P
Patentcloud
Patentcloud is a patent analytics and search platform for IP professionals, patent counsel and the attorneys, brokers and analysts who work on patent portfolios, built by InQuartik Corporation of Taiwan. It is sold as six applications that share one patent database: Patent Search, Design Search, Patent Vault, Due Diligence, Quality Insights and SEP OmniLytics. Quality Insights produces one-click validity and quality assessments of a US patent, pulling file wrappers, supporting claim construction and generating prior art comparisons; its Semantic Prior Art function uses the vendor's own machine learning to predict relationships between a patent and potential prior art, returning up to 300 US, European, Japanese, Chinese, Korean and WIPO references ranked by semantic similarity, with references above the average similarity threshold marked so a reviewer can see which are most likely to matter.
A generative AI Patent Summary feature, introduced first in Due Diligence and later extended to Quality Insights and Patent Vault, condenses each reference and highlights the terms that align with the claims of the patent at issue, which the vendor presents as a way to categorise and narrow large reference sets rather than validating each document individually. Design Search matches design patents from an uploaded image.
Patent Vault is the portfolio management and collaboration workspace, holding up to 100,000 patents with customisable folders and up to twenty collaborators per account, with edit permissions controlling who can save findings into a shared project. Products are bought individually or in five bundled plans built around common patent scenarios, including due diligence, portfolio analysis, litigation and freedom to operate, on monthly or annual billing by credit card with free trials available.
InQuartik is ISO 27001 certified, with the certification scope covering all six Patentcloud applications and the company's internal systems, and hosts patent data on a hybrid infrastructure of Amazon Web Services data centres in Oregon and Tokyo and Taiwan Mobile data centres in Taipei. The terms of use are published and governed by the law of the Republic of China, with disputes heard in the Taipei District Court.
Capability grades
All 15 axes, graded from public sources on the date shown. Hover a grade to see what the letter means on that axis.
AI Centrality
How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.
The models are the engine of core capabilities layered on a product that would still function without them. What would survive their removal is substantial: a curated global patent database, Patent Search as a Boolean and bibliographic research tool, and Patent Vault as a portfolio management and collaboration workspace holding up to 100,000 patents with folders, projects and up to twenty collaborators. Those are database and workflow products in their own right.
What the models power on top is real and named rather than decorative: Semantic Prior Art, described as running on the vendor's own machine learning to predict relationships between a patent and potential prior art and return ranked references across six jurisdictions; a generative AI Patent Summary that condenses each reference and highlights terms aligning with the claims of the patent at issue, shipped into Due Diligence in June 2023, Quality Insights in October 2023 and Patent Vault in February 2024; Design Search, which matches design patents from an uploaded image; and auto-generated prior art references with a calculated claim disclosure rate in Quality Insights.
A is unavailable because the platform is sold as six separately purchasable applications and several of them are not model-dependent, so a buyer can hold a working product with the models subtracted. The vendor's own framing supports the reading, positioning the offering as high-quality patent data first and analytics over it second. Checked 8 September 2026.
Citation Accuracy and Hallucination Disclosure
Whether the vendor publishes measured accuracy on citations and assertions, grounds output to primary sources, and says plainly what its system does when it does not know. Legal has a documented public record of fabricated citations reaching filed briefs, so an untested claim of accuracy is not evidence.
Grounding is real, documented and openable, with a defined relevance threshold, short of the A band on the limb that asks for failure modes to be named. Every output the buyer acts on is a patent document the reader can open and check: Semantic Prior Art returns up to 300 US, European, Japanese, Chinese, Korean and WIPO references, Quality Insights surfaces file wrappers and supports claim construction, and Prior Art Analytics compares references against the claims of the patent at issue and calculates the proportion of claim elements previously disclosed.
The retrieval method is described in part rather than merely asserted, and the threshold is unusually specific: references are ranked by relevance and starred where semantic similarity exceeds the overall average similarity between all patents and their family members and section 102 prior art, which is a stated and non-arbitrary line rather than a vendor adjective. Figures exist and were weighed. A model-upgrade post publishes a before-and-after classification measurement on named portfolios, reporting a pairing success rate for novelty prior art rising from 77 to 86 per cent with the error rate reduced by 40 per cent, and marketing claims prior art search time cut by over 50 per cent.
Those were not counted toward A because they measure an internal classification improvement on two third-party portfolios rather than the accuracy of the product's output, and because the A band also requires the failure modes to be named and nothing on the estate names any. What the agreement says instead runs the other way and is recorded here: clause 11 disclaims liability for errors or omissions and states that the client assumes that risk. Checked 8 September 2026.
Autonomy and Oversight Model
What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.
Oversight is asserted, and the assertion takes the form of assigning the risk to the customer rather than describing a control. Clause 11 of the Terms of Use states that no legal advice is intended or offered, that the vendor disclaims any and all liability related to any decision taken in reliance on the content, and that the client acknowledges it assumes the risk of possible errors or omissions. That tells a buyer where responsibility sits and nothing about how the system is supervised.
No published material describes what the system does unaided, what constrains it, what review surface a professional works through, or what route an output takes back to human judgement, and no mode or setting is documented. One real control exists and is graded here rather than ignored: the starred relevance threshold on Semantic Prior Art tells a reviewer which references exceed average semantic similarity, which is a disclosed signal a professional can act on.
It is a single indicator rather than a control structure. The product class softens the question without removing it, and that is said plainly: this platform retrieves, ranks and summarises patent documents for a professional to assess rather than drafting or acting on a matter, so the D band's premise of a system that drafts, advises or acts does not describe it and D would overstate the gap. B is unavailable because no written commitment describes the models working alongside a supervising professional with real review surfaces. Checked 8 September 2026.
Operational and Outcome Evidence
Named, dated evidence that the product works in production at real firms or legal departments. Case studies with figures and identified customers count. Unattributed testimonials and launch announcements do not.
Results are quoted with the analysis subject named but the customer never is, which is the C band's second limb. The published figures are concrete and traceable to worked examples: a model-upgrade post reports the pairing success rate for novelty prior art on an Nvidia portfolio improving from 77 to 86 per cent with a 40 per cent reduction in the error rate, and works through a second example on a Masimo portfolio, while marketing claims prior art search time cut by over 50 per cent.
Those examples name the portfolios analysed, not customers who deployed the product, so they evidence the tool working on public patent data rather than production use by a buyer. The About page adds a scale claim of more than 440,000 patents handled over 35 years, which is leadership experience rather than deployment evidence. A retrieval limit belongs on this row and is not graded against the vendor: the product and customer pages at the vendor's own domain are robots-refused to the fetcher on both www and non-www forms, and no first-party named customer, case study or attributed testimonial was recovered through the search index across several queries.
A software-directory listing names two large manufacturers as customers; that is an aggregator source, excluded outright as evidence of anything, and is recorded only so a later reader knows it was seen and refused rather than missed. If a customer page exists behind the refusal, this row is the one to revisit. Checked 8 September 2026.
Privilege and Confidentiality Posture
How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.
Confidentiality is asserted in general terms, and the one confidentiality article in the agreement does not cover the customer's work. Section 9 of the Terms of Use binds both parties to hold confidential the contents, terms and all information related to the Agreement, for ten years after termination. That protects the deal, not the client's matter material. Nothing in the Terms commits InQuartik to hold in confidence the searches a client runs, the portfolios it assembles, the notes it takes or the reports it generates.
What is published elsewhere is a security posture rather than a confidentiality commitment: ISO 27001 certification with a named scope, and a statement of commitment to protecting information disclosure, both of which are graded on Security Certifications where they answer the question directly. Two clauses run the other way and belong on this row for a buyer whose search strategy is itself privileged work product. Clause 3.2 records that the vendor collects information on the client's searches and browsing history, the notes the client made and other platform activity; clause 3.11 adds the terms, phrases and choices made while searching; and clause 3.4 reserves the right to analyse and compile statistics from information collected in order to improve existing services and develop new services.
Privilege and work product are addressed nowhere, which under R33 forecloses A independently. No position is stated on what any model provider may retain, none being named. Checked 8 September 2026.
UPL and Professional Responsibility Posture
Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point.
A real position on advice versus tooling is published, and it sits in the customer agreement rather than in a website footer, which is what separates this from the C band's boilerplate-in-the-terms pattern. Clause 11 of the Terms of Use states in terms that no legal advice is intended or offered by InQuartik in making any of the content available, and that the vendor disclaims all liability related to any decision taken by any party in reliance upon the content.
The surrounding sentence adds that the client assumes the risk of possible errors or omissions, so the allocation is explicit rather than implied. The audience is not left ambiguous, which removes the other route to C: the vendor addresses IP professionals, attorneys, patent counsel, brokers and analysts by name, and clause 5.5 contemplates institutional subscriptions for law firms, corporations and government entities, so this is a professional tool sold to professionals rather than a consumer-facing product.
Two A limbs fail. Nothing addresses how the product supports a practitioner's competence or supervision duties, and no jurisdiction limit is named anywhere despite a database spanning six patent jurisdictions and a product whose validity analysis is US-specific. Checked 8 September 2026.
AI Governance and Bias Disclosure
Published governance over model behaviour: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
No governance position is published for a system whose output informs patent validity and freedom-to-operate decisions. Nothing identifies who inside the vendor is accountable for the models, no pre-release testing regime is described as governance, no policy or principles statement exists, no certification such as ISO 42001 is claimed, and nothing addresses uneven output across jurisdictions, technology fields, filing languages or portfolio sizes, which is where bias would actually bite on a patent analytics product.
The C band was tested and does not fit, because C describes responsible AI principles published without a mechanism and no principles are published at all. One item was considered and rejected under the no-double-spend rule. A model-upgrade post does publish a before-and-after evaluation of classification performance on two named portfolios, which is genuine measurement disclosure; it is spent on Citation Accuracy, where it answers the question the band asks, and it is not a governance framework, an accountability structure or a testing regime that a buyer could audit.
The ISO 27001 certification covers information security management and is spent on Security Certifications. The grade records what is locatable on this axis as of the date rather than asserting the vendor has no internal practice. Checked 8 September 2026.
AI Safety and Data Stewardship
Retention, deletion, access control, and what happens to prompts and documents after they are processed. Whether the vendor states its subprocessors and its incident practice, or leaves the buyer to assume.
A privacy policy and a set of personal-data rights cover the product without addressing what happens to the work itself after processing. What is published and real: clause 3.7 gives the client the right to request export, update, correction or deletion of personal information and to require the vendor to stop collecting, processing and using it, with a commitment to comply in a timely manner; clause 3.8 commits to systematic deletion of identity documents once identity is confirmed; clause 3.5 addresses onward transfer to third-party processors with a requirement that they comply with data protection regulation; and clause 3.6 acknowledges cross-border transfer with security safeguards.
Access control and infrastructure protection sit inside the ISO 27001 scope. What is absent is the substance this axis asks about for a product that ingests professional work. No retention period is stated anywhere for searches, notes, projects or generated reports, and clause 3.1 states instead that the vendor may at all times receive and store such information. Nothing describes what happens to a client's Patent Vault projects on termination; clause 9.4 addresses return or destruction of confidential information, which section 9 defines as the agreement's own terms.
No subprocessor is named, clause 3.5 referring only to trusted third-party processors and service providers. No incident or breach notification practice was located on any surface. B is unavailable because two of the elements it names, a subprocessor list and a stated incident practice, are both missing rather than one. Checked 8 September 2026.
AI Liability and Recourse
What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.
Liability is addressed only through disclaimer, and the indemnity runs one way. Clause 11 disclaims liability for errors or omissions throughout the services, states that the client assumes that risk as a subscriber, and disclaims all liability related to any decision taken by any party in reliance upon the content; it further disclaims liability for interruptions and third-party infrastructure failures. Clause 15 is the only indemnification article and it obliges the client to indemnify, defend and hold harmless InQuartik across five heads including breach, third-party IP claims arising from client submissions, and the client's activities in connection with the database and reports.
There is no vendor indemnity of any kind. The warranties at clause 10.1 are corporate rather than performance warranties: that the services contain no libelous matter, that the vendor operates lawfully and holds necessary permits, and that it has capacity to contract. Nothing warrants the accuracy, completeness or fitness of the analysis. No insurance is referenced. One retrieval limit belongs on this row and is not graded against the vendor: section 12 is headed Limitation of Liability, Cap and Force Majeure and the page runs from that heading directly to clause 12.4, a force-majeure notice provision, so the limitation and cap text at 12.1 to 12.3 did not render and was not read.
The grade rests on what is published rather than on the missing clauses. D was tested and does not fit, because the allocation of loss is stated rather than absent; it is simply stated against the buyer. Checked 8 September 2026.
Practice Systems Integration Depth
How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.
No integration into practice systems was located, and the vendor's own positioning is that none is needed. The bundle FAQ states that the plans allow a team to work on the Patentcloud platform without needing to introduce additional third-party tools, which is a claim of self-sufficiency rather than of interoperability. No docketing system, document management system, IP management platform or patent office filing interface is named anywhere as a connection, no API is documented or referenced, and no integrations page was recovered.
What does exist is export and sharing inside the platform's own boundary: PatentMatrix dashboards and Due Diligence and Quality Insights reports can be shared with colleagues and partners, and the KEEP function saves findings into a Patent Vault project for team members. Those move work between the vendor's own applications and between its own users, not into the systems a firm already runs. A retrieval limit is recorded and did not drive the grade: the product pages are robots-refused to the fetcher, so an integrations page could exist behind the refusal.
The grade rests on located material rather than on the refusal, because the vendor's published account of its bundles positively asserts that third-party tools are unnecessary, which is affirmative evidence rather than silence. Checked 8 September 2026.
Deployment Model and Data Residency
Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.
The deployment model is stated clearly with partial residency detail, which is the B band. Published: the platform runs on a hybrid cloud infrastructure, and the locations are named rather than left as a cloud gesture, comprising Amazon Web Services data centres in Oregon and Tokyo together with Taiwan Mobile data centres in Taipei. The vendor states the AWS facilities are themselves certified to ISO 27001, PCI DSS Service Provider Level 1 and SOC 2, and that the Taipei facilities carry Tier III certification in design, facility and operations.
Physical and environmental security sits inside the ISO 27001 scope. Three named locations across two continents is more residency detail than most records in this lane publish. Three limbs are missing and each matters to a buyer with client obligations. No tenancy model is stated anywhere, so whether the deployment is single or multi-tenant and how customer separation is achieved cannot be established. Processing location is not distinguished from storage location, so a buyer cannot tell which of the three sites analysis runs in.
And no region choice is offered or described, the geography being disclosed as the vendor's architecture rather than as a customer election, with clause 3.6 confirming that information collected in one jurisdiction may be transferred to servers in another under unspecified safeguards. Checked 8 September 2026.
Security Certifications and Trust Center
Independent attestation a buyer can pull without a sales call: SOC 2, ISO 27001, penetration test summaries, a trust center with current reports and named scope rather than a badge image.
Independent attestation with a named scope, reachable without a sales call, on a dedicated trust surface. The vendor publishes a Trust and Security page stating that InQuartik is ISO 27001 certified and setting out the scope of the certification with unusual specificity: all six Patentcloud applications enumerated individually, Quality Insights, Due Diligence, SEP OmniLytics, Patent Search, Design Search and Patent Vault, together with internal systems spanning development, maintenance, physical security, and the management of application systems, data protection systems and network communications.
The certificate itself is offered for direct download rather than through an account executive or an NDA gate, which under R5 is the self-serve tier and reaches the top band. The page also names the hosting estate and the certifications held by those facilities. Two limits are named because they are real and because the grade should be read with them. The certificate was not opened: the trust page path is robots-refused to the fetcher on both URL forms and its body was recovered through the search index, so the auditor and the certificate's validity dates are not established, and the most recent certification announcement located is a second-year renewal rather than a current-year confirmation.
Grading down on that basis would convert a fetcher constraint into a finding against the vendor, which rule 6.6 forbids, so the grade records what the vendor's own text establishes and this note carries what it does not. No SOC 2 report of the vendor's own is claimed; the SOC 2 reference belongs to AWS and is infrastructure. Checked 8 September 2026.
Model Supply Chain Disclosure
Which models sit underneath, whose they are, where they run, and whether the vendor commits to telling customers when that changes. A legal buyer inherits every dependency it cannot see.
The vendor refers to its models as proprietary and identifies nothing underneath them, which is the C band exactly. Semantic Prior Art is described as powered by Patentcloud's machine learning technology, which is a claim of ownership rather than an identification; the Patent Summary feature is described as generative AI without naming a model, a provider or an architecture; and a model upgrade is discussed in terms of classification performance without saying what was upgraded.
No foundation model is named anywhere, no model provider is named, and nothing states whether the generative component is built in-house or supplied by a third party, which is the single question a buyer would ask of a feature that sends patent text to a model. The Terms of Use are silent on the point across their whole length. Amazon Web Services and Taiwan Mobile are named, and both are spent on infrastructure: under R29 test 1 naming a cloud host says where a model runs rather than whose model it is, and the same fact cannot also answer this axis.
D was tested and does not fit, because the vendor does disclose that machine learning and generative AI sit behind named features and characterises the machine learning as its own; what it withholds is everything below that. Change notification is not reached, nothing being named that a change could be notified about. Checked 8 September 2026.
Commercial Transparency
Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.
The unit and the structure are published in real depth and the figures did not resolve, which is the B band's second limb. Published: five named bundle plans, Premium, Patent Due Diligence, Patent Portfolio Analysis, Patent Litigation and Freedom to Operate, each built around a stated patent scenario and each with a feature comparison table setting out precisely what it includes and excludes; separate standalone pricing pages for the individual products; the unit of charge stated as one account per plan with additional seats available on request; quantified entitlements including a 100,000-patent storage limit, up to 1,000 customisable folders and up to twenty collaborators; a stated bundle discount of up to 79 per cent against standalone subscription; monthly and annual billing cycles with automatic credit card billing at the start of each cycle; accepted payment methods; a published switch-and-cancel policy operating from the next billing cycle; and free trials including seven-day trials and two free validity analysis reports.
Custom bundles and additional seats are withheld to a sales conversation, which is the enterprise-tier withholding the band contemplates. Two things are named rather than glossed. The plan figures render client-side and returned empty cells on fetch, so no rate was established first-party, and that is a retrieval limit under R85 rather than a finding that no figure is published. And a software-directory listing publishes four annual figures; it is an aggregator, excluded as evidence, nothing here is graded on it, and it is named only so a later reader knows it was seen and refused. Checked 8 September 2026.
Firm and Practice Coverage
Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.
Segment and practice coverage is described with real substance, with the boundaries left open. The buyer segments are named individually rather than gestured at: IP professionals, attorneys, dealers, brokers, scientists, engineers, financiers and designers, and clause 5.5 of the Terms contemplates institutional subscriptions for law firms, corporations and business, charitable or government entities. Practice coverage is unusually concrete because the five bundle plans are themselves named practice scenarios, each mapped to the work it supports: due diligence before a buying or licensing decision, portfolio analysis at any scale, litigation response where the task is finding weaknesses in the patent at issue, freedom to operate where the task is clearing products in target markets, and an all-scenario enterprise plan.
Standard-essential patent work is served by a dedicated application, with 5G and 6G claim charting and FRAND issues named specifically, and design patent clearance by another. Jurisdictional coverage is stated where it is testable, with US, European, Japanese, Chinese, Korean and WIPO documents named as the semantic prior art scope. A is unavailable because the limits are not stated: nothing says which jurisdictions are outside coverage, and the validity and file wrapper analysis in Quality Insights is described against US patents without a statement of what is unavailable elsewhere. Checked 8 September 2026.
4 public documents
The public pages on file for Patentcloud, with the recorded signals each one supports and the date it was last read. Open any of them and check the reading against the record.
-
inquartik.com/terms-of-use7 signals
Client Data in Training, Prompt and Output Retention, Ethical Walls and Matter Segregation and 4 more
Read Sep 8, 2026
-
inquartik.com2 signals
Fabricated Citation Record, Court Disclosure Support
Read Sep 8, 2026
-
inquartik.com/patentcloud2 signals
Good Law Verification, Billing and Fee Posture
Read Sep 8, 2026
-
Refusal and Uncertainty Behaviour
Read Sep 8, 2026
No published figure
- Patentcloud is sold as six products you can buy on their own or in one of five bundles built around common patent jobs: due diligence, portfolio analysis, litigation, freedom to operate, and an all-scenario Premium plan. Bundling is where the discount is, up to 79 per cent against buying the same products separately. Each plan is one account, billed monthly or annually to a credit card, and every bundle includes Patent Vault with a 100,000-patent store, up to 1,000 folders and up to 20 collaborators. Free trials are available, including two free validity reports on Quality Insights. Extra seats, more storage or a different product combination are quoted individually.
Published plan structure across five named bundles with the unit of charge stated, and the figures withheld from the fetch rather than from the page. Plans: Premium, described as the all-scenario plan for enterprise-level teams; Patent Due Diligence; Patent Portfolio Analysis; Patent Litigation; and Freedom to Operate. Each carries a feature comparison table setting out what it includes, covering patent bibliography data, the full Patent Search function set, the Patent Vault Premium function set, unlimited Due Diligence reports, unlimited Quality Insights reports, and image-based design patent search.
Standalone pricing pages are published per product for Patent Search, Design Search, Patent Vault, Due Diligence and Quality Insights. Unit of charge is stated as one account per plan, with additional seats available by contacting the vendor. Quantified entitlements are published rather than described: a 100,000-patent storage limit, up to 1,000 customisable folders and up to 20 collaborators per account. Billing terms are published: monthly plans billed automatically to a credit card at the start of each monthly cycle, annual plans once per year at the start of the cycle, all major credit and debit cards accepted, alternative payment methods on request.
The bundle discount is quantified at up to 79 per cent against standalone subscription. A published switch-and-cancel policy allows a change or cancellation by contacting the Client Success team, with the new plan charged from the start of the next billing cycle. Free trials are offered, including seven-day trials and two free validity analysis reports on Quality Insights. The agreement adds the commercial mechanics: payment within thirty days of invoice, all charges non-refundable once a subscription is activated, payment in US or Taiwan dollars, automatic renewal for successive twelve-month periods with ninety days' notice required to prevent renewal, no termination for convenience during the initial twelve-month term, and liquidated damages of up to one thousandth of the amount due for each overdue day.
Two limits are named. The plan figures render client-side and the comparison table returned empty cells on fetch, so no rate was established first-party; that is a retrieval limit and not a finding that no figure is published. And a software-directory listing publishes four annual figures for the bundles; that is an aggregator source, excluded as evidence, nothing here rests on it, and it is named only so a later reader knows it was seen and refused rather than missed.
Surfaces read on 8 September 2026: the bundle pricing page and its FAQ, per-product pricing pages via the search index, the get-started page, and the Terms of Use in full.
Legal Signals
What each signal meansA signal records what public sources say on the date shown. It is not a grade and it is not a recommendation. Where a signal reads Not addressed, it means the index did not locate the material in public sources on that date, which is a statement about disclosure rather than about the product.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
A published agreement or policy exists and none of it addresses the question either way, or the document that would answer it could not be read and the summary names the retrieval limit. The summary states which shape the silence takes: an improvement right granted that never names training, or no improvement right granted at all.
The agreement is published and was read in full, and no provision addresses training, model improvement or machine learning. The Terms of Use, last updated 1 May 2022, run to twenty-one articles and none of them mentions the models the product ships, despite the vendor describing Semantic Prior Art as running on its own machine learning and Patent Summary as generative AI. no-agreement-published is false because the agreement is published and complete.
The clause quoted here is the closest located and is recorded in full rather than credited, because under the rule that a clause must name the thing it does not reach this signal: it permits analysis and statistical compilation to improve and develop services and names neither training nor machine learning. It is nonetheless recorded prominently because its subject matter is broader than the usual usage-statistics provision.
Clause 3.2 states that the vendor collects information on the client's searches and browsing history, the notes the client made and other platform activity, and clause 3.11 adds the terms, phrases and choices made while searching, so what may be analysed under the quoted clause includes a patent professional's search strategy and working notes rather than telemetry alone. Clause 3.4 also permits sharing analysis or statistics publicly or with third parties in aggregated and non-personally-identifiable form. policy-never was tested and declined: no surface states that client material is not used to train.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
Retention is acknowledged in public materials with no stated period.
Retention is disclosed without a period anywhere. The agreement is candid that material is stored and open-ended about how long: the clause quoted here states that the vendor may at all times receive and store usage information, and clauses 3.2 and 3.11 describe what is stored as including IP address, software and hardware attributes, searches and browsing history, the notes the client made, and the terms, phrases and choices made while searching, configuring alerts or bookmarking.
No retention schedule, deletion cadence or maximum period appears in the Terms of Use or the privacy policy. disclosed-fixed was declined because no window is stated. customer-configurable was declined because nothing gives the client a retention setting to operate. What the client does have is a request-based route over personal information rather than over the work: clause 3.7 allows the client to require export, update, correction or deletion of personally-identifiable information and to require the vendor to stop collecting, processing and using it, with a commitment to comply in a timely manner, though the vendor may charge a reasonable cost for review or duplication.
Clause 3.8 commits to systematic deletion of any identity document once identity is confirmed, which is the one deletion commitment on the estate with a defined trigger. Nothing addresses what becomes of Patent Vault projects, saved reports or search history on termination.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
The product maintains its own permission model, documented, requiring the firm to keep it aligned.
The product maintains its own documented permission model, and it is a collaboration model rather than an ethical wall. What is published: Patent Vault organises work into projects with customisable folders and up to twenty collaborators per account; project members must be authorised to edit before they can save findings into a shared project through the KEEP function; reports and dashboards are shared deliberately with named colleagues rather than being open by default; and the clause quoted here restricts an institutional subscription to individuals designated as subscribers at the time of purchase, so access is named-user rather than open to the firm.
That is a real permission structure the firm configures and must keep aligned, which is what this value records. What it is not is a segregation mechanism. Nothing addresses walls between matters or between a firm's clients inside a single account, nothing describes what prevents a colleague on one engagement seeing work saved from another, and no conflicts or screening capability is published. That limit is stated here rather than left to inference. inherits-dms-acl is not reached because the product holds its own repository and retrieves over a patent database rather than over a customer document system, so there is no source access model to enforce.
The structural risk this signal targets is also lower here than for a document AI, since retrieval runs over public patent records rather than across client files.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
Published terms or policy address disclosure to authorities or in response to legal process, and no commitment or reservation regarding customer notice is located anywhere. The vendor has told the customer that data can leave and has said nothing about whether the customer hears of it.
Compelled disclosure is addressed and notice is never reached, which is this value precisely. Clause 9.1.2 carves compelled disclosure out of the confidentiality obligation on the terms quoted here, and clause 9.1.3 adds a further exception for disclosure required in connection with litigation, arbitration or administrative or other investigations and proceedings. Both permit disclosure and neither commits the disclosing party to tell the other in advance, to give it an opportunity to seek a protective order, or to notify it afterwards.
The limitation to the extent of the legal obligation is a real constraint on scope and is recorded as such, but it is not notice. Clause 14.6 points the same way from a different direction, stating that InQuartik may cooperate with legal authorities and third parties in the investigation of any suspected or alleged crime or civil wrongdoing, again with no notice commitment attached. discretionary was tested and declined: it asserts that the terms reserve discretion over whether to notify, and these terms do not reach the question of notice at all, so its words would be false.
One scope limit compounds the position for this buyer and is recorded: section 9 defines the confidential information it protects as the contents, terms and information related to the Agreement, so the client's searches, projects and notes are not obviously inside the article that contains these exceptions in the first place.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
Coverage is described by jurisdiction with no identification of the underlying corpus.
Coverage is described by jurisdiction and the underlying corpus is never identified. Jurisdictional scope is stated where it is testable: Semantic Prior Art returns US, European, Japanese, Chinese, Korean and WIPO documents, Quality Insights validity analysis is described against US patents, and the platform is presented as holding global patent data. What is absent is any identification of where that data comes from or on what footing it is held.
No data supplier, feed, national office agreement or open-data source is named; no update cadence or currency date is published; and no licence or rights basis is stated for any part of the database. The nearest the estate comes is clause 2.5, which excludes from the vendor's service commitments any restrictions imposed by the database source such as a jurisdiction's patent office. That acknowledges the category of source and identifies no source.
Clause 5.7 runs in the opposite direction, asserting that all right, title and interest in the Database and the Content are and shall continue to be the exclusive property of InQuartik and prohibiting the client from it or building a searchable database from it, which asserts the vendor's rights downstream without addressing the basis on which the underlying records were obtained upstream. named-no-licence was declined because it requires the sources to be identified, and a category is not an identification.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
The vendor computes and surfaces subsequent history itself, with the method described.
The vendor computes and surfaces the equivalent of subsequent history itself, with the method described in part. This is the one product class in the index where this signal's question has a genuine analogue: for a patent rather than a case, the question of whether the authority you are relying on still stands is a question of validity and legal status, and this platform is built to answer it. Quality Insights is described as a one-click solution providing consolidated data to evaluate the validity and the quality of a US patent, surfacing file wrappers, supporting claim construction and generating prior art comparisons.
Its outputs are treatment signals in substance: auto-generated prior art references that bear on validity, a Prior Art Analytics function that compares references against the claims of the patent at issue and calculates the proportion of claim elements previously disclosed in the file wrapper, and a semantic similarity ranking whose starring threshold is defined against the average similarity between all patents and their family members and section 102 prior art. Method is described in part rather than fully, which is what holds this at the neutral value. licensed-citator was declined because no third-party citator or commercial treatment service is named; the computation is the vendor's own. verification-prompt-only was declined because the product performs the analysis rather than instructing the reader to go and check elsewhere.
Refusal and Uncertainty Behaviour
What does the product do when the answer is not in the corpus?
The product exposes a confidence or grounding score without an explicit abstention path.
A confidence signal is exposed and no abstention path is described, which is this value. The confidence signal is better specified than most records that carry it. Semantic Prior Art ranks every reference it returns by relevance between the patent at issue and the abstract and first claims of the reference, and marks with a star those references whose degree of semantic similarity exceeds the overall average similarity between all patents and their family members and section 102 prior art. That is a stated, non-arbitrary threshold a reviewer can reason about rather than a vendor adjective, and the user can narrow the search scope by adding selected claim text or keywords.
What is missing is the other half. Nothing published describes what the product does when it cannot ground a result: no statement covers what a user sees when no reference exceeds the threshold, whether an empty or low-confidence result is distinguished from a failed search, or what happens where coverage in a jurisdiction is thin. No abstention or no-answer path is documented, so documented and documented-and-demonstrable are both unavailable.
The agreement addresses the same territory only by allocating risk, clause 11 stating that the client assumes the risk of possible errors or omissions, which is a disclaimer rather than a description of behaviour.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
No court order, opinion or disciplinary record addressing fabricated or hallucinated legal citations produced by this product has been located as of the date shown. This is a statement about the public record on that one subject, not a finding about the product, and this signal is not a litigation history.
No matter naming this vendor or its product was located. Searches were run on both the product name, Patentcloud, and the company name, InQuartik, against the AI Hallucination Cases database maintained by Damien Charlotin and against general search on 8 September 2026, and nothing returned any filing, sanction, order or judicial finding involving either. Searching on both names follows the standing instruction that filings may name either the product or the company.
The value records the state of that search on the date rather than a claim that no such matter could exist. One structural observation is recorded for a later reader and is not a substitute for the search, which was run regardless. The product does not draft legal argument or generate citations to case law; it retrieves patent documents from a database and ranks and summarises them, and every reference it surfaces is a real patent record the user can open.
The failure mode this signal tracks is therefore unlikely to arise from its output in the ordinary course, though the generative Patent Summary feature does produce synthesised descriptions of those documents.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
Public materials refer to professional responsibility in general terms without naming guidance.
Professional responsibility is engaged in general terms and no guidance is named. The engagement is real and it sits in the agreement rather than in marketing: clause 11 states in the terms quoted here that no legal advice is intended or offered, and adds that the vendor disclaims all liability related to any decision taken by any party in reliance upon the content and that the client, as a subscriber, assumes the risk of possible errors or omissions.
For a product sold to patent attorneys whose validity analysis informs opinions, that is a deliberate statement about the boundary of the tool rather than incidental boilerplate. What is absent is any named authority. No bar ethics opinion is cited, no guidance from a patent bar or an office of enrolment and discipline is discussed, no regulator is named, and nothing maps the product against the professional conduct obligations of the practitioners it is sold to in any jurisdiction. named-guidance requires engagement with at least one named ethics opinion and nothing on the estate reaches that.
The signal measures engagement with AI-specific professional guidance in particular, and an advice-line disclaimer written into terms last updated in May 2022 predates the guidance a buyer would now want addressed.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
Public materials claim time savings without addressing billing or disclosure, and the product sits inside a fee relationship between a lawyer and a client where those savings would change the bill.
Savings are the frame and nothing addresses what happens to the bill when the work compresses. The claims are quantified and directed at cost: prior art search time cut by over 50 per cent, and the pitch that with the platform in place a buyer will never have to manually collect patent data or outsource a prior art search again, which is explicitly a claim about displacing outside spend. The bundle material adds savings of up to 79 per cent against standalone subscription, which is a pricing discount rather than a claim about matter economics but points the same way.
What the signal asks is whether the vendor addresses what happens to the bill when a task that took six hours takes one, and nothing does. No per-matter record of AI-assisted work is produced, no artefact exists that a firm could put in front of a client showing what the analytics contributed, and no guidance is published on fee or disclosure treatment where the tool has compressed chargeable time. audit-record and audit-record-and-guidance both require a per-matter record and neither is available.
The direction of sale is worth recording: the buyer here may be a corporate IP department, a law firm or a broker, and the outside-counsel displacement claim is addressed to the client-side buyer, so for a firm the question the signal asks arises squarely and is unaddressed.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
No located public material supports a client side disclosure obligation.
A firm could not get this vendor through a client's AI clause without a bespoke negotiation, because neither of the two artefacts the signal requires exists. No subprocessor list is published anywhere, and nothing indicates one exists on request: the clause quoted here refers only to trusted third-party processors and service providers as a category, with a commitment that they will protect the information and comply with data protection regulation, and no party is named.
And no statement identifies which model providers, if any, see client content; no foundation model or provider is named on any surface, while the vendor describes its machine learning as its own and its Patent Summary feature as generative AI. Two named third parties were tested and neither satisfies the signal. Amazon Web Services and Taiwan Mobile are named as data centre operators, and under the coverage test naming a host says where a model runs rather than whose model it is, so infrastructure alone never satisfies this signal.
One structural point compounds it and is recorded rather than inferred: clause 20.1 provides that both parties act as data controllers, so there is no processor relationship and consequently no data processing addendum, subprocessor annex or forwardable client-facing disclosure pack anywhere on the estate. on-request was tested and declined because nothing suggests the material exists behind a request or an executed agreement.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
No located public material addresses court disclosure or verification certification.
No located public material addresses court disclosure or verification certification. Nothing on the estate discusses judicial standing orders on the use of artificial intelligence, no disclosure template or guidance is published, and no export produces a per-document record of the model used, the sources drawn on and the human verification applied. The question has more purchase here than on a purely transactional product, and that is recorded rather than glossed: the platform is expressly marketed for patent litigation, with a dedicated bundle built around finding weaknesses in a patent at issue, and its claim charts and validity reports are prepared for use in disputes where an opponent or a tribunal may ask how the analysis was produced. partial-record was tested and declined.
Reports are exportable and shareable and they do list the prior art references relied on, which is a record of sources in a substantive sense, but nothing identifies the model used, distinguishes generative Patent Summary output from retrieved document text, or records what a human verified. Nothing published describes any of this as serving a disclosure purpose. The vendor's position on reliance runs the other way, clause 11 placing the risk of errors or omissions on the client without offering anything the client could file.