IP Author vs XLSCOUT: how they compare in 2026

Every module is generative and there is no underlying platform that would function without them. Drafting produces claims, specification and figures from a disclosure, prior art search runs semantically from the disclosure text rather than from Boolean queries, office action responses are generated with structured arguments and claim mapping from the application history, classification is model driven, and the Evidence of Use assistant is an analysis product. Multimodal ingestion is a real model capability rather than a feature claim: the platform is stated to read images, tables and diagrams inside Word documents, PDFs and slide decks, and to generate flowcharts and block diagrams where a disclosure lacks them. Remove the models and there is no product at all, not even a database, since the corpus is searched rather than owned. Fourth A on this axis in ip-and-patents out of five records, with PatSnap the only exception and that on age.

The models are the entire product and the naming says so: seven of the nine modules carry LLM in the product name. Novelty Checker LLM, Invalidator LLM, ClaimChart LLM, PatDigger LLM, Standigger LLM, TechScaper LLM and Drafting LLM each exist only as a generative or retrieval-model application, and Ideacue generates invention concepts outright. Underneath them sits ParaEmbed, an embedding model XLSCOUT fine-tuned on patent and technology data and published openly on Hugging Face, combined with in-house RAG models and Azure endpoints. There is no document management system, no docketing, no renewals engine and no services layer beneath the AI that would survive its removal; strip the models out and nothing remains that a buyer could use. This is the second A on this axis in the ip-and-patents lane, alongside IPRally, against three incumbents at C.

Grounding is shown to the user at several points and one accuracy claim is published with a comparator but no method. The review surface is the substantive part: similarity scores against the invention, AI generated novelty summaries, explicit explanations of overlaps and differences between the invention and each cited reference, highlighted passages, and side by side disclosure comparison. That lets a practitioner test a retrieval result rather than accept it, and it is more than most records in this category describe. The published accuracy claim is that compared to a human searcher the tool returns similar or the same results 19 out of 20 times. Naming the comparator as a human searcher is meaningful and is credited, and everything that would make it checkable is missing: no sample size, no evaluator, no definition of similar, no date and no methodology. Held at B on that basis. Recorded for the next reader: the vendor also publishes its own comparison content benchmarking itself against named competitors, which is self published competitive material and is not credited as evidence.

Grounding is documented, the retrieval method is described, and figures are published without a test set behind them. Output maps to real patent documents a reader can open: novelty reports carry key feature mappings against the references, Invalidator LLM performs ParaEmbed-based contextual mapping of prior art against the claims and surfaces the top fifty most relevant references, and ClaimChart LLM maps product features to claims. The retrieval mechanism is named rather than gestured at, and unusually the underlying embedding model has been open-sourced on Hugging Face, so an outsider can evaluate part of the stack directly, which is more than any other vendor in this lane offers. Four numbers are published, including a candid 60 per cent accuracy figure for PatDigger LLM in identifying top licensing opportunities and a 95 per cent noise reduction claim for TechScaper. None carries a described test set, a method or a date. The master agreement is franker than the marketing, defining automated patentability and invalidity searches as first-pass results delivered on an as-is-where-is basis that shall in no manner be a comprehensive or full search, and clause 4.5 disclaims any warranty as to accuracy, completeness or correctness.

An explicit oversight commitment, stated twice in separate surfaces and in plain terms. The product material states that users retain full review, editing and decision making authority at every stage of drafting, and the published FAQ states that the platform keeps attorneys and agents in full control of the final work product. Two independent statements of the same commitment, one of them in a FAQ answering a direct question, is a clearer position than the implied oversight most records in this pull offer. Output is framed as a first draft and first pass response throughout rather than as finished work. Held at B rather than A because nothing is bounded operationally: no statement of what the system does unattended, no confidence threshold, no escalation behaviour, and no description of what happens in the agentic configuration where the platform connects its tools to external orchestration frameworks, which is precisely the setting where a stated commitment to full user control is hardest to maintain and is not addressed.

Automation is the entire pitch and no control structure is published to sit against it. The modules are described as automated drafting, automated novelty reports and automated invalidation reports, with Ideacue accelerating ideation tenfold, and nothing located states what runs unattended, where a reviewer enters, what the system does at a low-confidence threshold, or what happens after it is wrong. The nearest thing to a control is contractual rather than functional: the master agreement defines the search outputs as a first pass that is in no manner comprehensive, which tells a buyer the work is unfinished without describing any mechanism inside the product for finishing it. One module names a human step, the AI-assisted source code review being described as expert-validated, and that is a service design rather than a published oversight model for the platform. Searched the home page, the security and privacy controls page, the SOC 2 page, the master agreement and the module descriptions in the site navigation on 1 September 2026. This matters because Drafting LLM produces patent applications that a practitioner will file.

Nothing located. No customer or firm is named anywhere in vendor material, no case study, no usage figure, no funding announcement, no independent analyst placement, no award and no third party review were found. The vendor publishes a substantial content library including competitive comparisons and benchmark posts, and all of it is self authored, so the evidence base for this record is entirely the vendor describing its own product. That is the weakest operational position in ip-and-patents by a distance: Patlytics names Am Law 100 penetration and a Quinn Emanuel partner, Solve Intelligence names DLA Piper, Siemens and Finnegan, DeepIP publishes dated funding and traction figures, and PatSnap names NASA, Tesla and MIT in independent trade coverage. Checked the home page, the product pages, the FAQ, the blog library and the site navigation on 29 Aug 2026. Recorded as a documented absence across those surfaces rather than as a judgement about adoption, which cannot be assessed from public material.

Logos and unattributed testimonials stand in for evidence, and one contract clause makes that worth reading carefully. Four customer logos appear, Mitsubishi, NTT, IIT Madras and IDAIR, against a claim of more than 200 clients worldwide with no basis stated. Five testimonials are published and every one is anonymous, attributed only by role and sector: a patent attorney at a US law firm, an IP manager at a machinery manufacturer, a senior IP manager in industrial automation, a senior innovation manager at an engineering company and a partner at a US law firm. No organisation is named against any quote, no quote carries a date, and the published figures for time saved and efficiency gained are product claims rather than measured customer outcomes. Clause 10.5 of the master agreement is relevant to how much weight this evidence carries: customers agree in advance to participate in marketing activities including writing references and case studies, speaking at XLSCOUT events and appearing in webinars. A case studies library exists in the navigation and was not opened on 1 September 2026, so a named dated outcome was neither found nor excluded.

One genuinely thoughtful confidentiality feature, and no professional confidentiality position. The feature is worth naming because no other record in this pull has anything like it: prior art results export in masked versions that withhold filing dates and raw excerpts, stated expressly so findings can be shared with clients or outside counsel without oversharing. That is a product designed around the reality that a prior art report circulated too widely is itself a disclosure risk, and it addresses onward sharing rather than only storage. Against that, nothing located addresses attorney client privilege, work product, or the specific consequence this category carries, being that exposure of an unpublished application can affect patent rights rather than merely creating a data incident. Both Patlytics and Solve Intelligence engage that directly and both reached A. Research limitation stated: a Trust Center and a privacy policy are both published on the site and neither was read in this pass, so this grade is a correction candidate.

The core commitments are contractual rather than promotional, which is rare in this lane, and two limbs are still missing. Clause 6.4 of the master agreement undertakes that customer data is not used to train, retain or improve the models of XLSCOUT or of its third-party suppliers, and clause 6.3 confirms the customer retains all right, title and interest in its data. Clause 7 is a mutual confidentiality regime with a defined category of Confidential Information and a restriction on disclosure to employees, contractors and third-party suppliers who are themselves bound. Tenant separation is documented on the SOC 2 page: each customer environment sits in a dedicated trust zone described as a patented isolation approach, customer data is separated using unique encryption keys, and role-based access with least privilege is applied. What is absent is segregation below the tenant, so nothing addresses walls between users, teams or matters inside one account, which matters because the published testimonials come from US law firms. Privilege and work product are never mentioned on any surface read, and the retention picture is genuinely muddled, as set out on the retention signal.

Not located. The product drafts complete patent applications and generates office action responses spanning statutory grounds under sections 101, 102, 103 and 112, all of which are filed under a registered practitioner's signature, and it offers a self serve 14 day trial which means the buyer is not necessarily an attorney. The vendor states that attorneys and agents retain full control of the final work product, which is a product positioning statement graded on the Autonomy axis and is not a professional responsibility framework. No reference to USPTO Rules of Professional Conduct, 37 CFR, duty of competence or any bar or patent office guidance was located. Compare Patlytics at B in this category, which names those authorities directly. Checked the home page, the product pages, the FAQ, the blog library and the site navigation on 29 Aug 2026.

A real limit on what the product is sits in the agreement rather than in a footer, and the professional dimension around it is untouched. The master agreement defines Automated Patentability search and Automated Invalidity as first-pass results obtained by running the proprietary algorithm on the database on an as-is-where-is basis, which shall in no manner be a comprehensive or a full search. That is a contractual statement of what the tool does not do, binding on the vendor and readable before signing, and it is a more concrete scope limit than most vendors publish anywhere. Clause 4.5 adds an express disclaimer of any warranty as to accuracy, completeness or correctness. Against that, nothing located states that XLSCOUT is not a law firm or does not provide legal advice, no jurisdiction limit is named, and there is no engagement at all with a practitioner's supervision or competence duties. The gap is sharpest on Drafting LLM, which produces patent applications, and on Ideacue, which is sold to R&D and innovation teams who are not practitioners.

Nothing published about how the models are governed, evaluated or monitored. No AI policy, no model card, no bias or fairness testing, no evaluation methodology, no accuracy monitoring beyond the single unmethodised search claim, no drift statement, no named governance body, no ISO 42001 and no EU AI Act positioning were located. The vendor publishes commentary on AI industry developments and on intellectual property questions raised by AI systems, which engages the subject matter externally rather than disclosing anything about its own models. This is the weakest governance position in the category alongside DeepIP, against Patlytics holding an ISO 42001 certificate and Solve Intelligence publishing a continuous evaluation process. Checked the home page, the product pages, the FAQ, the blog library and the site navigation on 29 Aug 2026.

No governance position of any kind is published for a platform whose entire output is model-generated. There is no responsible AI page, no AI policy, no ethics statement, no ISO 42001 or EU AI Act position, no accountable owner named, no pre-release testing regime for model behaviour, and no disclosure about uneven output across technology areas or document types. The full site navigation was read on 1 September 2026 and lists every page the company publishes, covering the nine modules, about, board members, team, advisors, partners, MoUs, careers, pricing, a Prevention of Sexual Harassment policy, two security pages, blogs, press releases, events, webinars, case studies, white papers, contact, FAQs, terms of service, privacy policy, content security policy, research and AI collaborations, and data coverage. **No governance or responsible AI entry exists anywhere in it.** The two nearest items are not governance: a founder's video on global AI governance at the United Nations is thought leadership about the field rather than a statement about this product, and the research collaborations with Mitacs, Hugging Face and four universities are development partnerships. Clause 6.4's no-training undertaking is a data commitment and is credited on the confidentiality row.

No stewardship position located in any surface read. Nothing states whether invention disclosures, draft applications, search histories or generated work product are used to train or improve models, no retention period is published, and no deletion right is described. The content at stake is the most sensitive in this category, being unpublished invention disclosures uploaded before filing, and the vendor's own masked export feature demonstrates awareness that disclosure control matters, which makes the silence on its own handling more noticeable rather than less. Research limitation recorded prominently rather than buried: a Trust Center and a privacy policy are both published on this site and neither was read in this pass, and either is the surface where a training or retention statement would ordinarily sit. This grade is flagged as the strongest correction candidate on the record and should not be read as an established absence of policy, only as an absence across the pages actually checked, being the home page, the product pages, the FAQ and the blog library on 29 Aug 2026.

Most of the ground is covered with real specificity and the two gaps this band names are both present. What is published: encryption at rest and in transit with unique per-customer keys, role-based access control with least privilege and periodic review, third-party penetration testing by external firms alongside internal engineers, static and dynamic application security testing including open source libraries, OWASP Top 10 development practice, background checks and signed NDAs for all employees and contractors, continuous platform monitoring, and a customer-facing statement that data can be deleted at any time. The SOC 2 Type II audit provides independent assurance over that control set. What is missing is a named subprocessor list, since the master agreement discloses at clause 6.2 that inputs are shared with trusted third-party suppliers without naming any of them and the security page names AWS and Microsoft Azure followed by the phrase and others; and any incident or breach practice, with nothing located stating notification timing, contents or route. No retention period is published for customer content, and the Content Security Policy, which the agreement makes the governing document for storage, was not opened on 1 September 2026.

No published position located. Nothing was found on liability for AI output, warranty, service levels or remedy. The exposure is the category's own and is concentrated by the speed claim: a complete first draft in fifteen minutes compresses the interval in which errors would ordinarily surface, and claim scope lost at grant is not recoverable. The office action module covering statutory rejection grounds carries the further risk that a mischaracterised traversal argument creates prosecution history estoppel. The vendor's answer to all of this is the stated retention of full attorney control, which allocates responsibility to the user by design rather than by contract, and no contractual position is public. Checked the home page, the product pages, the FAQ and the site navigation on 29 Aug 2026.

This is the only real published liability position in the ip-and-patents lane and it is more complete than most of the corpus. The master agreement publishes an indemnity with named scope: clause 8.1 has XLSCOUT defend and indemnify the customer against third-party claims arising from breach of the warranties at 4.1 and 4.3, which cover its right to grant the licences and its authority to contract, and clause 8.3 sets out the remedy path if an injunction is sought, being procure, modify or terminate with a pro-rata refund. Caps are stated and differentiated at clause 5.2, limited to the fees paid for the service in question, or for matters unrelated to a service the aggregate fees paid in the preceding twelve months. Carve-outs from the cap are named at clause 5.3, covering the indemnity obligations and the customer's IP infringement. Mutual exclusion of indirect and consequential loss sits at 5.1. What stops this reaching the top band is the subject the axis actually asks about: the indemnity covers title and authority, not output, clause 4.5 expressly disclaims any warranty as to the accuracy or correctness of the licensed materials, and no insurance position is published. A buyer therefore knows precisely who bears the loss when a draft or a search is wrong, and the answer is the buyer.

Internal integration is real and external integration is thin. Within the platform, prior art search is deliberately integrated into the drafting flow rather than sold as a separate product, so references surfaced during search carry forward into claims and specification without re-keying, and results export as Word reports or Excel sheets. Externally, the vendor states that its tools including patent search, claim charting, whitespace analysis and office action response connect to agentic orchestration frameworks, which is a forward looking integration surface and is the only one described. What is absent is everything a firm would ask for: no IP docketing system, no patent management platform, no document management system, no Microsoft Word integration despite drafting being the core product, and no API documentation was located. For a prosecution tool, the docketing gap is the material one, and DeepIP names IP management platform integration in this same category.

No integration into the systems legal work already lives in was located, and the product stands alone. The complete site navigation was read on 1 September 2026 and contains no integrations page, no API or developer entry, no documentation index and no partner integration listing; the Authorized Partners page is a reseller and channel listing by its title rather than a technical one. Nothing names a connection to an IP management or docketing system, a document management system such as iManage or NetDocuments, a patent office filing system, Word or Outlook. The modules are described as a connected suite, but that is integration among XLSCOUT's own components rather than into a firm's estate, and the master agreement's references to a Delivery Method and Documentation are contractual placeholders that describe no interface. A customer moves work into the platform and exports results out of it. Checked the home page, the AI-powered modules entry, the security pages, the master agreement and the full navigation.

Nothing located. No hosting provider is named, no region or data residency commitment is published, and no single tenant, dedicated or on premise option is described. The gap is material in this category because unpublished applications engage foreign filing licence requirements and export controls in several jurisdictions, and a practitioner would need to know where a disclosure is processed before uploading one, particularly given the platform supports prior art search across more than 100 jurisdictions and therefore serves a geographically distributed user base. Compare DeepIP at A for on premise deployment and Solve Intelligence at A for customer selectable jurisdiction, both in this category. Research limitation: the Trust Center was not read in this pass and is where a residency statement would sit. Checked the home page, the product pages, the FAQ and the site navigation on 29 Aug 2026.

The tenancy model and the residency choice are both stated, which is more than the rest of this lane manages, and the processing question is left open. The SOC 2 page describes a modern multi-tenant cloud architecture with complete customer isolation, each customer environment held in a dedicated trust zone under what the company calls a patented isolation approach, with unique encryption keys separating customers from each other and from XLSCOUT itself. The security page states that clients can choose to have their data stored on either United States or European servers and names AWS and Microsoft Azure among the providers relied on, followed by the phrase and others, which leaves the list open. The limb that holds this below the top band is the one this band reserves: residency is offered for storage without processing location being addressed separately. That distinction has weight here because the same page states that inference runs partly through Microsoft Azure endpoints, and nothing says whether a customer who selects EU storage also gets EU inference.

A Trust Center is published and its contents were not read, which is the whole of what can honestly be said. The site navigation carries a dedicated Trust Center alongside a privacy policy, so the vendor has built a trust surface rather than treating security as a marketing line, and under the three tier test a published trust surface is not an absence. No certification of any kind was located in the pages actually read: no SOC 2 of either type, no ISO 27001, no ISO 42001, no named auditor, no penetration testing partner and no examination date. Held at C rather than D on the same reasoning applied to TrialView earlier in this pull: grading D would assert an absence on the single surface most likely to contradict it, which is the error the trust portal sweep corrected twenty five times, and repeating it after writing the rule would be worse than the original. Flagged as a correction candidate in both directions, since reading the Trust Center could establish certifications that raise this grade or confirm their absence and lower it. Every category peer publishes at least one named certification.

The attestation is real, specific and independently named, and it is not obtainable. XLSCOUT publishes a dedicated page stating that it completed the AICPA Service Organization Control 2 Type II audit and naming the auditor, Prescient Assurance, described as a registered public accounting firm in the US and Canada, with the outcome given as an unqualified opinion. Naming the auditor puts this above most of the corpus, where a badge or a bare standard is the norm. The page also details the control areas the audit covers across personnel, development, testing and cloud security. What is absent is access and currency: no examination period or report date appears anywhere, so a buyer in 2026 cannot tell whether the attestation is live, and no route to request the report, no portal and no questionnaire pack was located. The related ISO 27001 claim on the security page is honestly qualified with an asterisk reading that it applies to the partners rather than to XLSCOUT, which is a disclosure most vendors would not volunteer. Note also that the SOC 2 page carries boilerplate about digital identity offerings and managing billions of digital identities that does not describe this business.

A provider is named, in passing, in a two year old marketing comparison, which is a genuine disclosure of a kind and not a maintained one. In a blog post comparing itself to general purpose tools the vendor states plainly that IP Author also used ChatGPT and that the difference lies in prompt engineering, the patent search engine and user experience. Naming the underlying provider outright is rare on this index and it is credited. What stops this reaching the grade Onspring received for the same act is that the disclosure is incidental rather than deliberate, appears in dated April 2024 competitive content rather than in current product or trust material, is not repeated anywhere current, and gives no model, version or scope, so a buyer cannot tell whether it still holds after two years of the platform being rebuilt around newer capabilities. Separately, the vendor names agentic frameworks its tools connect to, which discloses an orchestration relationship rather than the model layer. No subprocessor list was located.

The architecture is described and the third-party model is the piece left unnamed. The security page states that XLSCOUT uses a combination of its own proprietary custom fine-tuned LLMs, embedding models and RAG models together with Microsoft Azure endpoints to draft patents, identify infringements and search prior art, which distinguishes what it built from what it calls out to. Its own component is not merely named but published: ParaEmbed, an embedding model fine-tuned on patent and technology data, was open-sourced on Hugging Face and reached a second version, so a customer can inspect the actual retrieval model rather than take a description of it, which no other vendor in this pull offers. Two things hold it below the top band. The generative model reached through the Azure endpoints is never identified, so the question of whose model reads an unfiled invention disclosure is unanswered, and naming the cloud is infrastructure rather than model identification. And no commitment to notify customers when the model set changes was located on any surface read on 1 September 2026.

Two self serve elements exist and no figure was read. A dedicated pricing page is published in the site navigation, and a 14 day free trial is available directly from the website with the FAQ confirming it can be started without going through sales. A published pricing page plus a self serve trial is a materially more open commercial posture than the demo request pattern that governs every other record in this category, where Patlytics, Solve Intelligence and DeepIP all publish nothing and PatSnap publishes only a tier structure through an analyst listing. Held at C rather than higher because the pricing page contents were not read in this pass and no price, range, tier structure or unit of charge was captured, so the openness is established and the figures are not. Flagged as a correction candidate in the upward direction, and a record where the grade may understate the vendor.

A page titled XLSCOUT Pricing exists and contains no pricing information at any level. It offers the line that pricing models are flexible and tailored to diverse needs, a free trial request, a demo booking, and a lead capture form whose own text says that for detailed pricing information the reader should fill out the form and the sales team will provide details tailored to the organisation. No rate, no band, no floor, no currency, no unit of charge, no tier or package names and no statement of what implementation adds appear on it. The modules are individually named across the site and are presumably licensed separately, but no packaging, bundling or per-module commercial structure is published, so the shape of the offer is not visible either. This is a stronger finding than an absent pricing page, because the vendor has built the page and used it solely as a form. Checked the pricing page, the home page, the master agreement, whose fees are deferred entirely to an unpublished Cover Sheet at Annexure 1, and the full site navigation on 1 September 2026.

Coverage is enumerated on the dimension that matters most for prosecution, which is the statutory grounds the tool can actually handle. Office action support is stated across 102 and 103 prior art rejections, 101 subject matter eligibility, 112 written description, enablement and definiteness, and double patenting, with a traversal strategy always produced. No other record in this category enumerates rejection grounds, and it is the enumeration a prosecution practitioner would check first, because a tool that cannot handle a 101 rejection is unusable in software and biotech practice. Search coverage is stated across worldwide patent and non patent literature and technical standards spanning more than 100 jurisdictions, filterable by jurisdiction, CPC and IPC classification and assignee. Drafting is stated to align with USPTO, EPO and other office requirements, and technical domains include life sciences and chemistry with named sequence handling. Held at B rather than A because the drafting jurisdiction list stops at two named offices and others, no corpus date range or update lag is published for the search side, and the technical domain list was not fully enumerated in the material read.

Practice coverage is documented module by module with real precision, and segment coverage is thin by comparison. Nine distinct workflows are named and separately described: prior art and novelty search, invalidity search, infringement analysis through claim charts, portfolio licensing and monetisation, standard-essential patent essentiality and standard mapping, technology and competitor monitoring, patent drafting, ideation and continuation strategy, and source code review. Two analytics dashboards extend that into sustainability, mapping Sustainable Development Goals and green transformation technologies to the patent literature. Language coverage is stated as English, Japanese, Chinese and German, and the company operates from five countries. Who it is for is answered in a single FAQ sentence listing patent attorneys, legal teams, inventors, research and development departments and businesses protecting intellectual property, with no segment pages behind it and nothing on firm size, portfolio size or in-house versus firm fit. The one boundary that is stated is a capability limit rather than a coverage limit, the agreement's first-pass definition of what the searches are not.

Silent. The quoted phrase is the vendor's clearest statement about the relationship between the user and the system, and it concerns control over the work product rather than use of the data. No statement in either direction was located on whether invention disclosures, draft applications, search histories or generated work product are used to train or improve models. The content at stake is unpublished invention disclosures uploaded before filing, which is the most disclosure sensitive material in this category, and the vendor's own masked export feature shows it understands that disclosure control matters. Recorded as silent, not as a negative commitment. Research limitation stated prominently: a Trust Center and a privacy policy are both published on this site and neither was read in this pass, and either is where such a statement would ordinarily sit, so this value is the strongest correction candidate on the record. Checked the home page, the product pages, the FAQ and the blog library on 29 Aug 2026.

The prohibition is in the master agreement rather than on a policy page, and it reaches the suppliers as well as the vendor. Clause 6.4 of the XLSCOUT General Services Terms and Conditions undertakes that the data, defined earlier in clause 6.2 as including both input and generated output, is not used to train, retain or improve the models of XLSCOUT or the third-party suppliers, and adds that any sharing is done under confidentiality obligations. The marketing carries the same position, with the security page stating that XLSCOUT does not use customer data to train its LLMs and the FAQ saying the company does not train its AI models on customer-provided inputs, so the two surfaces agree here. Read alongside clause 6.1, which grants XLSCOUT a licence to use inputs and outputs exclusively to perform the services, the permitted purpose is service delivery and training is expressly outside it.

Not addressed in any surface read. No retention period is published for uploaded disclosures, generated drafts, search results or office action responses, and no deletion right is described. Invention Management is offered as a module, which implies durable storage of disclosures as a product function, and nothing states for how long or under what terms. Same research limitation as the training signal applies: the published Trust Center and privacy policy were not read in this pass and either is where a retention statement would sit. Checked the home page, the product pages, the FAQ and the blog library on 29 Aug 2026.

The marketing and the agreement point in different directions and the agreement governs. The security page runs a heading reading Zero Data Retention or Training and states that the model does not retain any user data and that the customer can delete data at any time. The master agreement grants XLSCOUT a worldwide sub-licensable licence at clause 6.1 to collect, process, store and otherwise use the customer's inputs and the outputs generated from them, and clause 6.2 spells out that this includes the right to store that data as per the content security policy. Both can be true, since a model not retaining a prompt is a different thing from a platform storing a project, but a buyer reading the headline would not expect the storage licence. No retention period is stated in either place. The Content Security Policy, which the agreement makes the governing document for storage, was not opened on 1 September 2026 and is where a period would sit if one is published.

Not addressed. No permission model, matter level access restriction or tenant segregation description was located. The conflicts scenario in patent practice is concrete, since a firm may prosecute in the same technical field for competing clients, and nothing published describes what separates one matter's disclosure from another inside the platform or what an Invention Management module exposes across a firm. One adjacent feature exists and is credited on the Privilege axis rather than here: masked export withholding filing dates and raw excerpts, which controls what leaves the platform rather than who can reach what inside it. Checked the home page, the product pages, the FAQ and the site navigation on 29 Aug 2026.

XLSCOUT documents its own permission and isolation model rather than inheriting one, and it operates at the tenant rather than the matter. The SOC 2 page states that all customer cloud environments and data are isolated using a patented isolation approach with each customer environment held in a dedicated trust zone to prevent accidental or malicious co-mingling, that each customer's data is separated from other customers and from XLSCOUT using unique encryption keys, and that role-based access controls and least privilege are applied with access reviewed and revoked as needed. The product does not index a firm's own document estate, so the classic wall question bites less here than on a document-based tool. What is not addressed is separation inside a single customer account, between users, teams or matters, which is the level a law firm would need and the published testimonials include two US law firms.

Not addressed. No government or law enforcement request clause, no commitment to notify a customer before producing their data, and no transparency report were located. The category stake applies: the platform holds unpublished invention disclosures and pre filing drafts, and disclosure of an unpublished application to a third party can bear on novelty and prior disclosure questions. Same limitation as above, with the privacy policy unread. Checked the home page, the product pages, the FAQ and the site navigation on 29 Aug 2026.

The commitment sits where this signal's evidence usually lives, in the confidentiality section of the master agreement rather than in a privacy policy. Clause 7.3 provides that where the receiving party becomes legally compelled or is requested by a regulatory body to disclose confidential information, it will give the disclosing party prompt written notice unless legally prohibited, expressly so that the disclosing party can seek an order preventing disclosure or another remedy, with a commitment to cooperate if such an order is sought. If no order is obtained, disclosure is limited to the portion legally required in the reasonable opinion of counsel. The obligation is mutual, and because customer inputs and outputs fall within the defined Confidential Information it reaches the material a firm would care about rather than only personal data. No transparency report was located on 1 September 2026.

Named by type and jurisdiction count, with no source, licence basis or currency stated. The search corpus is described as worldwide patent literature, non patent literature and technical standards across more than 100 jurisdictions, filterable by jurisdiction, CPC and IPC classification, assignee and publication details. Naming non patent literature and technical standards separately is worth crediting, because prior art is not confined to patents and standards documents in particular are both critical to prior art analysis and encumbered by publisher licensing. That last point is exactly what is unaddressed: no database, publisher or standards body is named, no licensing basis is stated for the non patent literature or standards, no date range is given, and no update lag or refresh frequency is published despite results being described as returning in near real time. A practitioner relying on a novelty summary cannot determine what was searched or how current it was.

Coverage is described by scale and geography with no identification of the underlying sources. The home page states patents from more than 100 countries alongside non-patent literature described as bibliographic and a jurisdiction count, though the numerals themselves render as script-driven counters that did not resolve on fetch. No database, aggregator or patent office is named as the source of any of it, no licence or rights basis is stated, and no update cadence is published. What is identified is the training rather than the corpus: ParaEmbed is described as an embedding model fine-tuned on patent and technology data and is published on Hugging Face. A Data Coverage page exists in the site navigation and was not opened on 1 September 2026, which is where source identification would sit if it is published anywhere.

Not addressed, and applicable in both forms this category presents. The office action module generates traversal arguments against statutory rejections under sections 101, 102, 103 and 112, which in practice involves citing and distinguishing case law, and nothing states whether cited authority is checked for current treatment before it enters a filing. The patent specific analogue applies to the search side, where the question is legal status: whether a surfaced reference is granted and in force, lapsed, expired, amended in reexamination or invalidated in post grant proceedings. Filtering is offered by publication details, which is bibliographic rather than status aware. Fifth consecutive record in this category to leave this signal unanswered, and the pattern is now unbroken across every ip-and-patents vendor built.

No citator applies and the row is recorded rather than skipped. The platform retrieves and analyses patent documents and non-patent literature; it does not return legal authority whose subsequent history could be checked, so there is no treatment signal to surface and no good-law question in the sense this signal asks. The nearest analogue would be the legal status of a retrieved patent, and nothing located states whether status or validity events are surfaced against references in a novelty or invalidity report. Searched the module descriptions, the home page, the FAQs and the master agreement on 1 September 2026.

Not addressed as a behaviour, with a partial answer on the search side that stops short. Similarity scores are surfaced against each cited reference alongside novelty summaries explaining overlaps and differences, which gives a user a graded signal about retrieval relevance and is genuinely useful. It is a relevance score for a returned reference, not a confidence signal about the system's own output: nothing states whether a low confidence generated claim is flagged, whether the drafting module will decline where a disclosure is too thin to support a claim, or what happens when the office action module cannot construct a traversal, despite the vendor stating a traversal strategy is always produced, which is a commitment to always answer and the opposite of an abstention path. Checked the home page, the product pages, the FAQ and the blog library on 29 Aug 2026.

Nothing located describes what any module does when it cannot support a result. Relevance ranking exists, with Invalidator LLM surfacing the top fifty most relevant references, but ranking is inherent to retrieval and no confidence or grounding score is exposed to the user as a signal, and no abstention or no-answer path is documented for the drafting, ideation or claim-charting modules. The FAQ acknowledges the general point, saying that no system can guarantee 100 per cent accuracy, and the master agreement defines the search outputs as first-pass and as-is-where-is, but both place the burden on the reader rather than describing a behaviour of the system. Searched the home page, the nine module entries in the navigation, the FAQs, the security pages and the master agreement on 1 September 2026.

None located, with the instrument named. General web searches combining the vendor and product names with court, order, sanction, fabricated citation and patent terms returned nothing on 29 Aug 2026, and no named docket database, patent office record system or court record tracker was searched. Recorded as a statement about what this search found, not as a clearance. The exposure shape is a mischaracterised prior art reference or a defective case citation inside an office action response filed at a patent office, which would surface in a prosecution file wrapper rather than in a published opinion. Worth noting that no customer of this vendor is publicly identifiable, so even a proper docket instrument would have little to anchor a targeted search against.

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 module names including Novelty Checker LLM and Drafting LLM. No court order, opinion or disciplinary record naming XLSCOUT or any of its modules was located. This is a statement about the public record rather than a finding about the product. The failure mode is a partial fit: the search modules return real patent documents from an index rather than generating citations, but Drafting LLM produces patent application text and ClaimChart LLM produces claim charts, both of which can reach a patent office or a dispute.

Not addressed. No named ethics opinion, no USPTO Rules of Professional Conduct reference, no 37 CFR citation, no duty of competence discussion and no bar guidance was located. The vendor publishes an extensive practitioner facing content library covering drafting workflows, tool comparisons and AI industry commentary, and that material addresses capability and efficiency without reaching the professional rules governing the practitioner who signs the filing. Fourth of five records in this category at this value, with Patlytics the only exception. Checked the blog library, the home page, the product pages, the FAQ and the site navigation on 29 Aug 2026.

No engagement with professional responsibility guidance of any kind was located. Nothing addresses ABA Formal Opinion 512, USPTO practitioner guidance on artificial intelligence, EPO or other national patent attorney conduct rules, or any state bar opinion, and unlike some vendors in this lane there is not even a general statement that the user's professional obligations survive use of the tool. Searched the full site navigation on 1 September 2026, including the home page, the nine module entries, the FAQs, the master agreement, the privacy and security pages and the about entries. This is a gap with weight for a platform that drafts patent applications and is sold to patent attorneys and law firms across five countries.

Savings claims only, denominated in time to first draft rather than as a percentage. The headline figure is a complete patent draft in fifteen minutes, against a manual baseline the vendor describes as weeks, and prior art results returning in near real time against days of manual database work. Those are concrete and testable framings and they carry no methodology, sample or definition of what a complete draft comprises. Nothing appears on the client's side of the equation: no position on how AI assisted drafting time should be recorded on an invoice, and no exportable record showing what portion of an application or office action response was machine generated, which matters for a product sold to firms billing prosecution hourly. Checked the home page, the product pages, the blog library and the FAQ on 29 Aug 2026.

Time and efficiency savings are the central marketing claim and nothing addresses what happens to the invoice. Drafting LLM is said to cut patent drafting time by up to 50 per cent, ClaimChart LLM to achieve 75 per cent greater efficiency than traditional manual methods, and Ideacue to accelerate ideation and brainstorming tenfold. No guidance accompanies any of it on how a firm should bill for or disclose AI-assisted drafting, searching or charting to the client paying for the work, and no per-matter record of AI-assisted work is described for any module that a firm could interrogate. The audience for the savings claims is explicitly the buyer's own cost base, with the monetisation modules framed around revenue rather than fees.

On request, on the strength of a published Trust Center whose contents were not verified. A dedicated Trust Center and a privacy policy are both published in the site navigation, which gives a firm a defined destination to point a client to, and that is the distinction this value turns on rather than the strength of what sits behind it. Nothing was located outside it: no named certification, no subprocessor list, no auditor, no data processing agreement and no downloadable summary. Held at on request rather than not addressed because a published trust surface exists and was not entered, and held no higher because nothing behind it is established. Note the contrast within this category: PatSnap and Solve Intelligence reach this value through trust centres whose named contents were verified, and this record reaches it on the existence of the surface alone. Correction candidate in both directions.

The material exists behind the executed agreement rather than in public. Clause 6.2 of the master agreement discloses that customer inputs may be shared with trusted third-party suppliers who may collaborate on the services and may access whole or part of the data, and clause 6.4 states that such sharing happens under confidentiality obligations mutually agreed and signed between the customer and XLSCOUT, or under the terms of the content security policy. No supplier is named anywhere. The security page names AWS and Microsoft Azure as cloud providers followed by the open phrase and others, and Azure endpoints as where generative inference runs, but that identifies infrastructure rather than the model providers a client's AI clause asks about, which is the same line applied to Anaqua in this pull. A firm can therefore obtain the disclosure by contracting, and cannot forward anything to its client today.

Partial record, with an unusually strong internal consistency mechanism and the familiar two gaps. The source limb is answered on the search side: similarity scores, novelty summaries, highlighted passages and side by side comparison mean a practitioner can show what a novelty position rests on and where in a reference it appears, and exports carry to Word or Excel. The drafting side adds something no other record in this category describes, being automatic propagation of figure references so that renaming a figure updates every reference across claims, specification and legends, which addresses a real and common source of procedural rejection and is a traceability feature in substance. The gaps are the standard two: nothing indicates that output records which model produced it, and no human verification record is captured, so a practitioner who reviewed and adopted a generated claim set or traversal argument cannot evidence that they did. The relevant forum here is a patent office file wrapper rather than a court.

Nothing located addresses producing an AI-use disclosure or a verification certification. No export tying a passage to the model that produced it, the references it drew on, or the person who reviewed it is described for any module, and no AI-generated content marking or labelling convention is published. The automated novelty and invalidation reports do record which references were surfaced against which claim features, which is a retrieval record of a kind, but nothing states that it identifies the model or captures human review. Searched the module entries, the FAQs, the security pages and the master agreement on 1 September 2026. The obligation fits prosecution work imperfectly, though Drafting LLM output can reach a patent office and ClaimChart output can reach a dispute, so the row is recorded rather than skipped.