer signs, and no customer facing agreement was located. Checked the Smokeball partner listing, the medical chronologies page, the comparison pages and the home page on 29 Aug 2026.

Checked the home page, solutions page, practice areas page, security page, terms of use in full and privacy policy in full on 2 September 2026. Nothing addresses whether customer content is used to train or improve models, in either direction. No clause names training, model training, machine learning or model improvement. Two provisions bear on it without reaching it, and a reader should see both. The privacy policy permits use of personal information to provide, administer, maintain, improve and analyse the services and to develop new features and services, and separately permits aggregation of personal information to improve and add features, with aggregated information shared with third parties or published. The website terms of use grant the company an irrevocable, royalty-free, worldwide, sublicensable licence over user content to reproduce, prepare derivative works of, incorporate into other works and otherwise exploit it, expressly limited to the purpose of including that content in the Site. Neither names training, so under the naming test neither moves the value. The gap is sharper here than on most records because no platform agreement or business associate agreement is published at all, so the document where a training position would sit does not exist publicly.

Not addressed. No retention period is published for uploaded medical records, generated chronologies or Case Chat queries, and nothing indicates whether a firm can delete data or require deletion at matter close. The vendor states that data is encrypted and stored on servers under its own control, which describes custody rather than duration. The gap sits oddly beside an otherwise strong data position: this vendor answers what it does with customer data more clearly than any peer in this category and says nothing about how long it keeps it. Checked the medical chronologies page, the comparison pages, the Smokeball partner listing and the site navigation on 29 Aug 2026.

Retention is acknowledged and no period is attached anywhere. The privacy policy states that personal information is kept for as long as necessary for the purposes it was collected for, unless a longer period is required or permitted by law, and adds that it may be retained for legitimate business interests including compliance, dispute resolution and record keeping, with deletion or anonymisation when no longer needed and isolation from further processing where deletion is impossible because of backup archives. No figure appears. The website terms point the other way and are quoted above: section 3.1 states that the company is not obligated to back up any user content and that content may be deleted at any time without prior notice, placing the burden of keeping copies on the user, and section 8 adds that terminating an account may involve deletion of associated user content from live databases. Nothing states how long chronologies, agent outputs, queries or uploaded records persist, and no retention setting is offered. Deletion is available on request by email. The absence matters more than usual because the material is medical records held under a business associate agreement that is not published.

Not addressed. No permission model, matter level access restriction or tenant segregation description was located. The Case Chat feature makes the question concrete: it answers plain English questions across a case file, and nothing states whether it is scoped to a single matter or can reach across a firm's uploaded estate, or whether answers respect the asking user's permissions. Integrations exist with Smokeball and CASEpeer, both of which carry their own permission models, and nothing states whether Legalyze inherits them on import. Checked the medical chronologies page, the integration pages, the comparison pages and the site navigation on 29 Aug 2026.

Checked the home page, solutions page, practice areas page, security page, terms of use and privacy policy on 2 September 2026. Nothing describes segregation between customers, between matters, or between users where a conflict requires it. The one access statement published is that granular access controls govern user roles and system-wide data visibility, which describes permissioning inside a customer account rather than a boundary between accounts, and system-wide data visibility as a phrase reads toward breadth rather than containment. Nothing states whether the platform is multi-tenant, how one firm's case files are isolated from another's, or whether a wall can be configured within a firm. The product does organise work into cases and folders, so the structure for separation exists, but nothing presents it as a security boundary. The gap is pointed in this practice area: mass tort work routinely involves firms on both sides of the same litigation, and Parambil markets to plaintiff and defense firms simultaneously as a neutral platform.

Not addressed. No government or law enforcement request clause, no commitment to notify a customer before producing their data, and no transparency report were located. The vendor holds claimant medical records including handwritten provider notes on servers under its own control, and business associate agreements with its AI providers govern permitted uses between those parties rather than constituting a public notice commitment to customers. Checked the medical chronologies page, the comparison pages, the Smokeball partner listing and the site navigation on 29 Aug 2026.

The position is stated expressly and it is the adverse one. Section 3.1 of the privacy policy opens with the quoted words, qualified only by the phrase unless required by the law, and then lists the circumstances, which include disclosure where required by law or in the good faith belief that it is necessary to comply with a legal obligation including national security or law enforcement requirements, to protect the company's rights or property, to prevent fraud, to act in urgent circumstances, or to protect against legal liability. Several of those limbs are the company's own determination rather than a compulsion. No notice commitment appears anywhere, no undertaking to seek confidential treatment or to give the customer an opportunity to intervene, and no transparency report was located. The website terms of use add nothing on the point. This is one of the clearer statements of the position in the corpus, and it applies to a platform holding plaintiffs' medical records.

Not addressed, and inapplicable in the usual sense with a live residue. The platform operates on the firm's own uploaded records rather than a published law corpus, so there is no external legal source to name, license or date. The residue is what the extraction models were trained on, and it is narrowed rather than answered by the vendor's own disclosures: customer data is stated never to be used for training, and business associate agreements exist with third party AI providers, which together indicate the models are supplied rather than built on customer records. Nothing states what those supplied models were trained on or whether any legal or clinical fine tuning corpus exists. Checked the medical chronologies page, the comparison pages and the Smokeball partner listing on 29 Aug 2026.

Checked the home page, solutions page, practice areas page, security page, terms of use and privacy policy on 2 September 2026. Parambil ships no corpus of its own and none is claimed. The material it works on is the customer's: medical records, billing records and case files uploaded by the firm, with the customer warranting its rights to that content under the website terms. The provenance risks this signal tracks therefore sit with the firm rather than the vendor. One dependency is worth recording because it is the exception. The live web search feature is described as giving real-time access to clinical guidelines and authoritative medical literature through a partnership with Perplexity, with every citation verified. That is third-party content entering the work product, and nothing published identifies which guideline bodies or literature sources are within scope, on what licence or terms that material is retrieved and reproduced, or what the verification of citations consists of.

Not addressed, and inapplicable on the facts. The product extracts and organises medical facts and answers questions about the case record. It produces no legal authority and no demand letter drafting was located in vendor material, so there is nothing for a citator to check. Recorded as a scope fact rather than a disclosure failure, consistent with the treatment on Tavrn and DigitalOwl, so a reader comparing this record against a research or drafting product does not mistake an inapplicable row for a gap. Checked the medical chronologies page, the home page and the comparison pages on 29 Aug 2026.

Checked the home page, solutions page, practice areas page and terms of use on 2 September 2026. No public material addresses subsequent history, treatment flags or good law checking, and none is claimed. The product does not retrieve primary law: it reads medical records and drafts from them, so a citator has nothing to operate on and the limb does not bite in its usual form. The analogous question does arise and is also unaddressed. The Drafting Agent generates complaints, demands, interrogatories and discovery requests, which are documents that cite law, and the live web search retrieves clinical guidelines, which are superseded and revised in much the way authority is. Nothing published states whether a retrieved guideline is checked for currency or whether a generated filing's legal citations are verified against anything.

Not addressed, with one genuine adjacent capability that is credited in the note and does not satisfy the signal. The product surfaces missing records, treatment gaps, prior injuries and inconsistencies, which tells a user where the record set itself is incomplete or self contradictory and is a real and useful form of flagging. It is a statement about the records, not about the model's confidence in what it extracted from them. Nothing published describes what happens when a handwritten page is illegible, whether a low confidence extraction is marked, or whether Case Chat will decline to answer a question the file does not support rather than answering from inference. The no human in the loop model makes this the load bearing question, because there is no reviewer positioned to catch a confidently wrong extraction before it reaches the attorney. Checked the medical chronologies page, the comparison pages and the home page on 29 Aug 2026.

Checked the home page, solutions page, practice areas page, security page and terms of use on 2 September 2026. Nothing describes what the system does when it cannot ground an answer, and no confidence or grounding indicator is described as shown to the user. Two adjacent behaviours are published and are recorded here because they are the closest the material comes without meeting the signal. The platform flags documents referenced within the record but missing from the file, and it visualises gaps in the medical record, which tells a user the evidence base is incomplete rather than telling them the model is uncertain. Company statements elsewhere describe claims being flagged when evidence is incomplete and conclusions re-evaluated as new records arrive. All of that concerns the completeness of the source material. Nothing addresses the agent that drafts a complaint or answers a query when the record does not support the conclusion sought.

None located, with the instrument named. General web searches combining the vendor and product names with court, order, sanction, fabricated citation and medical record terms returned nothing on 29 Aug 2026, and no named docket database or court record tracker was searched. Recorded as a statement about what this search found, not as a clearance. The exposure shape is a misstated or unsupported medical fact rather than a fabricated legal citation, since the product generates no legal authority, and the adverse finding to look for would be a court addressing a chronology that misrepresented the underlying records.

Searched the AI Hallucination Cases database maintained by Damien Charlotin, and reporting drawing on it, on 2 September 2026 on the product and corporate name Parambil and Parambil Corp. 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 structural note bounds it in an unusual direction. The tracker records fabricated legal citations in filings, and Parambil's primary output is a factual chronology drawn from medical records rather than legal authority, so its characteristic failure mode would be a misstated clinical fact, a wrong date or a missed record rather than an invented case. That failure would surface as an evidentiary dispute or an expert challenge rather than as a sanctions order, and no tracker records it. The Drafting Agent, which produces filings, is the part of the product this signal would reach in its usual form.

Not addressed. No named ethics opinion, no ABA Formal Opinion 512, no state bar guidance and no engagement with professional conduct rules was located, despite the vendor publishing comparison and ranking content aimed at attorneys evaluating platforms. That content engages performance and cost and not the professional duties attaching to reliance on the output. Sixth of seven records in this category at this value, with ProPlaintiff the only exception. Checked the home page, the blog material, the comparison pages and the site navigation on 29 Aug 2026.

Checked the home page, solutions page, practice areas page, security page, terms of use and privacy policy on 2 September 2026. No public material engages guidance from any professional body governing the platform's users. ABA Formal Opinion 512 is not named, no state bar opinion is cited, and nothing addresses the competence, supervision or candour duties of a lawyer relying on a machine-generated chronology or a machine-drafted pleading. The company does engage an external regime, but a different one: it supports HIPAA compliance and signs business associate agreements, which governs the handling of protected health information rather than the professional obligations of the lawyers using the output. The one governance-adjacent structure mentioned, an advisory board of experts from institutions and academia, is described as helping build trusted AI systems rather than as engaging professional rules.

Savings claims only, and unusually the savings are quantified as a rate rather than a percentage. Published: review speeded up by ten times, ten pages per minute, a thousand page file reduced from fifteen to twenty hours to roughly two, and a user report of hundreds of man hours of document review saved. Those describe the firm's own labour. Nothing appears on the claimant's side: no position on whether platform cost is a case expense or firm overhead in contingency work, no disclosure guidance, and no record a firm could produce showing what portion of a chronology was machine generated. Worth noting that per page pricing tied to record volume, which independent material attributes to this vendor, is the structure that would make a case cost most legible to a client, and the vendor does not connect the two. Checked the medical chronologies page, the comparison pages and independent review material on 29 Aug 2026.

Savings are quantified and the bill is not addressed. The home page publishes more than 20,000 hours saved and 95 per cent faster than traditional methods, and frames the commercial case explicitly around case economics, arguing that time is money and that improving the internal rate of return on cases will become a greater priority. Nothing addresses how AI-assisted work is recorded, billed or disclosed to a client, and no per matter record of AI-assisted work was located. The gap has a particular shape in this practice area. The buyers are contingency-fee plaintiff firms, so the hours saved are not billed to a client in the usual sense; the displaced cost is case expense, the paralegals, legal nurse consultants and retired physicians the platform is explicitly compared against, and that expense is customarily charged to the client's recovery. Nothing published addresses whether or how platform cost passes through as a case expense, which is the question a plaintiff's client would have.

On request, reached through a fourth distinct route in this category. The vendor states that an extensive third party security penetration test has been performed and that the report is available upon request, which is a named artifact with a stated request path. Alongside it, two statements a firm could actually use in a client questionnaire: business associate agreements are in place with the AI providers, and the platform has passed extensive security reviews at large law firms. Held at on request rather than higher because nothing is open and the pack is incomplete: no subprocessor list, no named model provider, no SOC 2 or other attestation, no customer facing BAA described, and no trust centre. Note the category now shows four routes to this value: EvenUp via a Trust Center portal, Tavrn via an offered BAA, DigitalOwl via both, and Legalyze via a pentest report on request.

Checked the home page, solutions page, security page, terms of use in full and privacy policy in full on 2 September 2026. No subprocessor list exists in any form. The security page refers to verified enterprise-grade cloud infrastructure partners and to regularly auditing vendors without identifying a single entity, and the privacy policy lists categories only, covering hosting providers, cloud service providers, other information technology services providers, event management, email and newsletter services, advertising and marketing services and web analytics. Two entities are named elsewhere on the site, Anthropic and Perplexity, but they appear in product copy describing capabilities rather than in any list a firm could forward, and neither is presented as a subprocessor. No data processing agreement is published and no forwardable client-facing disclosure material was located. The one document a firm could obtain is a business associate agreement, available on request, which addresses protected health information rather than the disclosure a client's outside counsel guidelines would ask for.

Partial record, with the strongest source limb in the category applied across both output types. Every medical event opens in one click to the exact page in the original record, and Case Chat answers are individually cited to their location in the source documents, so a party can show where any assertion came from and an opposing party can check it. Independent material identifies this as the detail that matters when a timeline is challenged. The gap detection capability adds something no peer offers here: surfacing missing records and treatment gaps means a party can also see what the record set does not contain, which speaks to completeness as well as provenance. The familiar two limbs remain absent: nothing indicates that output records which model produced it, and no human verification record is captured, which under the no human in the loop model means there is no verification event to record at all unless the attorney creates one outside the platform.

Checked the home page and its full feature list, the solutions page, practice areas page, security page and terms of use on 2 September 2026. Nothing addresses judicial standing orders, disclosure of AI use, or certification that output was verified by a person. The platform produces artifacts close to what such a record would require and does not describe them as one: every chronology entry hyperlinks to its source page, missing and referenced-but-absent documents are flagged, and complete case packages export with notes, ledgers and data tables. Nothing captures which model or agent produced a given passage, when, or who reviewed it. The gap is sharpest at the drafting layer, since the Drafting Agent generates complaints and discovery requests that are filed with courts, and the chronologies are marketed as courtroom-ready, so the artifacts most likely to attract a disclosure question are the ones with no disclosure record attached.