M

Mary Technology

Remove the models and nothing remains to sell. The product is a reading and extraction engine. It takes case files as they arrive, including poor scans and old formats, reads every page, and extracts each fact rather than storing documents as embeddings. Each fact is tagged with people, dates and events, linked to its source page and given a status of supported, contradicted or unconfirmed. The engine also surfaces contradictions, unexplained periods and material referenced elsewhere but missing from the production.

There is no document management system, review platform or matter management layer underneath that would be licensed alone. Documents are uploaded or pulled from systems the firm already has, and Mary adds the structured factual record and everything built from it. The vendor's own architectural claim contrasts extraction of each fact, with a traceable link to source, with approaches based on compression.

The verification design is documented in detail, and none of it is measured. Grounding works fact by fact rather than answer by answer. Every extracted fact traces to its source page, plain language questions return answers with one click citations to the exact facts and pages, and the original source can be opened against any entry. Each entry also carries a published status of supported, contradicted or unconfirmed, and competing accounts are kept separate rather than merged into one answer, with contradictions and unexplained periods left visible.

Coverage is reported too: what was processed, what informed an entry, what was reviewed but unused and what remains unprocessed. That combination addresses hallucination by design. No accuracy figure, evaluation, test set or error rate is published. The nearest item is a preference study in which eight of nine litigators preferred Mary's output in a matter review at an unnamed global firm, undated and with no method, which measures preference rather than accuracy.

The control structure is published in full, and there is no autonomous mode. Proposed updates stay pending until a person approves them, stated as a product mechanism rather than a promise. The published workflow makes human review its own numbered stage: the team inspects sources, reviews contradictions and gaps, sets status and approves changes before anything becomes work product. What the system does alone is stated: reading every page and structuring people, dates, events and facts.

Its limit is stated plainly. Mary does not decide the case; it shows the record, including what is missing and what does not add up, so the lawyer can, and legal judgment stays with the lawyer. The review surfaces are the product's core, since every entry can be opened against its source and its status set by hand. Clause 11(d) of the agreement requires all outputs to be reviewed and approved by a qualified legal professional before any action is taken.

Nothing describes what happens when an entry is wrong, or how a firm would detect a document set that has been misread throughout.

Aggregate scale claims and unattributed quotations stand in for a named deployment. The published figures are counts rather than outcomes: more than four thousand lawyers and more than one hundred and fifty firms, neither dated nor tied to any organization. One comparative claim comes closer: eight of nine litigators preferred Mary's output in a matter review at a global law firm. The firm is unnamed, and there is no date, matter description or method for assessing preference.

Testimonials on the self serve site carry no name, role or firm, including one saying work that took half a day to a full day now takes under fifteen minutes. No customer is named, and no dated deployment or measured outcome is published. A customer stories page sits in the navigation. Trade press names Australian firms as customers, but that is third party material.

Separation is published at the level this buyer needs, and privilege is not addressed. The security commitment is specific and stated on the product page. Customer data is isolated by matter, with separate access and data boundaries for each, encrypted in transit and at rest, with access following the firm's own permissions and single sign on. Documents, matter content, prompts and outputs are never used to train AI models.

Isolation by matter is what a litigation practice with screened matters needs. Clause 19 of the terms carries part of this contractually, describing tenanted AI services under which confidential client information is not accessible for model training or to unauthorized personnel. Three gaps remain. No privilege or work product treatment appears, on a product built to hold the factual record of contested matters. No model provider is named, so nothing addresses what any third party retains. And the isolation claim has no supporting documentation, since the security page defers detail to a trust center.

A real published position, stated in the agreement and repeated across the site. Clause 11(d) says the software helps with legal tasks but is no substitute for professional legal advice, analysis or judgment. All outputs and recommendations must be reviewed and approved by a qualified legal professional before any action is taken or decision made, and the vendor disclaims liability for actions taken without legal review.

It also declines to guarantee the accuracy, completeness or suitability of output for any purpose. Every page carries the same idea: lawyers remain responsible, and legal AI should make its work easy to check before they sign. The product material repeats that Mary does not decide the case and that legal judgment stays with the lawyer. That is a coherent account of where the product stops. The professional layer is absent: no bar association, rule of professional conduct or ethics opinion is named, no jurisdiction is identified, and nothing addresses a firm's supervision or competence duties.

No governance position is published. There is no responsible AI page, principles statement, accountable owner for model behavior, evaluation regime before release, management system or certification such as ISO 42001, and nothing addresses uneven output. The vendor publishes an editorial position on verification: that verification in legal AI is a design problem, and that legal work should be easy to check before a lawyer signs.

That position is held across the product, but it is a design philosophy rather than a governance framework, and it names no reviewer, criteria or results. A system that assigns each fact a status of supported, contradicted or unconfirmed makes a judgment on every entry. Nothing published says how that classifier was evaluated or how it behaves across document types, jurisdictions or the poor scans the product advertises handling.

Access and testing are addressed; retention, deletion and suppliers are not. Published: client data encrypted throughout its lifecycle, annual penetration tests and continuous vulnerability assessments, background checks and access reviews for staff, mobile device management, a cybersecurity training program, data boundaries at matter level, and access governed by the firm's own permissions with single sign on. Most of the rest is missing.

No retention period is stated for documents, extracted facts, prompts or outputs, only that data is kept as long as necessary and securely purged in line with best practices. No deletion or return of data commitment for customer material is published. No subprocessor list exists; the only third parties named are Stripe for payments and Meta among advertising networks, neither of which touches matter content. Clause 5 of the terms says the privacy policy covers the vendor's obligations on security breaches, but the privacy policy, last updated 22 August 2026, has no breach notification provision.

The allocation of loss is published, readable before signing, and leaves almost nothing behind it. Clause 12(a) limits total liability, however arising and under any theory, to resupplying the services, a remedy of performing again with no monetary ceiling behind it rather than a capped one. Clause 12(b) excludes direct, indirect, incidental, special, consequential and exemplary damages alike. Clause 11(c) provides everything as is and as available without warranty of any kind, with a long list of losses disclaimed, including loss of data and unauthorized access to records.

Clause 14 runs the indemnity one way from the user, covering their content and any use of the service. There is no indemnity from the vendor, no warranty on output, no service level commitment and no insurance position. One real floor exists: clause 11(a) preserves the guarantees of the Australian Consumer Law that cannot be excluded, a real statutory backstop for an Australian buyer that does not reach a US litigation customer at all.

Integration is described by category throughout and no counterparty is named anywhere. The product material states that a team can upload or connect the documents it already has, shows a document system and a supported AI tool feeding the same factual record, and offers to let approved AI tools and drafting workflows build from that record. The privacy policy confirms the mechanism, recording that data is collected from third party applications connected to the platform, such as document management systems, when authorized by the user, and shared back with those integration partners.

That is a real and reciprocal connection, and it is described only as a class. No document management system, review platform, matter management system or ediscovery tool is named, no API or developer documentation is published, and nothing states what data moves, in which direction or on what trigger. For a product whose value proposition is being the single factual record several tools build from, the absence of a named counterparty is the most visible gap.

The isolation model is stated precisely, and the region is not stated at all. Customer data is described as isolated by matter, with separate access and data boundaries for each, and clause 19 of the terms says the platform runs on tenanted AI services under which confidential client information is not accessible to unauthorized personnel. Boundaries at matter level are finer than separation by tenant and imply separation between customers.

No region is named for storage or processing, no cloud provider is identified, and no residency commitment appears in the terms, the privacy policy or the security page. That is a live question for a product run through Australian and United States entities, with an application host on a country specific subdomain and a trust center on an Australian one. No on premises or self hosted option is mentioned.

A real trust portal and a claim of recognized controls, with no standard named in text. The security page says the program is built around internationally recognized controls and independently tested safeguards, and links to a trust center on a third party compliance platform. No standard or assessor is named in the page text: no SOC 2 of either type, ISO certification, examination period, scope statement, certificate or report summary appears.

Annual penetration testing is asserted without naming the testing firm or publishing results. Nothing published says whether the trust center releases documents on an email address or requires a sales conversation.

The vendor describes its AI architecture and identifies nothing underneath it. The terms speak of a platform powered by industry leading AI processes and of tenanted AI services, and the product material contrasts extraction of each fact, linked to source, with systems that store documents as embeddings in vector databases. No model, version, provider or hosting arrangement is named, and no commitment to notify customers of changes is published.

The only third parties named are Stripe as payment processor and Meta among advertising networks, neither of which processes matter content. The question matters here because the material is the whole factual record of contested litigation, including privileged work product. The tenanted AI services claim in clause 19 cannot be traced to a named provider.

No pricing information is published at any level, including the unit of charge. The navigation carries only security, resources and company, and every route on the site is a demo request. The terms describe payment without a charging model. Clause 7 requires a subscription and a fee for a subscription period, clause 8 lets the vendor vary the fee at any time with effect from the end of the current subscription, and clause 9 offers refunds only under the Australian Consumer Law or at the manager's absolute discretion.

Nothing says whether the subscription is charged per user, matter, document or firm, and no rate, band, tier, term length or implementation cost appears. A self serve entry point is advertised as free to start, with nothing published about what it includes or what the paid tier costs.

Four practice areas are described with substance, and the boundary is left open. Commercial litigation is framed as rebuilding the sequence across contracts, emails, board materials, witness accounts and the production. Family law is framed as tracing finances, communications and parenting events across disclosures, bank statements, messages and filings. Personal injury is framed as rebuilding treatment, symptoms, diagnoses, work capacity and prior history across the medical file.

In house disputes and investigations are framed as keeping one reviewable record across investigations, business teams, outside counsel and changing tools. Each names the material that practice generates. Partners, associates and litigation support are named as working from the same record. No jurisdiction is stated, no firm size band is given, nothing says which matter types or document sets the product handles poorly, and supported languages are not addressed.