Anaqua vs Questel: how they compare in 2026

This is the incumbent-with-an-AI-layer case, and the layer is real but plainly peripheral. AQX is a twenty-year-old IP management system whose core is docketing, deadline calculation, renewals, prosecution workflow, document management and IP financial management, none of which depends on machine learning. Four AI features sit on top: a patent classifier, generative AI Patent Summaries in AcclaimIP, document auto-processing of patent office correspondence using Microsoft Azure AI Document Intelligence, and WiseTime's AI Narrative for invoice descriptions. Each is discrete and separately marketed, and each is an accelerant on a workflow that already existed. Remove all four and a full enterprise IPMS remains, which is the distinction this axis is asking about.

This is the incumbent-with-an-AI-layer case in its purest form, and the layer is genuine but plainly peripheral to what is being sold. The core is a patent and trademark database business, an IP management system handling docketing, renewals, forecasting and e-billing, and a large human services arm covering filings, EP validation, translation, recordals and renewals through a network described as 320 agents and 1,200 translators. The AI features are discrete and separately marketed: an AI-Classifier, Sophia as a cross-platform agent, Qthena for prosecution workflow, generative patent drafting, Markify ProSearch for trademark screening, and office action and goods-and-services drafting assistants. Strip all of them out and the databases, the IPMS and the services business remain intact. The one component that would not survive is the generative patent drafting module, which is model-native and arrived through the acquisition of qatent, but it is one product inside a suite of roughly thirty. Third consecutive ip-and-patents incumbent at this grade, against IPRally at A on the same axis.

Accuracy is asserted repeatedly and measured nowhere. AI Patent Summaries is described as providing a consistent, high-quality review of any patent, the classifier promises explainable AI insights, and the Adaptable principle commits to maintaining the accuracy and relevance of the AI tools through regular training and refinement. No benchmark, test set, error rate, or named failure mode was located on the AI initiatives page, the guiding principles page or the security page on 31 August 2026. The citator and primary-authority limbs of this axis do not apply, since the product summarises and classifies patent documents rather than citing legal authority. The applicable risk is a wrong summary or misclassification feeding a portfolio decision, and nothing published addresses how often that happens. The linked AcclaimIP article describing the summaries feature was not opened this session.

Accuracy is discussed at length and measured nowhere located. The Trusted AI page carries a heading on a duty of transparency in the accuracy of results, but the text under it is written as intention rather than description: it says users need to be informed of uncertainty and given the means to verify, and that this will be achieved by providing sources and explaining the reasons behind answers. Future tense is not evidence of a shipped behaviour and it is not credited here. The AI Policy is more candid than most, stating that Questel does not guarantee the absolute accuracy of AI content and acknowledging the potential for hallucinations, which is an honest disclosure but not a measurement. A press release on the home page announces that the Questel AI Lab created a proprietary model for patent search to benchmark and enhance semantic retrieval; that release was not opened on 1 September 2026 and no benchmark figure, test set or error rate was located on the home page, the AI Policy, the Trusted AI page or the AI-in-IP solutions entry. If the AI Lab release carries a measured number with the test set described, this row moves up.

One review surface is described and the rest is asserted. In WiseTime the timekeeper sees a timesheet and can adjust durations, activity codes, matter references and the activity description, either manually or using AI Narrative, which is a concrete and editable checkpoint. Against that, the Explainable principle commits to a clear understanding of how AI models perform and for what purpose without describing any mechanism that delivers it, and the highest-consequence automation on the platform has no published control at all: auto-processing takes correspondence from the US, EU, Japanese and WIPO offices into the docket, where an error becomes a missed deadline, and nothing states what a human checks or when the system stops. No threshold, no confidence gate, and no statement of what happens after the system is wrong was located.

Oversight is asserted repeatedly and no mechanism is described. The Trusted AI page states that full automation is ideal in some instances while in others AI should act as a supportive tool or human judgment is necessary, and that the balance must be found, which names the question without answering it for any product. AI Policy Article 5 places the duty on the buyer, saying use of the Services remains under the professional responsibility of each client and that clients are responsible for reviewing and verifying accuracy and completeness. The products are marketed as copilots and assistants, which implies supervision without documenting it. What was not located on any surface read on 1 September 2026 is a statement of what any feature runs unattended, where a reviewer sits in the workflow, what threshold stops the system, or what happens after it is wrong. This matters most on the office action response and patent drafting assistants, where the output is a document heading toward a patent office deadline.

The attribution is unusually senior and the measurement is absent. Named executives are quoted with titles and employers: Lars Kellberg, Corporate VP of Corporate Patents at Novo Nordisk; Toe Su Aung, Head of IP at British American Tobacco; Hans Pennings at NXP Semiconductors describing a 25,000-patent portfolio across nine innovation centres; Dr Udo Meyer, SVP of Global IP at BASF; and Christof Wolpert and Sara Halton at adidas on the HyperView analytics tool. Corporate and law firm logos run from TSMC, GSK, ARM and BD to Thompson Coburn, Michael Best, Lewis Rice and Conley Rose, and a 2026 announcement names Innomotics as a new client. What is missing is any figure for what changed: the quotes describe satisfaction and scale rather than a measured outcome, none carries a date, and several identify the speaker's former role. The client stories library and client reviews page were not opened this session.

Customers are named against specific products, and no figure attaches to any of them. The home page and product navigation carry dated client announcements pairing an organisation with a named product: ELT Group selecting Orbit Intelligence, Fideon selecting Equinox Law Firm, TALKUAL selecting Markify Watch, Pointon Partners selecting the Markify platform, and HGF expanding Qthena for firm-wide patent practice after what is described as years of successful deployment. A separate case study describes how Equinox Law Firm supports Redchip Lawyers. That is more than a logo strip, which is what keeps this above the band below, although the home page also runs one of those, showing Sanofi, PepsiCo, Dow, Novartis and Danone without attribution. The scale claim of more than 20,000 clients in 30 countries is stated with no basis. This grade rests on the announcement headlines and their product pairings as they appear on the pages fetched; the client stories library at questel.com/clients-stories and the individual announcement bodies were not opened on 1 September 2026, so no dated measured outcome was either found or excluded.

The architecture is specific and the central question is unanswered. Anaqua publishes single-tenant hosting with a dedicated instance per customer, record-level security with permissions by role, SSO integrated with HR systems so access is revoked when staff leave, AES-256 at rest and TLS 1.2 in transit, and a dedicated SFTP site for sharing files with outside counsel. The Secure AI principle adds that each client operates in an environment where no AI-driven insights or data are shared between unintended recipients. But nothing located states whether client IP data is used to train or refine Anaqua's models, which is the question a patent department asks first, and the Adaptable principle's reference to regular training and refinement of the AI tools makes the silence more pointed rather than less. Privilege and work product are never mentioned, no retention or deletion position is published for portfolio content as distinct from personal data, and there is no customer agreement on the site in which any of this could be read as a commitment.

The substantive commitments are published and readable in advance, and they sit in policy rather than in a contract a buyer can hold. The Trusted AI page commits that information a user submits to an AI inference is not used to train or fine-tune Questel models, gives AES-256 encryption at rest and TLS 1.2 or better in transit, limits access to authorised employees needing it for engineering support, abuse investigation and legal compliance plus named categories of contractor bound by confidentiality, confirms the customer owns inputs and outputs, and states zero data retention on leaving with deleted conversations removed within 30 days. AI Policy Article 3 adds that third-party providers undertake not to use client data to train their own models. The page even reaches the professional dimension, saying the design removes the risk of revealing information relating to the representation of a client. What is missing is the rest of the picture: nothing addresses segregation between users, teams or matters inside a customer account, which matters because Equinox Law Firm is sold to firms, and the Master Services Agreement and Data Processing Agreement that the AI Policy says govern this are referenced as signed with clients and are not published anywhere on the property.

Nothing published addresses the advice line for the product. The website Terms of Use of 19 May 2025 does say the materials do not constitute legal advice, but that document is scoped by its own opening to the anaqua.com website and governs visitors rather than AQX users, so it does not reach the software. Searched the home page, the AQX Corporate and Law Firm entries, the AI initiatives page, the guiding AI principles page, the security and hosting page, the privacy policy and the Terms of Use on 31 August 2026: no statement of what the product is and is not, no supervision or competence discussion, no jurisdiction limit, and no engagement with ABA Formal Opinion 512 or any bar guidance was located. This matters on a platform sold to law firms whose docketing output is a computed filing deadline. Not every AQX Law Firm sub-page was opened, so the absence is recorded against the surfaces named.

A real position on advice versus tooling is published, and it is unusually direct for this lane. AI Policy Article 5 states that Questel's Services are not intended to replace or substitute the expertise and judgment of legal professionals, that clients should seek professional legal advice before relying on AI-generated outputs, that the use of the Services remains under the professional responsibility of each client, and that clients should not submit confidential information they do not wish the AI engine to process. That is a published advice line rather than a boilerplate disclaimer buried in website terms, and it appears in a document about the product rather than about the site. What is absent is the rest of the treatment. No jurisdiction limits are named, nothing addresses a practitioner's supervision or competence duties, and no engagement with ABA Formal Opinion 512, USPTO practitioner guidance, EPO guidance or any national patent attorney regime was located on the AI Policy, the Trusted AI page, the law firm solutions entry, the legal notice or the privacy policy on 1 September 2026.

Anaqua publishes a dedicated guiding AI principles page setting out six principles, and it carries more than the usual list of adjectives. It names an external framework, stating that Anaqua follows the EU AI Ethics Guidelines for the responsible use and development of AI and that its principles will change as AI law does. It commits to data minimisation, saying models have access to only the minimum data necessary. It commits to explainability in specific terms, covering how models perform, how AI is used and for what purpose, with regular updates to clients. And it describes a real development process, with AI capabilities built through named client committees and working groups. What is absent is what would make it auditable: no owner inside Anaqua is named as accountable, no pre-release testing regime is described, and no results are disclosed. The claim that models are transparent, fair and unbiased is asserted without any supporting evaluation, which is the single largest gap on a tool that classifies patents into technology categories.

This is the most developed governance framework located in the ip-and-patents lane. Questel publishes a dedicated Artificial Intelligence Policy issued at group level across Questel SAS and its affiliates, structured in six articles. It commits to complying with the EU AI Act, naming the regulation and its entry into force on 12 July 2024, which is a named external regime rather than a slogan. Article 6 describes a real internal structure: an AI Committee and a dedicated AI taskforce drawing representatives from the Legal, Information Security, Data Privacy and Business Development functions, plus ongoing employee training on ethics, data privacy and security. The Trusted AI page adds an AI oversight committee and states that every conversation starts with whether the company should do something rather than whether it can. Article 4 commits to safeguards preventing bias among other misuse. What holds this below the top band is exactly what that band asks for: no individual inside Questel is named as accountable, no pre-release testing regime is described, and no evaluation results of any kind have been disclosed, so the commitment to prevent bias is asserted rather than evidenced.

Most of the ground is covered from two documents. The security and hosting page gives AES-256 at rest, TLS 1.2 in transit, regular Azure backups with restore capability, SSO with role-based provisioning, and record-level access control. The privacy policy of October 2023 sets a retention position tied to the customer relationship plus a reasonable period afterwards, notes that anonymised data may be kept indefinitely, confirms Anaqua's role as processor with the customer as controller, commits to EU Standard Contractual Clauses for onward transfers from the EU, limits internal access to those who need it, and names a Privacy Officer with a postal address and email. The California policy adds a commitment to notify affected parties and regulators of a suspected breach where legally required. The gap this band names is present: no subprocessor list is published anywhere. Retention and deletion are also framed entirely around personal data, with nothing stating what happens to portfolio documents or AI inputs after processing.

The AI-specific stewardship is specific and the general estate around it is thin, so this sits below the band above on three of the five limbs. What is published and good: the Trusted AI page gives zero data retention once a customer leaves, deletion of removed conversations within 30 days unless legally required otherwise, AES-256 at rest and TLS 1.2 or better in transit, and access restricted to authorised employees needing it for engineering support, abuse investigation or legal compliance plus specified contractors. Against that, the Data Privacy Policy updated May 2026 declines to state a retention period at all, saying data is kept as long as necessary and directing the reader to email for more information. No subprocessor list for the products exists: the policy's third-party section names thirteen website marketing and analytics tools including Google Analytics, LinkedIn Ads, Salesforce Pardot, AdRoll and SEMRush, and no processor touching customer IP data. Incident practice is described only as notifying the supervisory authority and, where required, data subjects, with no customer-facing commitment. A Questel IT Security Policy is referenced in the privacy policy and was not located as a published document. Worth flagging that AI Policy Article 2 claims full transparency on third-party providers as outlined in the personal data policy, and that policy does not deliver it for the product.

There is no customer agreement on the property. The only contract published is the website Terms and Conditions of Use, which states in its own opening that it governs use and access of the website at www.anaqua.com, and whose disclaimer and liability clauses are correspondingly about the materials on that site rather than about AQX. Searched the footer, the security and hosting page, the AQX Corporate and Law Firm entries, the privacy policy and the resource library on 31 August 2026: no master subscription agreement, service agreement, indemnity, liability cap, warranty on output or insurance position covering the software was located. A buyer cannot learn who bears the loss if the classifier misfiles a patent or auto-processing mishandles an office action until they are inside a negotiation.

No customer agreement is published, and unlike the comparable case in this lane something about the product's failure mode still is. AI Policy Article 5 states that Questel does not guarantee the absolute accuracy of AI content generated by its Services and expressly disclaims any liability for inaccuracies, given the potential for errors or hallucinations inherent to AI. That is an allocation of loss a buyer can read before signing and it is scoped to the Services rather than to the website, which is the distinction from Anaqua, whose only published contract was a website Terms of Use that did not reach its software. Everything else is absent. The Master Services Agreement and Data Processing Agreement are referenced in AI Policy Article 2 as signed between Questel and its clients and neither is published. No indemnity, liability cap, carve-out, warranty on output or insurance position was located. The legal notice is a French mentions legales scoped by its own text to the corporate website and disclaims liability only for the site and its materials. Searched the footer legal set, the AI Policy, the Trusted AI page and the privacy policy on 1 September 2026.

The connections are named and several are described in terms of what actually moves. Document auto-processing ingests correspondence from the four largest patent offices, identified as the US, WIPO, European and Japanese offices, with a separate integration to USPTO TSDR for trademark teams and prior art import into the IDS process. SSO integrates with HR systems specifically so that permissions are removed when employees leave, which is a described direction of sync rather than a logo. Anaqua also owns and sells SeeUnity as its application integration product, and a dedicated SFTP site handles file exchange with outside counsel and partners. What was not located is a public integrations directory or API reference on anaqua.com that an implementer could work from, and the SeeUnity property itself was not opened this session, so the depth of the document management connections it provides is not established here.

Integration is a named commercial position rather than a logo wall, and the direction of movement is stated for the specific cases. Questel markets an Integrated IP Ecosystem entry and describes its IPMS as directly integrated with leading IP renewal services and top-tier IP databases to eliminate data gaps. Two connections are described concretely enough to assess: PAVIS Connect, sold as a renewals connector explicitly for companies and law firms that already have an IP management system, which is an integration into a competitor's system rather than a demand to migrate; and the Markify Watch integration into Equinox, described as accessing trademark watches directly from within Equinox. An IP Services Portal handles service ordering and monitoring, and the IPMS entry lists e-billing and blockchain tooling. What is missing is depth of the kind an implementer could work from: no public API reference, developer portal or integration documentation index was located on the site navigation, the integrated ecosystem entry or the IP management software pages on 1 September 2026, and no named connection to a document management system such as iManage or NetDocuments was found.

This is the most complete deployment disclosure located in the pull so far. The security and hosting page states single-tenant hosting, defined as a single dedicated instance of the application deployed for each customer, on Microsoft Azure, with data location options in the United States, Canada, Europe, Japan and India. It also describes a second deployment tier and what changes in it: AnaquaGov runs AQX in the Azure Government environment, restricts access to US citizens, residents and legal permanent residents, holds CUI and export-controlled data, has been independently assessed against NIST SP 800-171, and adds configurable US export control procedures. A hosting data sheet is published alongside. The limb that keeps it here is the one this band reserves for the top: residency is offered without processing location being addressed separately from storage, and in particular nothing states where the AI features process data, which matters when one of them runs on a Microsoft AI service.

Cloud delivery is implied throughout and neither the tenancy model nor a region is stated. The products are described as SaaS software solutions in the AI Policy, and the Trusted AI page refers to business data stored on Questel systems, but nothing located names a hosting provider, a data centre location, a country or region option, or whether customers share infrastructure or receive a dedicated instance. The only geography published is in the privacy policy's transfer section, which says Questel is a global business, that data may be transferred to countries with different protection laws, and that transfers out of the European Union are covered by intra-group agreements incorporating Standard Contractual Clauses and by clauses executed with third parties. That describes the legal basis for movement rather than where anything sits, and it is scoped to personal data rather than to the IP portfolios and unpublished applications the platform holds. Searched the home page, the AI Policy, the Trusted AI page, the privacy policy, the legal notice and the integrated ecosystem entry on 1 September 2026.

The certifications are real, named and scoped. The security and hosting page states ISO 27001 certification and a Type 2 SOC 2 examination of Anaqua's information security management system supporting infrastructure and services, which is more scope detail than most of this corpus gives. Anaqua's own press releases name the auditor, Schellman & Company, for both the SOC 2 examination and the 2017 ISO 27001 certification, and name Schellman again as the C3PAO that assessed the AnaquaGov enclave against NIST SP 800-171; ISO 9001 is also claimed. What is missing is access and currency. There is no trust portal, no report request route, and no examination period or certificate date on any current page, so a buyer reading in 2026 can see which standards are claimed but cannot confirm the attestation is live or obtain it without a sales conversation.

The standard is named in prose rather than shown as a badge, which is what separates this from the band below. AI Policy Article 3 states plainly that Questel is an ISO 27001-certified company adhering to international standards for information security management, and an ISO mark also appears on the home page against a line about transparency and consistency. That is a real, stated certification. Everything that would let a buyer act on it is absent: no scope statement covering which entities, products or infrastructure are certified, no certificate date or validity period, no auditor or certification body named, and no SOC 2 or other attestation claimed. There is no trust centre or security portal anywhere on the property, and no route to request a report or a completed questionnaire was located. The privacy policy points to a Questel IT Security Policy that is referenced but not published. Checked the footer, the AI Policy, the Trusted AI page, the privacy policy and the home page on 1 September 2026.

One component is identified and the rest are not. The AI initiatives page names Microsoft Azure AI Document Intelligence as the service behind prior art import for the IDS process, which is a real disclosure. For the other three capabilities, the AI patent classifier, the generative AI Patent Summaries, and WiseTime's AI Narrative, no model, model family, provider or version is named anywhere located on 31 August 2026, and no commitment to notify customers when any of that changes was found. The Azure hosting relationship is described in detail but that is infrastructure and it is credited on the deployment row rather than here, since naming the cloud a system runs in does not identify the model inside it. A firm asked by a client which model reads its unpublished patent text cannot answer from published material.

The providers and the model families are named, and the currency of that disclosure is the problem. The Trusted AI page states that Questel leverages generative models from OpenAI through API integrations, linking the OpenAI API Platform, and that it also uses open-source models identified as Mixtral, Falcon and Llama 2, fine-tuned and trained on proprietary data sets, with prompts optimised in-house. It adds that all models are fine-tuned using public patent, trademark and design text or Questel's own data. That is more than most of this corpus publishes. Three things keep it below the top band. No commitment to notify customers when the model set changes was located. Where the models run is only partly answered, since the OpenAI relationship is described as an API integration while nothing states where the open-source models are hosted. And the named models are dated: Llama 2 and Falcon are generations behind, while a 2026 press release on the home page announces a new proprietary Questel AI Lab model for semantic patent search that appears nowhere on this page, so the disclosure a buyer reads may no longer describe the system they would use.

No pricing information is published at any level. There is no pricing page, no pricing entry in the navigation, no tier or package names, no unit of charge, and no indication of what implementation adds. Every commercial route on the site leads to Contact Us or Schedule a Call With an Expert. Checked the home page, the AQX Corporate, AQX Law Firm and PATTSY WAVE entries, the services pages, the security and hosting page and the site footer on 31 August 2026. The absence is consistent with an enterprise sales model and it is still an absence: a buyer cannot learn the unit Anaqua charges on, let alone a figure. Note that QuantifyIP, listed under software as IP cost forecasting, forecasts a client's own patent office costs and says nothing about the price of Anaqua.

The shape of the offering is unusually visible and the number is nowhere. Tiering is published across the range: Equinox is sold as Corporate, Corporate+, Law Firm and Law Firm+, with the plus editions marked for enterprise organisations, and Orbit splits into Intelligence, Express, Insight, Intellixir, BioSequence and Chemistry, with Markify separated into ProSearch, Comprehensive Search, Full Pharma Search and Watch. A buyer can therefore see the packaging and roughly which segment each tier addresses. No rate, band, unit of charge, currency or statement of what implementation adds was located anywhere, and the commercial route on every page read is Request a call or Contact. Two products carry a BUY ONLINE label in the navigation, Defensive Publication and Markify Comprehensive Search, indicating a self-serve purchase path that would ordinarily display a price; neither of those pages was opened on 1 September 2026. This grade is therefore rebuttable upward if either carries a published figure, and that check is the single cheapest thing a future pass could do on this record.

Segment coverage is described with real substance and separate products behind it. Anaqua sells AQX Corporate to in-house IP departments, AQX Law Firm to firms, AQX Pharma as a vertical edition, PATTSY WAVE for docketing-led buyers, and AnaquaGov to US government contractors handling controlled unclassified information, which is an unusually explicit fifth segment. The client roster demonstrates both main segments, running from adidas, TSMC, GSK and BASF on the corporate side to Thompson Coburn, Michael Best and Conley Rose on the firm side. Practice coverage spans patents, trademarks, domain names, innovation and partnership management, with analytics and annuity services alongside. The boundary is what is missing: no statement of firm size, portfolio size, or the practice areas or jurisdictions where the product is not the right fit was located.

Segment coverage is documented with more substance than anything else in this lane, and the boundary is still left open. Questel publishes dedicated solution entries for law firms, life sciences and R&D and innovation teams, and the product line itself encodes the segmentation rather than merely describing it: Equinox ships as separate corporate and law firm editions, each with an enterprise variant, and the trademark side mirrors the same four-way split. Practice coverage is genuinely broad and specific, spanning patents, trademarks, designs, domain names, standard-essential patents, biosequences, chemistry, innovation management and legal translation, with named use cases including patentability, freedom to operate, invalidity, clearance, watching, prosecution, renewals and recordals. The life sciences entry names the products that serve that vertical individually. What is not stated anywhere located is where the product stops: no statement of firm size, portfolio size, minimum engagement, or the practice areas and jurisdictions the suite does not serve was found on the solutions pages or the home page on 1 September 2026.

Nothing located states whether client IP data is used to train or refine Anaqua's models, either way. Searched the AI initiatives and capabilities page, the guiding AI principles page, the security and hosting page, the global and California privacy policies and the website Terms of Use on 31 August 2026. The nearest statements point in opposite directions and neither answers the question: the Secure principle says no AI-driven insights or data are shared between unintended recipients, which is about segregation, while the Adaptable principle says the AI tools are maintained through regular training and refinement, without saying what they are trained on. There is no customer agreement published in which a prohibition could sit.

The commitment is clear and it sits in policy rather than in an agreement, which is what decides this value. The Trusted AI page lists it as the first of several commitments covering information a user provides for AI inference, described as a question, an invention description or an action query and acknowledged as often related to confidential client information. AI Policy Article 3 extends the same requirement outward, stating that third-party providers undertake not to use client data to train or improve any of their own AI or machine learning models. Separately the page says all Questel models are fine-tuned using public patent, trademark and design text or the company's own data. No published agreement containing a matching term was located: the Master Services Agreement and Data Processing Agreement are referenced in AI Policy Article 2 as signed with clients and neither is available on the site, so the prohibition cannot currently be read as contractual.

No located material states how long AI inputs or outputs are kept, whether a customer can configure that, or whether zero retention is available. The privacy policy addresses retention only for personal data, tying it to the length of the customer relationship plus a reasonable period afterwards, and noting that anonymised data may be kept without time limit; it does not reach patent text, classifier inputs or generated summaries. Searched the AI initiatives page, the guiding AI principles page, the security and hosting page and both privacy policies on 31 August 2026.

Customer control is stated directly and a concrete deletion window comes with it. The Trusted AI page says the customer controls the retention period, that zero data retention applies once a customer leaves, and that any deleted conversations are removed from Questel systems within 30 days unless retention is legally required. Zero retention is not offered as a setting available during use, which is why this is not the value above. Note that the Data Privacy Policy updated May 2026 takes the opposite approach for personal data, stating only that it is kept as long as necessary and directing the reader to email for details, so the specific commitment here belongs to AI inference data rather than to the wider account.

AQX maintains its own permission model rather than inheriting one from a document system. The security and hosting page describes record-level security with permissions based on each user account's role, alongside SSO and role-based security integrated with HR systems so that permissions are removed when employees leave, which reduces the drift that a separate permission model usually creates. Hosting is single-tenant, giving separation between customers as well as within them. What is not addressed is whether the AI features honour those record-level permissions at query time, which is the question that matters for the classifier and the summariser.

Access control is claimed and no enforcement detail is published. The privacy policy lists roles-based data access alongside encryption, firewalls, intrusion-detection software and physical access controls to data centres, and the Trusted AI page limits access to authorised employees and specified contractors. Neither describes a permission model, how it is administered, or whether it operates per user at query time. Nothing addresses walls or matter-level segregation inside a customer account, which is a live question here because Equinox is sold in dedicated law firm and enterprise law firm editions, and because Sophia is marketed as a cross-platform agent reaching across a range of Questel software. Searched the privacy policy, the AI Policy, the Trusted AI page and the law firm solutions entry on 1 September 2026.

Disclosure is addressed and notice is not. The privacy policy states that Anaqua may provide personal data to third parties to comply, when required by law, with court or administrative orders, civil or criminal subpoenas, warrants from law enforcement, regulatory requirements, mandatory governmental audits and grand jury investigations, and separately that it may be required to disclose personal data in response to lawful requests by public authorities. No commitment to notify the customer first, and no transparency report, was located. Note also that the policy governs personal data rather than the IP portfolio content a firm would most want notice about. Searched both privacy policies, the Terms of Use and the security page on 31 August 2026.

Disclosure is contemplated and notice is not. The Data Privacy Policy lists duly authorised public authorities, judicial and control, among the recipients of personal data in the framework of Questel's legal and regulatory obligations, and separately names regulated professions including lawyers and bailiffs who may intervene in guarantees, collection or litigation. No commitment to notify the customer before or after such a disclosure was located, no carve-out for where notice is legally prohibited, and no transparency report exists. The documents that would ordinarily carry this, the Master Services Agreement and the Data Processing Agreement, are referenced as signed with clients and are not published. Searched the privacy policy, the AI Policy, the Trusted AI page and the legal notice on 1 September 2026.

The product does not retrieve case law or statutes; the corpus behind it is patent and trademark registry data, and the sources are identified by office. Document auto-processing is stated to handle correspondence from the four largest patent offices, named as the US, WIPO, European and Japanese offices, with a separate integration to USPTO TSDR for trademarks. No licence, rights basis or update cadence is stated for any of it, and the corpus behind AcclaimIP patent search and analytics is not described on the pages read on 31 August 2026, the AcclaimIP property itself not having been opened this session.

The product does not retrieve case law; the corpus is patent, trademark and design registry text, and Questel is itself a database publisher, describing its own offering as a unique collection of IP databases. The Trusted AI page identifies what the models learn from, saying they are fine-tuned and optimised using public information covering patent, trademark and design texts, or Questel's own data. It also states a rights principle rather than a rights position, committing to acknowledge the source of training data and confirm permission to use it, giving open-source data or data provided by users as the examples. No licence or rights basis is stated for any individual source, no coverage figure such as jurisdictions or document counts was located on the pages read, and no update cadence is published. Checked the home page, the Trusted AI page, the AI Policy and the Orbit product entries on 1 September 2026.

No citator applies. The product does not return legal authority whose subsequent history could be checked; it manages registrations and summarises patent documents. The nearest analogue is registry currency rather than treatment, since docketing keeps status aligned with patent office records. Searched the AI initiatives page, the security page and the AQX Corporate entries on 31 August 2026.

No citator applies and the row is recorded rather than skipped. The suite manages and searches registered rights rather than returning legal authority whose subsequent history could be checked, so there is no treatment signal to surface. The nearest analogue is registry currency, which the product does address: Prosecution Pack is described as providing grant statistics and detailed legal status analysis, and the IP management system exists to keep portfolio status aligned with patent and trademark office records. That is the status of a registration rather than a judgement about whether an authority still stands. Searched the Orbit entries, Prosecution Pack, the AI Policy and the Trusted AI page on 1 September 2026.

Nothing located describes what any of the four AI features does when it cannot produce a reliable result. No abstention path, no confidence score and no low-certainty flag is published for the classifier, the summariser or document auto-processing. The guiding AI principles page commits to explaining how models perform, which is a commitment to describe behaviour rather than a description of it. Searched the AI initiatives page, the guiding AI principles page and the AQX Corporate IP operations entry on 31 August 2026.

Nothing located describes what any Questel AI feature does when it cannot support an answer. The Trusted AI page comes close to the subject and then states an intention rather than a behaviour, saying users need to be informed whenever there is AI use and uncertainty regarding the AI's responses and given the means to verify them, and that this will be achieved by providing sources and explaining the reasons behind answers. That is written in the future tense and describes what the company considers necessary, not what any product currently does. No abstention path, confidence score, grounding indicator or low-certainty flag is documented for Sophia, Qthena, the AI-Classifier, the drafting assistant or the office action assistant. Searched the Trusted AI page, the AI Policy, the AI assistants entries for patent and trademark, and the AI-in-IP solutions page on 1 September 2026.

Searched the AI Hallucination Cases database maintained by Damien Charlotin at HEC Paris, along with 2026 sanctions trackers and trade coverage, on 31 August 2026, on both the company name and the AQX product name. No court order, opinion or disciplinary record naming Anaqua or AQX was located. This is a statement about the public record rather than about the product, and the failure mode this signal tracks is a poor fit: the AI features classify and summarise patent documents and process office correspondence, and none of them generates citations to legal authority for filing.

Searched the AI Hallucination Cases database maintained by Damien Charlotin at HEC Paris, together with 2026 sanctions trackers and trade coverage, on 1 September 2026, on the company name and on the product names Orbit, Sophia and Qthena. No court order, opinion or disciplinary record naming Questel or any of its products was located. This is a statement about the public record rather than a finding about the product. The failure mode fits this suite only partly: most of the AI here classifies, searches or manages registry data, but the generative patent drafting and office action response assistants do produce text that enters patent office proceedings, so the risk is not entirely absent even though it sits outside the litigation filings this tracker principally captures.

No engagement with ABA Formal Opinion 512, USPTO practitioner guidance on AI, or any state bar opinion was located. The guiding AI principles page names an external framework, the EU AI Ethics Guidelines, and commits to adapting as AI regulation evolves, but that is technology regulation rather than the professional responsibility guidance binding the firms Anaqua sells to. Searched the AI initiatives page, the guiding AI principles page, the AQX Law Firm entry and the resource library on 31 August 2026.

Professional responsibility is engaged in general terms with no guidance named. AI Policy Article 5 states that the use of Questel's Services remains under the professional responsibility of each client, that the Services are not intended to replace or substitute the expertise and judgment of legal professionals, and that clients should seek professional legal advice before relying on AI-generated outputs. That is a real acknowledgement of the obligation and it names no source for it. No reference to ABA Formal Opinion 512, USPTO practitioner guidance on artificial intelligence, EPO or national patent attorney conduct rules, or any bar opinion was located, which is a notable gap for a vendor selling prosecution assistants into law firms across roughly 30 countries. The EU AI Act is named in Article 1, but that is technology regulation binding Questel rather than the professional conduct guidance binding its buyers.

WiseTime produces a reviewable per-matter time record: users see a timesheet and can adjust durations, activity codes, matter references and the activity description, either manually or using the AI Narrative feature that drafts invoice summaries. That is a usable record a firm could interrogate, and the AI-drafted element is identified as such to the timekeeper. No guidance on how AI-assisted work should be billed or disclosed to a client accompanies it. The marketing runs the other way from a savings claim, describing WiseTime as increasing the amount of billable time captured, and the client-side implication of that is not addressed.

Time savings are claimed and the client's side of the bill is never addressed. The home page invites the reader to imagine how much time could be saved with an AI-driven integrated IP management system, and the resource library carries pieces on saving time in prosecution work and on how Equinox Law Firm helps a firm save time and build relationships. Nothing accompanies this on how a firm should bill for or disclose AI-assisted drafting, office action responses or search work. Worth noting that Questel sells invoice management, cost management and fee audit services, but those address the client's own IP spend with outside agents rather than the treatment of AI-assisted time on a law firm's invoice, and no audit record of AI-assisted work was located for any product.

No subprocessor list, data processing addendum or client-facing disclosure pack is published, and no route to request one was located. What a firm can point to is thin and partial: Microsoft Azure is named as the hosting provider, Microsoft Azure AI Document Intelligence is named as the service behind prior art import, and the privacy policy commits to EU Standard Contractual Clauses for onward transfers and describes categories of recipients without naming them. The model providers behind the classifier, the generative summariser and the AI narrative generator are not identified, so the question an outside counsel guideline actually asks cannot be answered from public material. Searched the security page, both privacy policies, the AI pages and the Terms of Use on 31 August 2026.

A model provider list is published and a subprocessor list is not, so a firm can answer half of what a client's AI clause asks. The Trusted AI page names OpenAI as the generative provider, links the OpenAI API Platform, and identifies the open-source families Mixtral, Falcon and Llama 2 as additionally used and fine-tuned in-house, which is the disclosure that carries this value. Against that, no subprocessor list covering the products exists: the Data Privacy Policy's third-party section names thirteen website marketing and analytics tools and no processor touching customer IP data. No data processing addendum or client-facing disclosure pack is published and no route to obtain one without an agreement was located. AI Policy Article 2 claims full transparency on third-party providers as outlined in the personal data policy, and on the surfaces read that claim is not met.

Nothing located addresses producing an AI-use disclosure or verification certification. No export covering which model produced a summary or classification, what it drew on, or who reviewed it is described for any of the four AI features. Searched the AI initiatives page, the guiding AI principles page, the security page and the reporting and analytics entry on 31 August 2026. The obligation fits this product poorly, since its output supports portfolio management and prosecution administration rather than court filings, but the row is recorded rather than skipped.

Nothing located addresses producing an AI-use disclosure or a verification certification. The Trusted AI page states a principle adjacent to it, that Questel needs to clearly disclose when content has been autonomously generated by an AI, but that is user-facing labelling described as an obligation the company recognises rather than an exportable record, and no product is said to implement it. No export tying a passage to the model that produced it, the sources it drew on, or the person who reviewed it was described for Sophia, Qthena, the drafting assistant or the office action assistant. Searched the Trusted AI page, the AI Policy and the patent and trademark AI assistant entries on 1 September 2026. The obligation fits prosecution work imperfectly, but the row is recorded rather than skipped.