C

Caseway

AI Centrality

How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.

Every product a customer pays for is model work, and the company was built as one. Casey is a generative research assistant that answers legal questions in conversation and returns cited answers grounded in Canadian and United States case law, with citation-linked analysis of uploaded documents. CaseForm auto-completes court and regulatory forms by extracting required fields from source documents and prior filings and applying jurisdiction-specific logic before submission, which is extraction and rule application rather than a template. Synthium is a retrieval and validation layer built for complex regulated work. The agreement is framed around the same premise, defining Output as material generated by the platform from Input, acknowledging in its warranty section that the platform uses probabilistic prediction and pattern analysis with inherent limitations, and defining third-party large language models as a category of dependency. One non-model surface exists and is described honestly rather than credited: CaseLite is a keyword and citation search interface over the same corpus, offered free and without an account, and the vendor positions it as the direct search front end from which a user upgrades to Casey when AI answers are wanted. That is an on-ramp to the paid model product rather than a separate business, and it does not survive as a commercial proposition on its own. Terms of service of 7 February 2026, pricing page and methodology page read 7 September 2026.

Citation Accuracy and Hallucination Disclosure

Whether the vendor publishes measured accuracy on citations and assertions, grounds output to primary sources, and says plainly what its system does when it does not know. Legal has a documented public record of fabricated citations reaching filed briefs, so an untested claim of accuracy is not evidence.

Grounding is documented and verifiable by the reader, and nothing is measured. What is published: answers are cited to real court decisions and the vendor states on its methodology page that the product does not generate opinion text or fabricate citations, and that opening a result shows structured metadata and documents tied to a real source identifier from the upstream publisher. The same corpus is exposed free through CaseLite without an account, and the vendor invites a reader to run a known citation and see how the record appears, which makes the grounding claim independently testable in a way most records on this axis are not. Limitations are published rather than implied: not every court, tribunal or historical period is equally represented; absence of a case does not mean it does not exist; metadata quality depends on the upstream source and older records may lack citations or dockets; and critical citations should be confirmed against the official reporter or court website when filing. The agreement carries the same instruction, telling users always to confirm that the sources and citations provided in any Output accurately support the statements made, and warning that outputs are generated as is and may be incorrect or incomplete. What is absent is measurement. No accuracy figure, error rate, hallucination rate, test set, benchmark or evaluation result is published for the product, so a reader can establish that the answers point at real documents and cannot establish how often the reasoning over them is right. Methodology page, terms of service and pricing page read 7 September 2026.

Autonomy and Oversight Model

What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.

Responsibility is allocated to the user in detail and the system's own behaviour is not described. The allocation is thorough and appears in more than one place: the agreement states that the user is solely responsible for reviewing and verifying the accuracy and appropriateness of any Output before using it in any legal matter, that outputs are provided as is and may be incorrect or incomplete, that the user should not rely on Output without first obtaining advice from a licensed attorney, and that the vendor disclaims liability for actions taken or not taken on the basis of Output. The methodology page repeats that the search product is a research aid rather than legal advice. What none of that supplies is an oversight model. Nothing published describes what runs without a person, at what point the system acts on its own, what a review step looks like inside the product, or whether any human checkpoint exists. The question is live rather than theoretical for one product in the family: CaseForm is described as applying jurisdiction-specific logic and validation before submission of court and regulatory forms, which places automated work directly against a filing deadline, and nothing states who presses submit, what the validation checks, or what happens when it passes something wrong. No confidence signal, escalation path, audit view or reviewer workflow is described for any product. Disclaiming responsibility is not the same as describing oversight, and only the first is published here. Terms of service, methodology and pricing pages read 7 September 2026.

Operational and Outcome Evidence

Named, dated evidence that the product works in production at real firms or legal departments. Case studies with figures and identified customers count. Unattributed testimonials and launch announcements do not.

Institutional association is evidenced and customer outcomes are not, on the surfaces read. What is established: a research collaboration with the University of British Columbia, reported in the legal trade press and described by the vendor in its own posts as anchoring AI systems in real court decisions; a named integration between CaseForm and MyCase, a legal practice management product; a free public search surface that any reader can exercise without an account; and sustained trade coverage in Canadian Lawyer, CBC and regional technology press across 2024 to 2026. What is not established is outcome evidence of the kind this axis asks for. No customer is named on any surface read, no deployment is quantified, no time or cost saving is measured, and no adoption figure, user count or matter volume is published. The vendor's own methodology page cuts against unmeasured scale claims, telling readers to treat any size claim as point in time and to verify current availability inside the product rather than relying on marketing adjectives, which is candid and also leaves the record without a number. Recorded so the grade is read correctly: a Success Stories page and a Testimonials page both exist in the site navigation and neither was opened on this channel, so this row records what was established from the surfaces read and those two pages are the rebuttal route on it. Site navigation, methodology page and trade coverage checked 7 September 2026.

Privilege and Confidentiality Posture

How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.

A real contractual confidentiality regime sits beneath an express and unusually blunt denial that anything entered is privileged. The regime is substantive and it is in the agreement rather than on a policy page. Clause 7 makes customer data the customer's confidential information, binds each party to use the other's confidential information only for the purposes of the agreement, restricts disclosure to employees and contractors who need to know and are bound by equivalent obligations, and sets a standard of the same care each party uses for its own confidential information and in no event less than reasonable care. Around it sit a zero-retention default for customer data, automatic deletion of uploaded documents within twenty-four hours, chat history saved only where the user opts in and deletable at any time, and a prohibition on training public models on customer input. Against all of that, clause 3 states in terms that because the user is not receiving legal advice, information input into the platform is not protected by any attorney-client privilege or work product doctrine, and instructs the user not to input confidential or sensitive information unless entitled to and accepting that it will not be legally privileged. So the limb this axis exists for is engaged by name and answered against the buyer, which is why this is not the top grade, and the surrounding regime is genuinely strong, which is why it is not lower. Nothing addresses segregation between matters or between users inside one customer. Terms of service and privacy policy of 9 March 2026 read in full 7 September 2026.

UPL and Professional Responsibility Posture

Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point.

The boundary is drawn repeatedly and in the agreement, and no governing instrument is named. What is published: the platform is not a law firm and does not provide legal advice; use of it creates no attorney-client relationship; Output is general information generated by AI and not a substitute for professional legal counsel; the user agrees not to rely on Output as legal advice for any course of action without first obtaining advice from a duly licensed attorney in the relevant jurisdiction; and the vendor disclaims all responsibility for actions taken or not taken on that basis. The warranty section restates it, and the methodology page repeats that the product is a research aid and not legal advice. A jurisdictional caveat is included, stating that no guarantee is made that the platform is appropriate or available for use in every jurisdiction and that a user accessing it where it imposes additional obligations does so at their own risk. That last point matters because the product is sold to members of the public as well as to legal professionals, which is where the unauthorised practice question actually bites, and the vendor addresses it with disclaimers rather than by engaging the prohibition. What is absent is any named authority. No provincial law society rule, no statutory prohibition on unauthorised practice, no bar guidance and no court rule appears anywhere, in a market where the governing rules are made province by province. Terms of service and methodology page read in full 7 September 2026.

AI Governance and Bias Disclosure

Published governance over model behaviour: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.

Limitations are acknowledged candidly and no governance apparatus stands behind them. The acknowledgements are real and specific. The agreement states that the platform's use of AI involves probabilistic predictions and pattern analysis with inherent limitations, and that the vendor does not guarantee that Output will be correct, complete or suitable, that defects will be corrected, or that any particular legal result will follow. The methodology page publishes five named limitations of the corpus, including uneven representation across courts and historical periods, dependence on upstream metadata quality, and the warning that absence of a case does not mean it does not exist. The privacy policy commits that legal content is not used for model training, profiling or advertising. Taken together that is an honest account of what the system cannot do. What is missing is everything that would make it governance rather than disclaimer. No AI governance policy, framework or standard is published or claimed, and no certification such as ISO 42001 is asserted. Nobody is named as accountable for the system's behaviour. No pre-release evaluation, red-team exercise, benchmark or monitoring practice is described. And no bias, fairness or representativeness disclosure exists at any level, which is worth naming for a product whose corpus is judicial decisions about identifiable parties and whose own methodology page concedes that representation across courts and periods is uneven. Terms of service, methodology page and privacy policy read 7 September 2026.

AI Safety and Data Stewardship

Retention, deletion, access control, and what happens to prompts and documents after they are processed. Whether the vendor states its subprocessors and its incident practice, or leaves the buyer to assume.

Substantive published policy on the data questions, with the supply chain and incident practice absent. What is published, and much of it contractually rather than on a policy page: a zero-retention default, stated in clause 7 of the agreement, under which session content, uploaded documents, search queries and prompts are not stored except temporarily as needed to generate Output unless the user chooses to save them; automatic deletion of uploaded documents and context files within twenty-four hours, with a statement that they are not permanently stored; chat histories saved only on opt-in and deletable at any time; no retention on the vendor's servers of sensitive, case or matter data processed by the form automation product; a commitment not to train public AI models on customer input and, in the privacy policy, not to use legal content for model training at all; no sale of personal data and no profiling for advertising; and a stated minimum-data approach under which a category not operationally needed is not kept. Security is described in general terms: encryption controls for data in transit and at rest, access controls and permission boundaries, and service hardening, monitoring and abuse protections. Two gaps are named. No subprocessor or service provider is identified anywhere, the privacy policy referring only to contracted service providers acting on instructions. And no incident or breach notification practice is published, no notification timeline is committed and no security contact route is given. The agreement also states plainly that the vendor is not HIPAA or PCI-DSS certified and instructs users not to process protected health or payment card data, which is an honest negative disclosure rather than a gap. Terms of service and privacy policy read in full 7 September 2026.

AI Liability and Recourse

What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.

A complete two-way allocation, published and readable before signing, including something most records on this axis do not have: an indemnity running to the customer. Clause 12.1 commits the vendor to defend the customer against a third-party claim that the platform as provided and used in accordance with the terms infringes a valid copyright, patent or trademark or misappropriates a trade secret, and to indemnify against damages, costs and reasonable fees finally awarded or included in an approved settlement, with a remedy ladder of modifying the platform, obtaining a licence, or terminating with a pro-rated refund. Liability is capped mutually at the greater of twelve months of fees paid or CAD 500, with a stated floor rather than an open figure, and carve-outs for fraud, gross negligence and wilful misconduct capped at twice that. Termination for convenience by the vendor carries thirty days notice and a pro-rated refund of prepaid fees. What holds it below the top grade is the substance of what is not warranted. The platform and all content are provided as is and as available, with all implied warranties disclaimed and an express statement that the accuracy or reliability of Output is not warranted, so the product's core function carries no promise. No service level is committed, no insurance position is published, and the cap's floor of CAD 500 is nominal against the exposure a wrong fact in a filed document creates. Liability for third-party AI model failure is expressly disclaimed under clause 8. Terms of service read in full 7 September 2026.

Practice Systems Integration Depth

How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.

One named practice system and an API, with no documentation read and no connector inventory established. What is established from surfaces read: an integration between the form automation product and MyCase, a legal practice management platform, published on its own path in the site navigation; API access stated for the research assistant, described on the pricing page as available for research and document workflows; and API or embedded user interface options offered for the form product for teams and integrations. For the enterprise data hub the vendor describes custom pipelines, integrations and workflows tailored to a customer's stack, which is a services statement rather than a named connector. What is absent is depth and evidence of it. No API or developer documentation was located, no authentication or scope model is described, no document management system, email client, word processor or e-signature product is named, and nothing states what moves in which direction across the MyCase boundary or how permissions travel with it. Recorded so the grade is read correctly: an Integrations and Partnership page exists in the site navigation and was not opened on this channel, so this row records the integration surface as established in outline and not in depth, and that page is the rebuttal route. Pricing page, site navigation and terms of service checked 7 September 2026.

Deployment Model and Data Residency

Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.

Neither limb is stated for the product this record turns on, and what is published belongs to a different product in the family. For the research assistant, nothing states where data is stored or processed, no country or region is named, no residency option is offered, no tenancy model is described and no cloud provider is identified. The privacy policy addresses what is retained and for how long without ever saying where, and the agreement's only geographic content is a governing law clause and an arbitration seat, which are a forum rather than a data location. What does exist is published for the enterprise data hub: secure deployments with data residency options and compliance support, and deployment on premises or in an isolated cloud. That is a genuine deployment and residency disclosure and it is described here rather than credited across, because the product it attaches to is the enterprise data platform and not the legal research assistant that a lawyer buys at the published seat price. Credit follows scope, and no statement connects the two: nothing says the research assistant can be deployed on premises, in an isolated tenancy, or in a nominated region. A buyer of the research product can establish that the company knows how to offer these things to enterprise customers and cannot establish what applies to their own matter data. Pricing page, privacy policy and terms of service read 7 September 2026.

Security Certifications and Trust Center

Independent attestation a buyer can pull without a sales call: SOC 2, ISO 27001, penetration test summaries, a trust center with current reports and named scope rather than a badge image.

No certification is claimed, none is held so far as anything published shows, and no trust centre exists. There is no SOC 2, no ISO 27001, no independent audit, no named auditor, no report of any kind, no penetration test result, no compliance page and no portal from which documentation could be requested. The site navigation carries no security or trust entry at all, its legal section listing only the privacy policy and the terms of service, so this is established from the vendor's own page inventory rather than inferred from a failed search. Two things are recorded so the floor is read correctly rather than as an accusation of concealment. First, the vendor makes an honest negative disclosure that most records at this level do not: the agreement states plainly that it is not HIPAA or PCI-DSS certified and instructs users not to process protected health information or payment card data through the platform, which tells a buyer where the boundary is instead of leaving a badge to imply otherwise. There are no unsupported badges anywhere on the estate. Second, a document called a Statement of Confidentiality is referred to in clause 7 of the agreement as providing more technical and legal information on how customer data is protected; no location is given for it, it appears nowhere in the navigation, and it was not located. The general security controls described in the privacy policy are credited on the stewardship axis and are not counted again here. Site navigation, terms of service and privacy policy checked 7 September 2026.

Model Supply Chain Disclosure

Which models sit underneath, whose they are, where they run, and whether the vendor commits to telling customers when that changes. A legal buyer inherits every dependency it cannot see.

The supply chain is acknowledged to exist and not one participant in it is named. Clause 8 of the agreement is explicit that the platform may rely on or incorporate third-party services including third-party large language models or data sources, that the AI engine may rely on underlying models or APIs provided by third-party AI vendors, and that some customer input may be processed by third-party AI engines under the vendor's contract with those providers. So a reader is told plainly that outside models see their material. No model is named, no model family is named, no provider is named, no inference host is named and no cloud provider is named, on any surface read including the privacy policy, which refers only to contracted service providers acting on instructions. Change notification runs in the opposite direction to what this axis asks: the same clause reserves the right to replace, update or disable third-party services at any time and states that the unavailability or modification of any such service, expressly including a third-party AI model being discontinued or changed, does not entitle the customer to any refund or relief. The vendor does commit that third-party processing will be subject to appropriate confidentiality and data protection obligations, which is a flow-down promise rather than a disclosure, and it disclaims liability for the acts or omissions of those providers. Terms of service and privacy policy read in full 7 September 2026.

Commercial Transparency

Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.

Published figures for the priced products, a stated unit of charge, a free tier reachable without an account, and the commercial mechanics in the agreement rather than withheld. The research assistant is published at CA$49.00 per month with a monthly and yearly toggle; the form automation product at $49 per seat per month with its own purchase path; and the enterprise data hub as a contact route. The unit of charge differs between the two priced products and is stated for each, per user per month against per seat per month. Free access is genuine rather than a trial: the search product is offered with no account required, and the vendor frames the upgrade path from it to the AI assistant explicitly. Two discount programmes are published with eligibility checks, fifty per cent for students and fifteen per cent for veterans. The mechanics come from the agreement: fees stated in Canadian dollars and exclusive of taxes, monthly or annual billing by card or invoice, interest on late payment at 1.5 per cent per month, fees non-cancellable and non-refundable with no pro-rata on mid-term cancellation, a thirty day window to dispute a charge, and a pro-rated refund where the vendor terminates for convenience on thirty days notice. Two gaps are named rather than smoothed. No usage limits, quotas or overage terms are published for any tier. And the assistant's own feature list refers to uploads for premium users while only one price is shown for it, so a buyer can read that a premium tier exists and cannot find what it costs. Pricing page and terms of service read 7 September 2026.

Firm and Practice Coverage

Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.

Coverage is stated by jurisdiction with named collections, and the vendor publishes what it does not cover, which is rarer than what it does. What is published: case law for the United States, federal and state, and for Canada, federal, provincial and territorial; primary law browsing for four named United States jurisdictions, being California, Florida, New York and Texas, plus Canadian federal, provincial and territorial collections as released; records carrying party names, courts, decision dates, docket or court-file identifiers and citations where present in the source. Update practice is described rather than asserted, with ongoing ingest as source systems publish and an express statement that depth and lag vary by court and jurisdiction and that recent filings may appear after a delay. The limitations are the strongest part: not every court, tribunal or historical period is equally represented; absence of a case does not mean it does not exist; older records may lack citations, dockets or complete party strings; and primary-law browse outside the listed jurisdictions may be incomplete or unavailable. Buyers are named across law firms, enterprises and individuals, with legal sitting alongside defence, manufacturing, energy, mining and utilities as addressed sectors, and a practice areas surface exists. What keeps this off the top grade is that no per-court depth, date range or historical start point is given for any collection, so a reader knows the boundaries are uneven without being able to locate them. Methodology page, pricing page and site navigation read 7 September 2026.