LEGALFLY vs Noxtua: how they compare in 2026

The models are the product. Every surface sold is a generative or machine learning capability: research agents, contracting agents, compliance agents and a studio for building custom agents. The anonymisation layer that differentiates the product is itself a fine tuned model rather than a rules engine. Remove the models and nothing remains.

The artificial intelligence is the product, and further down the stack than any other record on this index. The vendor states it is built as a complete system with control over infrastructure, model, data and interface, running proprietary models trained on licensed legal data rather than calling third party foundation models. It also publishes its own legal embedding model, Noxtua Voyage Embed, built with Voyage AI and dejure.org. Remove the models and nothing remains.

Accuracy is asserted and the grounding behind it is not documented. Vendor material states that answers are grounded in verified case law and legislation across more than 130 jurisdictions and that anonymisation preserves full legal and contextual accuracy of every review, which is an accuracy claim with nothing behind it. Searched the site, the six product pages, the Knowledge platform page, the security page and the resources index on 29 Aug 2026 and located no description of the retrieval method, no statement of how output links back to primary sources a reader can open, no accuracy figure, no hallucination rate and no evaluation. Two jurisdiction counts appear on the same home page, 110 plus in one place and 130 plus in another, which is the kind of inconsistency that matters on an accuracy axis.

Grounding is real and documented, with published comparative measurement on one component, short of accuracy figures for the product itself. The retrieval method is described in unusual detail: the agent analyses a query, develops a solution plan, and researches systematically across licensed publisher content, with version 5 generating a knowledge graph per query that surfaces relevant sources and maps how they connect. Output carries verifiable source references into a licensed corpus a reader can open. Uniquely on this index the vendor publishes measured figures for its own retrieval model, stating Noxtua Voyage Embed outperforms a named OpenAI embedding model on legal text benchmarks with 1.7 times better search accuracy and 2.2 times better ranking quality at three times lower dimensionality. That is retrieval quality rather than answer accuracy. Not located as of 29 Aug 2026: an accuracy or hallucination rate for generated output, a test set for it, or any abstention behaviour.

A real published commitment with described control surfaces, short of the full structure. Agent Studio documentation states that workflows trigger automatically from email, Slack, Teams or manually, that conditional logic routes work and handles follow ups, that the system escalates when needed, and that approval steps keep legal in control without bottlenecking every decision. That is an oversight mechanism described at the point where an agent acts, and the escalation and approval language is more specific than most of this market publishes. Not located as of 29 Aug 2026: the threshold at which an agent escalates rather than proceeds, who configures it, and what the vendor commits to when an agent is wrong.

Autonomy is claimed and oversight is asserted without a mechanism. The vendor describes agentic AI that independently develops a solution plan and executes systematic research, and states results are traceable at every stage, which is transparency of process rather than a control structure. Searched the site, the three product pages, the security page and the press releases on 29 Aug 2026 and located no description of what the system does on its own versus what a lawyer must approve, no review surface, no threshold at which the agent stops, and no statement of what happens after an output is wrong. Traceability shows a reader what happened; it does not establish who decides.

Real deployment evidence with substance, short of measurement. Three named customer stories are published, ECS on streamlining legal review across departments, PIF Group on speed, consistency and quality, and Wealins of Foyer Group on broader market coverage and reduced external legal spend, alongside a customer logo wall and named enterprise references reported in funding coverage including Slaughter and May and Allianz. Not located as of 29 Aug 2026: figures for what changed at any named customer, dates, or a method a reader could assess. The reduced external legal spend claim carries no number.

Customer logos and an unattributed selected clients strip stand in for evidence. A customer stories page exists in the navigation, and a named endorsement appears from Dr Markus Kaulartz, partner at CMS, though CMS is also a co-initiator and investor rather than an arm's length customer, which is disclosed on the page and weakens it as independent evidence. Corpus scale figures are published, 130 million plus searchable documents and 8.5 million plus court decisions, but those describe the database rather than a deployment. Searched the site, the customer stories entry point and the press centre on 29 Aug 2026 and located no named deployment with figures, dates and an assessable method.

Substantive published commitments built around an unusual architectural claim, short of the full picture. The vendor states that documents are anonymised before analysis begins, that in the on premise configuration sensitive data never leaves the customer environment in identifiable form, that environments are isolated per client, and that the customer retains data ownership throughout. Anonymisation is configurable, with adjustable levels and term whitelisting. Segregation is documented at the level this buyer segment requires under the amended band, through per client environment isolation and single tenant deployment. Two gaps hold it off an A. Attorney client privilege and work product handling is not addressed directly in located material. No training prohibition on customer content was located on the public pages as of 29 Aug 2026, which for a vendor whose entire pitch is data protection is a conspicuous absence rather than a small one.

The most complete confidentiality posture on the index so far, and the only one that engages the professional secrecy question in statute. The vendor states compliance with Section 43e BRAO and Section 203 StGB, the German provisions governing what a lawyer bound by professional secrecy may use, and states this permits use by confidentiality bound professionals without requiring anonymisation. Training is addressed directly and in the negative: inputs and outputs are never used to train the AI and never shared with third parties. Segregation is documented at the level this segment requires under the amended band, with every project encrypted in transit and at rest under its own cryptographic key, strictly controlled access, and every action logged. The vendor further states it has no persistent access to plaintext data. Certification covers AI governance as well as security. Short of a full A only in that the underlying customer agreement is not published, so these commitments were read from the security page rather than a contract.

The audience is unambiguous, in house legal and business teams at enterprises, and the vendor is explicit that it is built for that rather than for law firm workflows. But the product is sold to non lawyers by design, with dedicated solution pages for HR, procurement, sales, claims and compliance, and a product called Discovery that delivers legal research and advice. Searched the site, the solution pages, the published terms and conditions and the resources index on 29 Aug 2026 and located no position on the advice line, no treatment of competence or supervision duties, and no statement of jurisdiction limits. Selling a legal advice surface to procurement and sales teams is the configuration where a published position matters most, and none was located.

A real position is published and it is grounded in named law rather than a disclaimer. The vendor addresses the professional rules that govern whether a lawyer may use the tool at all, citing Section 43e BRAO and Section 203 StGB and alignment with the EU AI Act, and states the audience precisely across law firms, in house teams, auditors and tax advisers, courts and public authorities. Jurisdiction limits are handled structurally rather than in prose: the product ships as jurisdiction specific editions and the site requires a jurisdiction selection before granting access, which is a real published boundary. Short of an A because competence and supervision duties are not addressed, and because the position is framed around the vendor's own permissibility rather than around what the lawyer remains responsible for.

Responsible AI is claimed without a mechanism. The vendor describes itself in published material as setting the global standard for responsible legal AI and states that controls are continuously audited against international standards, and it publishes governed execution and guardrails as product concepts. Searched the site, the security page, the platform pages and the resources index on 29 Aug 2026 and located no named internal owner of model governance, no pre release testing regime for model behaviour, no ISO 42001 or equivalent AI management certification, and nothing on uneven output across matter types, parties or populations. The security governance is real and is a different subject.

A published governance framework with real substance and independent validation, short of testing results or a named owner. ISO 42001 certification covers the AI management system specifically and is stated alongside ISO 27001, 27017, 27018 and 9001, BSI C5 and TISAX, with the vendor stating controls are independently and regularly audited and publishing a trust center for the full current list. Alignment with the EU AI Act is stated. That is a governance regime for AI rather than a principles page, and it is externally audited. Not located as of 29 Aug 2026: a named internal owner of model governance, published pre release testing results for model behaviour, or any disclosure about uneven output across matter types, parties or populations.

Substantive published policy covering most of the ground. Encryption at rest and in transit, per client environment isolation, attribute based access control, multi factor authentication and session governance, SSO across Okta, Azure AD and Google Workspace with SAML 2.0, OIDC and OAuth, annual independent penetration testing, and a trust center published at a stable URL. The anonymisation layer is a genuine data minimisation control rather than a policy statement. Not located as of 29 Aug 2026: a stated retention period or deletion control for documents and prompts, a named subprocessor list, and an incident or breach notification practice. Retention is the notable one, since it is the question this axis asks first.

Substantive published policy covering most of the ground. Per project encryption in transit and at rest with a distinct cryptographic key per project, strictly controlled access, full action logging, a stated position that the vendor has no persistent access to plaintext data, regular security reviews and risk assessments, and a Vanta backed trust center at a stable URL. Infrastructure providers are named individually with their own certifications, which is effectively a partial subprocessor disclosure at the hosting layer. Not located as of 29 Aug 2026: a stated retention period or deletion control for documents and prompts, a full subprocessor list, and an incident or breach notification practice.

Liability is addressed through published terms and conditions a buyer can read before signing, which puts this above a pure absence, but what was located is the standard shape rather than a position on AI output. Searched the published terms and conditions, the privacy policy, the security page and the trust center on 29 Aug 2026 and located no indemnity running to the customer for third party claims arising from output, no warranty on output, no stated liability cap and no insurance position. For a vendor selling into banking, insurance and regulated manufacturing, the absence of a published position on who bears the loss is a live buyer question.

Searched the site, the security page, the imprint, the data privacy statement, the FAQ and the trust center entry point on 29 Aug 2026. No published indemnity, liability cap, carve out, warranty on output or insurance position was located, and no customer terms or general conditions are published at all. Access to the product itself runs through a jurisdiction gated request form, so the agreement is reached through a sales process. Notable given how much else this vendor publishes: the compliance posture is documented in statutory detail while the allocation of loss when output is wrong is not addressed anywhere public.

Real integrations, named and documented at feature level, short of implementer depth. Microsoft Word and Outlook add ins each have their own product page, workflows trigger from email, Slack or Teams, Multi Review reads across data rooms directly from the customer's document system, and a dedicated integrations page describes the product as embedded across existing enterprise systems. Orientation is toward enterprise business systems rather than legal specific document management, which fits an in house buyer. Not located as of 29 Aug 2026: named document management connectors such as iManage or NetDocuments, and per integration documentation of what moves in which direction and what an administrator configures.

Integrations are named without documentation an implementer could use. A Microsoft Word integration is stated on the drafting product page, with work possible either in Noxtua directly or in Word. Searched the site, the three product pages, the security page and the FAQ on 29 Aug 2026 and located no other integration: no document management connector such as iManage or NetDocuments, no Outlook, no contract lifecycle or matter management, and no integrations page at all. Nothing describes what the Word integration moves, in which direction, or what an administrator configures.

The strongest deployment disclosure in the index so far, and the first A on this axis. Three models are published with the differences between them spelled out: fully managed SaaS; single tenant, a dedicated isolated instance deployed in the customer's preferred Azure region, fully managed and completely segregated from other customers; and single tenant plus on premise anonymisation, where the anonymisation component runs inside the customer environment and only anonymised data crosses to the dedicated tenant over private encrypted connections. Residency is stated as customer selected Azure region, and the architecture separates where identifiable data is processed from where the tenant sits, which is precisely the processing versus storage distinction this axis asks for. The public FAQ adds private cloud, hybrid and full on premise. Short of naming the specific regions available.

Where the software runs and where the data sits is published in more concrete detail than by any other vendor on this index, because it is the product's entire proposition. Hosting providers are named individually with their certifications and their roles: IONOS as a BSI C5 certified critical infrastructure provider operating the Bundescloud; Deutsche Telekom's Industrial AI Cloud, the AI Factory in Munich underpinning the Deutschland Stack for regulated sectors including the judiciary; T Cloud Public, BSI C5 certified for critical infrastructures; and T Cloud Public in Switzerland, BSI C5 certified and FINMA compliant, offering Swiss data sovereignty. The vendor states data stays in Europe on infrastructure operated by European providers independent of US cloud providers and shielded from the US CLOUD Act, which addresses jurisdiction of processing and not merely geography of storage. Short only of a per tier statement of what changes between deployment options.

Certification is real and stated with an open access route, short of accessible evidence. ISO 27001 and SOC 2 Type II are both stated as certified, annual independent penetration testing is stated, compliance frameworks across the EU, UK and Middle East are named, and a trust center is published at a stable URL, which under the three tier test is materially better than absent. What was not located as of 29 Aug 2026 is any coverage period, audit scope, report date or named auditor. Worth flagging one noun problem: the security page opens by saying the compliance framework is aligned with ISO 27001, SOC 2 Type II and GDPR, and a later section says certified. Alignment and certification are different claims, and the page makes both about the same standards.

The broadest certification set on the index, with a trust center reachable without a sales call. Named and current: BSI C5, TISAX, ISO 42001, ISO 27001, ISO 27018, ISO 27017 and ISO 9001, with the vendor stating controls are independently and regularly audited and describing itself as the most comprehensively certified legal AI in Europe. The trust center is a Vanta hosted portal at a stable URL carrying the full current certification list and policies, which is a self serve route rather than a sales conversation. BSI C5 and TISAX are meaningful here rather than decorative, being the German federal cloud computing criteria and the automotive industry information security assessment. Short of the very top only because no coverage period, report date or named auditor was located on the public pages as of 29 Aug 2026.

The vendor refers to model selection without identifying what sits underneath. The architecture is disclosed at a structural level and it is a real disclosure: the platform is LLM agnostic and selects the best foundation model for each legal scenario, and a fine tuned anonymisation model runs ahead of any external model call. That tells a buyer how the supply chain is shaped. What it does not tell them is which providers are in it. Searched the site, the security page, the anonymisation platform page, the trust center entry point and the published legal pages on 29 Aug 2026 and located no named model provider, no subprocessor list, no statement of where models run, and no commitment to notify customers when the selection changes.

The clearest supply chain answer on the index, because the vendor owns the chain. Models are proprietary and trained in house on licensed legal data rather than sourced from a third party foundation model provider, which is stated plainly and is the point of the sovereignty positioning. Where models run is named to the specific provider and facility, IONOS and Deutsche Telekom's Industrial AI Cloud in Munich among them. One named external component is disclosed with its partner: the retrieval model Noxtua Voyage Embed, built with Voyage AI and dejure.org, with the model identifier published. A customer therefore inherits a dependency chain that is both short and named. Not located: a commitment to notify customers when the model or its hosting changes.

Searched the site, the navigation, the product pages, the industry and solution pages and the published terms on 29 Aug 2026. There is no pricing page anywhere on the property, no published rate, no stated unit of charge, no tier structure and no seat minimum. The only commercial entry point on every page is a demo booking, which is sales gated and earns no credit. Note that three deployment models are published in detail with no indication of what any of them costs or how they differ commercially.

Searched the site, the navigation, the three product pages, the jurisdictions page and the FAQ on 29 Aug 2026. No pricing page exists, no rate is published, no unit of charge is stated and no tier structure appears. The only commercial entry point is a Get Access request that first requires selecting a jurisdiction, so a buyer cannot reach a figure or even a product edition without entering a sales process. Third party pricing figures were not located either.

Who the product serves is documented precisely and the boundary is stated rather than implied. Five industries carry dedicated pages, banking, insurance, transportation and mobility, technology and professional services, and six functions carry their own, legal, claims, compliance, HR, procurement and sales, which is unusually explicit about reaching beyond the legal department. The vendor states plainly that it is built for enterprise legal and business teams rather than law firm productivity workflows, which is a published limit on who it is not for. Jurisdictional coverage is quantified, though inconsistently at 110 plus in one place and 130 plus in another. Named enterprise references span the segments claimed.

Segment coverage is described with substance and with an unusual axis of precision. Four segments carry their own published sections: in house legal teams, law firms, auditors and tax advisers, and courts and public authorities, the last being the first explicit judiciary and public administration positioning on this index. Jurisdictional coverage is stated concretely as named editions per country rather than as a count, spanning Germany, Austria, Switzerland, Poland, Czech Republic, Slovakia, Bulgaria and Sweden, each tied to its national legal publisher. Short of an A because practice areas supported are not enumerated beyond a general research, analysis and drafting framing, and because what is not supported is nowhere stated.

Searched the security page, the anonymisation platform page, the published terms and conditions, the privacy policy and the home page on 29 Aug 2026. No located term or policy states whether customer content may be used to train models, either way. The vendor's published position is architectural rather than contractual on this point: documents are anonymised before analysis and, in the on premise configuration, sensitive data never leaves the customer environment in identifiable form. That reduces what a model could receive but is not a statement about training. Under the rule that a value is never inferred from the absence of a contradiction, this records as silent. Conspicuous for a vendor whose entire positioning is data protection.

The security page states directly that inputs and the product's outputs are never used to train the AI and are never shared with third parties, and separately that the vendor has no persistent access to plaintext data. The commitment as located sits on a public security page rather than in a customer agreement, and no customer terms are published on this property, so it records at the policy level. Worth noting the training relationship runs the other way as well and is disclosed: the vendor states it trains its proprietary models on legal data licensed from publishers, which is publisher content rather than customer content.

Searched the security page, the platform pages, the published terms and conditions, the privacy policy and the trust center entry point on 29 Aug 2026. No public material states how long documents, prompts or outputs are retained, whether the customer controls the window, or whether deletion is available. The vendor states that customers maintain data ownership throughout and that environments are isolated per client, neither of which answers the retention question. First vendor on this index to record an absence rather than a period here.

Searched the security page, the data privacy statement, the FAQ, the imprint and the trust center entry point on 29 Aug 2026. No public material states how long inputs, outputs or uploaded documents are retained, whether the customer controls the window, or whether deletion is available. The vendor does publish related architecture, per project encryption under a distinct key and a statement that it holds no persistent access to plaintext data, but neither answers the retention question. Second vendor on this index to record an absence rather than a period.

The product maintains its own documented permission model rather than inheriting one from a document management system. Published controls are attribute based access control, multi factor authentication and session governance, configurable to the customer's own standards, with SSO through Okta, Azure AD or Google Workspace enforcing the customer's access policies. Environments are isolated per client, and single tenant deployment segregates a customer completely from others. What is not documented is segregation between users or matters inside a customer, and no document management integration was located whose permissions retrieval could enforce at query time.

The product maintains its own documented segregation model rather than inheriting one from a document management system. Every project is encrypted in transit and at rest under its own cryptographic key, access is strictly controlled and every action is logged, and the vendor states it has no persistent access to plaintext data. Per project cryptographic separation is a stronger mechanism than most and it is documented, but it is the vendor's own model, which the firm must keep aligned with its walls. No document management integration was located whose permissions retrieval could enforce at query time, and no material addresses conflicts or ethical walls as such.

Searched the published terms and conditions, the privacy policy, the security page and the trust center entry point on 29 Aug 2026. No clause addressing government or law enforcement requests for customer data was located, and no transparency report was located. Worth noting for a future reader that the on premise anonymisation configuration materially changes what a vendor could produce in response to such a request, since identifiable data never reaches it, but the vendor does not make that argument in published material and it is not recorded as a value here.

Searched the security page, the data privacy statement, the imprint, the FAQ and the trust center entry point on 29 Aug 2026. No clause committing to notify a customer of a government or law enforcement request for their data was located, and no transparency report was located. The vendor does address the adjacent question of foreign jurisdiction directly, stating data stays in Europe shielded from the US CLOUD Act on infrastructure operated by European providers, which speaks to which state could compel production rather than to whether the customer would be told.

Coverage is described by jurisdiction with no identification of the underlying corpus. Vendor material states that answers are grounded in verified case law and legislation across more than 130 jurisdictions, with a separate home page figure of more than 110 jurisdictions for global translation and jurisdiction coverage. Neither number is accompanied by a named source or publisher, a licence or public domain basis, or an update cadence. A Knowledge platform page describes verified legal knowledge without saying where it comes from.

The strongest corpus disclosure on the index so far and the first to state a licence basis. Sources are named individually as national legal publishers and tied to the jurisdiction each serves: C.H.Beck for Germany, Poland, Czech Republic and Slovakia, MANZ for Austria, Ciela for Bulgaria and Blendow for Sweden, plus a Swiss edition, with dejure.org named in the retrieval model partnership. The rights basis is stated as exclusive publisher partnerships, described as the first Europe licence of its kind. Scale is quantified at more than 130 million searchable documents and more than 8.5 million court decisions. Short of the top value only because no update cadence or lag for the corpus was located as of 29 Aug 2026.

Searched the site, the Discovery product page, the Knowledge platform page and the resources index on 29 Aug 2026. No material was located addressing whether authority returned by the product carries a treatment signal, whether subsequent history is checked, or whether any commercial citator is licensed. The vendor describes case law as verified without stating what verification means or when it was performed.

Searched the site, the three product pages, the jurisdictions page and the press releases on 29 Aug 2026. No material was located addressing whether authority returned carries a treatment signal or whether subsequent history is checked. The Understanding product is described as assessing the validity of clauses, which is contract clause analysis rather than the standing of cited authority, and was not treated as evidence here. Noted for context: continental civil law jurisdictions do not use citators in the Anglo American sense, so this signal reads differently for a vendor selling into Germany, Austria and Switzerland than for a US product.

Searched the site, the six product pages, the Agent Studio page and the resources index on 29 Aug 2026. No published material describes what the product does when it cannot ground an answer, and no explicit no answer path or confidence signal was located. Agent Studio documents escalation when needed as a workflow routing behaviour, which is about handing a task to a person rather than about the system declining to answer, and was not treated as evidence for this signal.

Searched the site, the three product pages, the security page and the press releases on 29 Aug 2026. No published material describes what the product does when it cannot ground an answer, and no explicit no answer path or confidence signal exposed to the user was located. The vendor publishes that results are traceable at every stage and that version 5 surfaces a knowledge graph of the sources behind an answer, which lets a reader inspect the basis of an answer that was given rather than telling them when the system found nothing.

No court order, opinion or disciplinary record naming this product has been located as of 29 Aug 2026. Instrument searched: the AI Hallucination Cases database maintained by Damien Charlotin, which tracks court decisions worldwide involving AI generated hallucinated content and records the AI tool implicated where it is known. Also checked published 2026 sanctions summaries and secondary sanctions trackers. The entries located name filers, and in some rows other products, rather than this one. This is a statement about the public record on the date shown and not a clearance, and it is bounded by what that database covers, which is weighted toward US and other common law filings while this vendor sells primarily into European enterprises.

No court order, opinion or disciplinary record naming this product has been located as of 29 Aug 2026. Instrument searched: the AI Hallucination Cases database maintained by Damien Charlotin, which tracks court decisions worldwide involving AI generated hallucinated content and records the AI tool implicated where it is known. Also checked published 2026 sanctions summaries and secondary sanctions trackers. The entries located name filers, and in some rows other products, rather than this one. This is a statement about the public record on the date shown and not a clearance. The bound is worth stating plainly for this vendor: that database is weighted toward US and other common law filings, and this product sells into German speaking and central European jurisdictions whose decisions are less comprehensively covered by it.

Searched the site, the security page, the solution pages and the resources index on 29 Aug 2026. No engagement with any named ethics opinion or bar guidance was located. The vendor publishes compliance material aligned to information security and data protection standards across the EU, UK and Middle East, which addresses its own regulatory posture rather than the professional responsibility obligations its buyers are bound by. Noted for a future reader: this vendor sells primarily into European enterprises, where the applicable guidance is issued by national bars and law societies rather than the ABA, and none of those was engaged either.

Public materials engage with named professional obligations by statutory citation rather than in general terms: Section 43e of the German Federal Code for Lawyers and Section 203 of the German Criminal Code, the provisions governing what a lawyer bound by professional secrecy may use, plus stated alignment with the EU AI Act. The vendor states this permits use by confidentiality bound professionals without requiring anonymisation, which is a specific claim about the rules its buyers are bound by. Short of the mapped by jurisdiction value because the engagement is with German law only, despite the product shipping in eight national editions, and no equivalent Austrian, Swiss, Polish, Czech, Slovak, Bulgarian or Swedish provision was addressed in located material.

Vendor material is framed around speed and cost reduction, including a published customer story headlined on broader market coverage and reduced external legal spend, and product copy promising review and negotiation in minutes rather than hours. Searched the site, the product pages, the customer stories and the resources index on 29 Aug 2026 and located no per matter record of AI assisted work intended for fee purposes, and no published guidance on billing, fee or client disclosure treatment. The buyer here is an in house team rather than a firm billing a client by the hour, so this signal reads differently for this segment.

Searched the site, the three product pages, the customer stories entry point and the press centre on 29 Aug 2026. No published guidance on billing, fee or client disclosure treatment was located, and no per matter record of AI assisted work intended for fee purposes was located. Distinct from the other vendors on this index in that no time savings or efficiency figures were located either: the vendor's published framing is precision, verifiability and permissibility rather than hours saved, so there is no savings claim to weigh against the client's side of the equation.

Searched the security page, the trust center entry point, the published terms and conditions and the privacy policy on 29 Aug 2026. No subprocessor list was located, no statement of which model providers see customer content was located, and no client facing consent or notification material was located. A trust center exists at a stable URL, which is an access route rather than the artifacts themselves. Worth recording that this vendor's architecture arguably answers the underlying question differently, since the on premise anonymisation option means identifiable client content need not reach a model provider at all, but the disclosure a firm could forward is not published.

A trust center is published at a stable URL, reachable without a sales conversation, carrying what the vendor describes as the full list of current certifications and policies. The certification set named on the public security page is broad and specific: BSI C5, TISAX, ISO 42001, 27001, 27018, 27017 and 9001. Infrastructure providers are named individually with their own certifications, which covers the hosting layer of the supply chain, and the model layer is proprietary rather than third party, which removes the model provider question a firm would normally have to answer for its client. Short of the full disclosure pack because no formal subprocessor list was located and no client facing consent or notification material was located as of 29 Aug 2026.

Some elements of a record are available, short of a document level export. Multi Review is documented as exporting audit ready reports intended for deal teams, regulators or internal stakeholders, and the same review logic is stated to apply across every document so results are comparable. That is an exportable artifact, but it is a review output rather than a record of how AI produced it. Not located as of 29 Aug 2026: any export covering model used, sources retrieved and human verification per document, and the model used is not identifiable in any case since the platform selects different models per task.

Several elements of a disclosure record are available and unusually well documented. The vendor states every action is logged, that agentic research is traceable at every stage, and that version 5 generates a knowledge graph per query surfacing the sources relied on and mapping how they connect, with verifiable source references throughout. Sources retrieved and the reasoning path are therefore recoverable. Two elements are missing: no per document export covering model used, sources retrieved and human verification together was located, and human verification is not recorded as such. The model question is simpler here than elsewhere, since the model is the vendor's own proprietary system rather than a rotating third party.