Huski.ai vs MarqVision: how they compare in 2026
Huski.ai and MarqVision both apply image recognition to protecting a brand online, and they differ sharply in what they publish. MarqVision publishes measurement and terms. It reports 97 percent on enforceable item detection using SKU level comparison against the brand's own product data, 99.8 percent across more than 48,000 domain impersonation incidents and 94 percent SKU matching on unauthorized sales, and states openly that these are internal benchmarks. Its agreement sets caps, an infringement indemnity running to the customer, and a clause stating that the services include no legal representation. Huski.ai publishes its price openly: 299 dollars a month for unlimited text and image clearance searches, no contract, no card to start. Beyond that it publishes little, with no certification, no training or retention position and no model provider.
At a glance
All 15 axes, side by side
The same grid applied to every vendor in the index, graded from public sources. 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 image recognition model is the product and the founding insight, not a layer over an existing service. The company was built in 2020 around applying visual similarity matching to trademark work, and the vendor says the technology removes the need for design search codes. The USPTO design coding system exists because computers could not compare images, so a model that compares them directly replaces the taxonomy rather than speeding a search within it. The same recognition engine runs across all four products: matching filed applications in watch, clearing candidate designs in search, and identifying infringing product listings online, including images the vendor says are blurred or obscured. Remove the model and there is no search, no watch and no brand protection, only a docketing tool.
The models are the product, with nothing conventional beneath them. MarqVision sells detection and enforcement at a scale no human team can reach. It scans more than 1,500 channels across 118 countries, compares findings with a brand's own product data at SKU level, analyzes video frame by frame, and generates the notices that follow. The named capabilities are all machine learning. They include Atomic Product Detections, described as using generative AI for SKU level precision, and a multi agent system that decides whether an infringement is valid and classifies the enforcement grounds. There is also an LLM framework that generates search keywords in any language, image recognition and text analysis, and models the company says are trained on successful enforcement data. There is no content asset underneath: the raw material is marketplace listings the crawler gathers, which only the models make usable. The stated alternative is not other software but the manual monitoring the product replaces.
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.
The mechanism is described and demonstrated, and nothing measured is published. Independent coverage by a named legal technology journalist describes running an image search against the USPTO database and getting visually similar results within seconds. It reproduces a worked example in which the tool found a filed design similar to an existing Louis Vuitton mark. A demonstrated example in third party coverage is better evidence than a claim, but it is not a measurement. No figure is published: no precision or recall on visual similarity, no false negative rate on clearance, no comparison against design search code retrieval, which is the method the vendor says it replaces, and no evaluation or test set. The gap matters because in clearance the failure is a silent miss, and a confusingly similar mark the model does not return produces an opinion that is wrong but looks complete.
Accuracy is measured, published and explained, and one clause in the agreement undercuts it. Three figures appear: 97 percent on enforceable item detection using SKU level comparison, 99.8 percent across more than 48,000 domain impersonation incidents, and 94 percent SKU matching accuracy on unauthorized sales. The company says openly that these are internal benchmarks. The method is described: SKU level comparison with genuine product data, detection on several signals, and a tiered model that routes borderline cases to analysts. A published piece names failure modes, setting out what a false positive costs, including misrepresentation liability under 17 U.S.C. 512(f), tortious interference and suspension of platform reporting. Only one figure carries a sample size and none carries a test set. Section 2.4 of the terms also says Marq Vision assumes no liability for the quality, accuracy or validity of the data gathered by the platform and that the customer shall in no event rely on it. The marketing measures accuracy and the agreement disclaims reliance on it.
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.
The professional is in the loop by design, and no oversight model is published. The product is sold only to trademark attorneys, paralegals and legal professionals, and the vendor says IP attorneys and their staff are the end users, so every output reaches a qualified reviewer. Output is framed as input to professional work: search results feed a report and a customized opinion letter the attorney issues, and watch results feed a decision on whether to file a Letter of Protest or an opposition. Nothing says what runs unattended. No confidence indication accompanies a visual similarity match, and no threshold governs what surfaces in a watch alert. Nothing describes what happens when the engine is unsure about a blurred or obscured listing image in brand protection, where an automated conclusion could support enforcement against a third party seller. Oversight built into the audience is real, but it is not a published model.
Real review surfaces are published, and the point at which the system acts alone is not. The company says a human reviewer confirms every significant decision and describes a tiered model that routes only borderline cases to analysts. Each classification comes with its reasoning and a risk score, so a team can see why a determination was made. Two customer controls are contractual. Section 8.3 of the terms requires qualified customer representatives to review, validate and verify all information transmitted with due care, and to report claims only with a good faith belief in their truth. The whitelist gives the customer a standing list of sellers it does not want enforced against. Against that sits the speed the product is sold on: 15 minutes from detection to enforcement, median responses under five hours for fake domains, responses in seconds for piracy, and enforcement actions started automatically. Both cannot be fully true at once. Significant and borderline are never defined, so nothing published says which determinations reach a person.
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.
Independent journalism and a demonstrated product, with no customer evidence at all. The strongest element is coverage by a named legal technology publication that reviewed the product at launch, described the founders and their reasoning, ran the search itself and published a worked example of a result. That independently confirms the core capability works as described. Company facts are checkable: founded in 2020, based in Middletown, Delaware, with named cofounders, and independent directory and review listings exist. Against that, no law firm or customer is named in located material, and no case study, usage figure, funding announcement or award was found. No outcome or efficiency claim is published with a figure beyond the pricing comparison.
The customer list is large and spans categories seldom seen together. Roughly forty logos run across automotive (Nissan, GM), electronics (Panasonic), consumer goods (Henkel, Persil, Miele), beauty and fashion (Lush, Wella, Stussy, Courreges, Lemaire, Ader Error), Korean commerce and finance (Coupang, Shinsegae, Toss, KT&G), and gaming and media (Smilegate, Class101). Six case studies are published and five people speak with name and title. They include Tae Hyun Kim, Director of Ecommerce at Miele, Ryan Dahlstrom, Global Director of Digital Commerce at Darn Tough Vermont, and John Belcaster, General Counsel of MSCHF. Operational figures are plentiful: 180 times faster enforcement for its own teams, 50 percent faster than traditional providers and a 15 minute mean time to resolution. It also reports more than ten million sellers tracked, dozens of offline raids in Shenzhen, Guangdong and Yiwu, and one million dollars of counterfeits seized. Method and dates are missing. No figure carries a measurement basis or date, and none is tied to a named customer.
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.
Nothing located. The product holds candidate marks and unlaunched brand designs uploaded for clearance, which are commercially sensitive before filing, and it generates opinion letters, which are the archetypal attorney work product in trademark practice and the document that becomes contested in later infringement litigation on the question of what the client knew before adoption. Nothing published addresses attorney client privilege, work product, the confidentiality of a clearance search or of a generated opinion letter, or the conflicts question that arises when one platform serves many firms clearing marks against each other's clients. The vendor's exclusive focus on legal professionals makes the omission more surprising than it would be for a general enterprise tool.
One part is strong and the rest are thin or point the wrong way. Section 5 of the terms is a proper mutual confidentiality regime with notice before compelled disclosure, and it survives termination. Section 4.2 confirms the customer owns all data it sends and grants no license beyond providing the services, and section 4.3 limits Marq Vision's license to the term and to service delivery. Against that, section 2.5 grants a broad right to use information the customer submits, expressly including information about its customers, distributors, inventory and products. The uses include improving the platform, developing updates, and improving the business models and services of Marq Vision, its subsidiaries, its vendors or its other customers. Anonymized data is excluded from confidential information at section 5(d) and owned by Marq Vision under section 4.1. Nothing addresses separation between customers, and neither document gives a retention or deletion position beyond general reasonableness. Privilege and work product are not mentioned. Section 1.2 disclaims any attorney client relationship, but that leaves the question open for a customer whose in house counsel routes enforcement through the platform.
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. Where the advice line is not the duty a product raises, the axis is read through the nearest professional duty it does raise: judicial conduct rules and the reviewing duty for products sold only to courts, and the duty to bill for time actually spent for products that draft time entries.
A clear scope statement exists, and no professional framework sits behind it. The vendor says plainly that Huski is for trademark attorneys, paralegals and other legal professionals, that IP attorneys and their staff are the end users, and that clients benefit indirectly. That limits the user population to qualified professionals, and since the tool never reaches the public, it is a real answer to the unauthorized practice question. Everything else is absent. No statement says that generated output, including the customized opinion letter, is not legal advice. Nothing addresses the attorney's duty of competence over an opinion letter generated by machine and issued under their name, and no bar guidance is engaged. Generating an opinion letter in a few clicks raises that question sharply, and it is not addressed.
The agreement draws the line clearly and the marketing crosses back over it. Section 1.2 of the terms, headed Authorization; No Legal Representation, says the services include no legal representation and that Marq Vision and its staff do not and cannot give legal guidance or advice. It says the services never create an attorney client relationship, that soft notices are not a substitute for legal claims, and that legal questions should go to the customer's own counsel. Where offline services involve proceedings, third party law firms or agencies bring them. That is a complete statement of what the product is and is not, backed by the customer warranties in section 8.3 requiring due care in review and a good faith belief before any claim is reported. Against it, the anticounterfeit page claims the AI driven solutions comply fully with the ABA's ethics guidelines, ensuring the highest standards of legal integrity in every action. The American Bar Association's rules govern lawyers, and the vendor's own agreement says it is not acting as one. No jurisdiction limits are stated.
AI Governance and Bias Disclosure
Published governance over model behavior: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
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, no drift statement, no named governance body, no ISO 42001 and no EU AI Act positioning are published. The untested risk is specific to image recognition and worth naming rather than listing generically: visual similarity models are known to perform unevenly across image types, and this engine is applied to logos, product photographs and deliberately obscured listing images alike, with results feeding both clearance opinions and enforcement actions against third party sellers who never see the model. Nothing published indicates that performance across those conditions has been examined. Checked the home page, the solutions pages, the FAQ, the published rankings material and the site navigation on 29 Aug 2026.
One transparency commitment is published and nothing stands behind it. The anticounterfeit page says a customer will always have full visibility into when and how AI is used to enforce and protect their brand, which is a governance principle rather than a feature. The impersonation material adds that every classification carries its reasoning and a risk score, which explains individual decisions. Nothing else is published. There is no responsible AI page, framework or set of principles, no person or function named as accountable for model behavior, nothing on what is tested before a model change ships, and no certification such as ISO 42001. Nothing addresses uneven output. That is a live question for a system working across 118 countries and 1,500 channels, where detection quality plausibly varies by language, script, marketplace and product category. Here a false positive is not an inconvenience but a takedown against a real business.
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.
No stewardship position located. Nothing states whether uploaded candidate marks, logo images, client portfolios or search histories are used to train or improve the recognition models, no retention period is published, and no deletion right is described. The gap has a distinctive edge on this product: the platform ingests unfiled logo designs at the clearance stage, which are both commercially sensitive and precisely the kind of labeled image data that improves a visual similarity model, so the incentive alignment is visible and the position is unstated. Docketing additionally holds entire client portfolios uploaded in one click, and the Customs and Border Protection integration means recordation status data flows through the platform.
A trust center exists and almost nothing beneath it is stated in a way a reviewer could test. Section 1.3 of the terms commits Marq Vision to using information security techniques consistent with good industry standards, and the privacy policy states that all appropriate technical, organizational and administrative security measures are used, both of which are reasonableness language rather than a described control set. Retention is acknowledged and never quantified, with the privacy policy saying data is kept no longer than reasonably necessary and the terms containing no deletion obligation for customer property on termination. No subprocessor is named anywhere: the privacy policy describes agents and affiliated businesses generically. No incident or breach notification practice is published on any surface. Access control is addressed only as the customer's responsibility to protect its own password. A trust center hosted on Vanta is linked from the footer and would ordinarily carry the control detail; it renders in the browser and returned no readable content, so it may say more.
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.
No published position located. Nothing was found on liability for AI output, warranty, service levels or remedy. Three distinct exposures arise from this product and none is addressed. A clearance search that misses a confusingly similar mark supports an opinion letter the attorney signs, and the opinion letter is the artifact a client relies on to launch a brand. A docketing system that fails to surface a deadline causes a lapsed registration, which is a classic malpractice claim and is not covered by any published service level. And a brand protection match that wrongly identifies a legitimate listing as counterfeit supports an enforcement action against a third party seller with no described recourse.
A complete allocation is published, and the exposure the product creates sits with the customer. Section 10.1 caps Marq Vision's total liability at the amounts paid in the twelve months before the event, and section 10.2 excludes consequential damages for both sides, with indemnification obligations carved out. Section 9.1 gives the customer an indemnity for third party claims that the platform infringes intellectual property. Section 8.2 warrants compliance with law, noninfringement of the platform and absence of malicious code. The catch is what the indemnity excludes. Section 9.1(d) removes any claim arising from the customer's use of the services to request removal of content or send soft notices, which is the core function of the product. Section 9.2 requires the customer to indemnify Marq Vision for actions taken at its instruction and for any consequence of a party being wrongly left off the whitelist. Section 8.4(b) says Marq Vision is never responsible for enforcement against an entity that should not have been targeted if it was not properly whitelisted. Wrongful takedown liability, which the vendor's own material names as the central risk, is allocated to the brand.
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.
One uncommon integration and none of the ordinary ones. The vendor states integration with US Customs and Border Protection to track recordation statuses, connecting the platform to the government system through which recorded marks are enforced at the border. Portfolio upload in one click, with clients and trademarks populated automatically from filing record details, implies a working connection to USPTO filing data. What is absent is everything a firm would ask about. There is no IP management or docketing system integration, which matters less because this product is itself the docketing system, and no document management or practice management connector. No API or export documentation is published beyond report generation in one click.
No integration into the systems legal or IP work already lives in is published, and none is claimed. No IP management or docketing system is named, no case management or document management system, no connector list, no API or developer documentation, and no integrations page exists in the navigation or the footer. What the platform does connect to is the enforcement estate rather than the customer's own systems, with the privacy policy noting that third-party account credentials may be provided so that account information transmits into the customer's MarqVision account, which is marketplace access rather than workflow integration. Trademark management is delivered through a separate product on a separate domain, and nothing describes data moving between it and the enforcement platform.
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.
Nothing is published. No hosting provider is named, no region or data residency commitment is published, and no deployment options are described. The vendor is a small United States company serving United States trademark practice against the USPTO, so residency is a narrower question than for multinational vendors. It is still unanswered for a platform holding unfiled logo designs and complete client portfolios.
Nothing published addresses where the platform runs or how customers are separated within it. No cloud provider is named, no hosting region or country is stated, no residency option is offered, nothing separates processing from storage, and no tenancy or isolation model is described. The only geography published is the office list: Los Angeles, San Francisco, New York, Seoul, Paris, Shanghai and Tokyo. The only cross border statement is the privacy policy's reference to the EU-U.S. Privacy Shield Framework. That reference is itself a problem, since Privacy Shield was invalidated in 2020 and replaced in 2023, and the policy carrying it took effect in February 2026. For a platform operated from Delaware with offices in Seoul, Shanghai and Paris, processing brand and seller data across 118 countries, the lack of any residency statement is a material gap.
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.
No certification, attestation, trust center or security page is published. No SOC 2 of either type, ISO 27001, named auditor, penetration testing partner, encryption statement or security documentation of any kind was found on the pages read. The material is absent rather than gated, and no trust surface of any kind is published.
A trust center is linked, and no attestation is stated on any page that renders. The footer links directly to a Vanta trust center for MarqVision, a real route rather than a badge image, which suggests a monitored compliance program. Nothing else a reviewer needs is published. No standard is named on the site itself: neither SOC 2 nor ISO 27001 appears on the home page, the product pages, the terms or the privacy policy, so a buyer learns of any certification only by leaving the site. No auditor, certificate number, report period, scope or trust services criteria is published, and no penetration testing or vulnerability program is described. The Vanta portal renders in the browser and returned no readable content, so it may say more. The vendor's own pages state no certification.
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.
Nothing located. No foundation model provider, model family or version is named, no distinction is drawn between proprietary and third party models, and no subprocessor list was found. The vendor describes advanced image recognition and AI powered brand recognition throughout, which characterizes capability rather than provenance, and given the image recognition engine is the entire technical proposition, whether it was built in house or licensed is the single most material supply chain question about this product and it is unanswered.
The capabilities are named in proprietary terms and nothing identifies what sits underneath. The published vocabulary is extensive: Atomic Product Detections using generative AI, a multi agent AI that decides whether an infringement is valid, an LLM framework that generates keywords in any language, image recognition and text analysis, and models said to be trained on successful enforcement data. None is attributed. No model, version or provider is named, and nothing says whether any component is built in house or reached through a third party, which a brand handing over its product catalog and seller data would ask. No subprocessor list exists in the terms, the privacy policy or anywhere else; the privacy policy describes third parties only as agents and affiliated businesses. Nothing says where inference runs, and there is no commitment to notify customers when the models change. Trademarked capability names stand in for provider identification.
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.
Everything a buyer needs is published without contacting anyone. The entry price is $299 per month on a dedicated pricing page, and the unit of charge is unlimited text and image clearance searches rather than a fee per search. Discounts are stated for twelve month commitments, with no contract and no credit card required to start. The vendor positions its price against the market openly, and independent coverage backs the comparison, noting that competing trademark search tools commonly charge $100 to $800 per search. A firm can work out total cost and compare it with incumbents from published material alone. The tiers above the entry price were not read, and it is unclear which of search, docketing, watch and brand protection the entry price includes; those pages may say more.
The pricing structure is published in the terms in detail, and no figure appears anywhere. Fees sit on an order form. Section 3.1 says they are based on services purchased rather than actual use and are not refundable, so this is a committed subscription rather than a consumption model. Under section 7.4 the term renews automatically for twelve month periods, with thirty days notice to end it. Two escalation terms are published. Section 3.5 says fees may be revised on renewal, resulting in an increase every renewal term. Section 3.6 caps increases on agreements longer than a year at seven percent on each anniversary unless the order form says otherwise. Late payment carries one percent a month, with suspension possible after fifteen days. Coverage is itself commercial: the order form sets the regions, and section 8.4(a) keeps discretion over which marketplaces come first. No number, tier or unit is published, no pricing page exists, and the entry routes are a free brand scan and a demo request.
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.
Coverage of the trademark lifecycle is complete, and jurisdictional coverage is narrow and largely unstated. Within its scope the product spans clearance search, docketing with deadline management, watch on new filings, Office Action handling, and enforcement after registration through online brand protection. Online coverage is described specifically: the vendor names independent Shopify stores as a capability it claims as unique, reasoning that Shopify has no central marketplace and over three million stores run on custom domains. Coverage appears limited to the United States. The USPTO is the only trademark office named across search, watch and docketing, the Customs and Border Protection integration is specific to the United States, and no international registry, Madrid Protocol handling or foreign filing coverage is published. A platform built for one office is a legitimate design choice and a real coverage limit for any firm with international clients.
Coverage is documented by industry and by right rather than by buyer role, and on its own terms it is thorough. Seven industry pages cover beauty, fashion, automotive, pharmaceuticals, food and beverage, network marketing and natural health products, each with its own counterfeiting patterns. Six solution lines are documented separately. The anticounterfeit page says the platform protects trademarks, copyrights, patents and designs. Reach is given as more than 1,500 channels across 118 countries, with offices in seven cities and the site in English, Korean and Japanese. What is missing is the buyer. No page addresses legal, IP counsel or brand protection teams as roles. The people named in customer stories are mostly in commerce and sales, with one general counsel, so a reader cannot tell whether the legal function or digital commerce buys the platform. No portfolio size is stated and nothing says where coverage stops.
The 12 legal signals, side by side
Recorded rather than graded. These are the questions a practitioner has to answer before a tool touches a client matter, and the answers are taken from public material only.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
No statement either way. The quoted line is the vendor's clearest statement about who the platform serves, and it says nothing about what happens to what they upload. No statement was located on whether candidate marks, uploaded logo images, client portfolios or search histories are used to train or improve the recognition models. The incentive is visible: the core technology is a visual similarity engine, and unfiled logo designs uploaded for clearance are exactly the labeled image data that improves such a model.
Docketing also takes in entire client portfolios in one click. This is not a negative commitment. Checked the home page, the FAQ, the solutions pages including docketing and brand protection, and the site navigation on 29 Aug 2026.
Training is described in product material and not addressed in the agreement, so the position is a policy rather than a contract term. The content protection page states the quoted position, describing an antipiracy method built on models trained on successful enforcement data. A published article on false positives describes borderline cases going to analysts, whose decisions are then fed back into training to shrink the review queue over time.
Enforcement data and analyst decisions come from work on customer matters using customer product data, so customer material is in the training loop, even though neither statement says customer data. The agreement does not name training. Section 2.5(b) of the MARQ AI terms grants a broad right to use information the customer submits, including information about its customers, distributors, inventory and products, to improve the platform and the business models and services of Marq Vision, its subsidiaries, vendors or other customers.
It also allows anonymized data to be used for new products. That is a wide improvement right that does not name training.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
Uploaded images, generated reports and portfolio data have no stated retention. No retention period is published for uploaded images, search queries, generated reports and opinion letters, or docketed portfolio data, and no deletion right is described. Retention is implicit in the product design rather than stated as a policy: docketing is a system of record that holds client portfolios and deadlines by function, and generated reports and opinion letters persist for later use, and nothing states for how long, under what terms, or what happens when a firm stops subscribing, which matters for a month to month product sold with no contract. Checked the docketing page, the home page, the FAQ and the site navigation on 29 Aug 2026.
Retention is acknowledged and never quantified. The privacy policy says personal information is kept as long as necessary for the purpose it was collected for or to meet legal or regulatory requirements, and no longer than reasonably necessary, with no period for any category. The terms are silent both ways. Section 7.3 says that on termination the license ends and the customer stops using and destroys documentation, with no deletion or return obligation on Marq Vision for customer property.
Nothing says how long detection records, evidence packages, seller intelligence or enforcement history are kept, and no retention setting is offered. Two provisions push the other way. The privacy policy says some information may remain in records after an account is deactivated, and that aggregated data derived from personal information may still be used after the customer updates or deletes it. Section 4.1 of the terms confirms Marq Vision owns anonymized data outright.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
Walls by client, by matter and between firms have no described mechanism. No permission model, client level access restriction or tenant segregation description was located. The vendor describes a dashboard giving a bird's eye view of all clients, their trademarks, related office actions and detected risks, which is a firm wide view across the whole book of business and raises rather than answers the question of whether access can be restricted by client or matter within a firm.
The cross customer question is equally open: this platform serves many trademark firms clearing marks against each other's clients, and nothing describes what separates one firm's searches and portfolios from another's. Checked the docketing page, the watch page, the home page and the site navigation on 29 Aug 2026.
Checked the home page, the anti-counterfeit, brand protection, content protection, impersonation and MarqLaw pages, the MARQ AI terms in full and the privacy policy in full on 2 September 2026. Nothing describes segregation between customers or isolation of one brand's data from another's. No tenancy model is stated and no access boundary is described beyond the customer's own obligation to protect its password. The question has a specific edge on this product that the published material does not reach.
Section 2.5(b) of the terms permits use of information a customer submits, including data about its customers, distributors, inventory and products, to improve the business models and services of Marq Vision's other customers, and the platform maintains a global seller database of more than ten million entries built from enforcement work. A brand would reasonably want to know what separates its catalog and channel intelligence from a competitor's on the same platform, and nothing published answers that.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
No notice promise protects customers if an agency seeks their data. No government or law enforcement request clause, no commitment to notify a customer before producing its data, and no transparency report were located. The question has a concrete edge here: the platform integrates directly with US Customs and Border Protection to track recordation statuses, an operational relationship with a federal enforcement agency.
Nothing published describes what data flows in that direction or what would happen if an agency asked for more. Checked the docketing page, the brand protection page, the home page and the site navigation on 29 Aug 2026.
The commitment is in the operative agreement and is specific. Section 5 of the MARQ AI terms allows disclosure of confidential information where a valid court order, judicial process or regulatory authority requires it. The receiving party must then use best efforts to preserve confidentiality and give the quoted notice promptly, unless a court order prevents it. Naming the nature, scope and contents of what must be disclosed to the customer goes beyond bare notification.
The obligation is mutual and survives termination. Two limits remain. Anonymized data is excluded from confidential information at section 5(d), so the notice commitment does not reach it. The privacy policy also takes a broader position for personal information. It reserves the right to access, read, preserve and disclose any information Marq Vision believes necessary to comply with law or court order, enforce its agreements or protect its own rights, property or safety, with no notice commitment.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
The source is named and public, with no license basis stated and none obviously needed. The searched corpus is the USPTO trademark database, including design and image records, a single named public authority rather than a set of undisclosed databases. The vendor's central technical claim rests on it, since image recognition is said to replace the USPTO design search code taxonomy. On the brand protection side, monitored platforms are named, including independent Shopify stores.
What is absent: no statement of the licensing or access basis for USPTO data, and no update lag or refresh frequency, although watch depends on catching applications at filing. Nothing says how online listing data is obtained or how current it is.
The sources are identified by platform and scale, and no rights basis is stated for any of them. Coverage is more than 1,500 marketplaces, social platforms, websites and app stores across 118 countries. Platforms named in the product material include Amazon, Alibaba, Taobao, TikTok, BiliBili, YouTube, Google and Meta, and there is a global seller database of more than ten million entries. The customer's own genuine product data is the comparison set for SKU matching.
The legal basis for gathering any of it is absent. Nothing says whether marketplace listings are collected under platform agreements, APIs, brand registry programs or by crawling against those platforms' terms of use. That matters commercially as well as legally. The company markets marketplace partnerships in Southeast Asia and China and holds direct takedown authority on YouTube, which implies negotiated access on some channels and not others. No license, agreement or permission is described for the underlying listing data.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
The product carries its own status tracking. Status is a core function rather than an optional layer: docketing tracks live deadlines and critical dates across a portfolio with a daily email, the dashboard shows office actions and detected risks, and portfolio data is populated automatically from filing record details. Customs and Border Protection recordation statuses are tracked, which is status monitoring at the enforcement layer.
Watch exists to catch newly filed applications at filing. No citator or status authority is named as a distinct source, and nothing says how current the status data is or how often it refreshes. No explicit statement says that dead, abandoned or canceled marks are flagged in search results, which is the case that matters for a clearance opinion.
Checked the home page, the anti-counterfeit, brand protection, content protection, impersonation and MarqLaw pages, the MARQ AI terms and the privacy policy on 2 September 2026. No public material addresses subsequent history, treatment or good law checking, and none is claimed. The product does not retrieve primary law, so a citator has nothing to operate on and the question does not bite in its usual form. The analogous currency question in this product class is whether the intellectual property right being enforced is still valid and in force, and it is addressed only as the customer's responsibility rather than as a system check: section 8.3(d) of the terms requires the customer to warrant that it is the rightful owner or licensee of all rights subject to the services, and section 8.4(b) states that the sufficiency of the customer's own registrations is outside Marq Vision's control. Nothing describes the platform verifying a registration's status before enforcing on it.
Refusal and Uncertainty Behavior
What does the product do when the answer is not in the corpus?
An unsure recognition engine has no described way to say so, and the gap matters most on the capability the vendor markets hardest. Nothing describes what happens when the recognition engine is unsure: whether a marginal visual similarity is flagged, whether a confidence score accompanies a match, whether a watch alert separates a strong conflict from a weak one, or whether the system will report that it could not reach a conclusion.
The brand protection claim sharpens this. The vendor says the engine finds counterfeits even in blurred or obscured product images, the condition under which a visual model is least certain and most likely to be wrong in both directions. No confidence signal is described for output that can support enforcement against a third party seller. Checked the brand protection page, the watch page, the search material, the FAQ and independent coverage on 29 Aug 2026.
Confidence is scored and shown, and nothing describes the system declining. The impersonation material says every classification from the multi agent system comes with its reasoning and a risk score, so a user sees a rationale and a confidence measure rather than a bare determination. A published article on false positives adds that a tiered model routes only borderline cases to analysts, which is a confidence threshold at work even though the threshold is never defined.
Nothing says what the system does when it cannot reach a determination at all, whether it abstains, holds an item or escalates by default. Nothing says what confidence is required before an automated enforcement action goes ahead. The customer's whitelist is a suppression list the customer supplies, not a behavior of the model.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
None located, with the instrument named. General web searches combining the vendor and product names with court, order, sanction, negligence and clearance terms returned nothing on 29 Aug 2026, and no named docket database, trademark 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 matches Corsearch rather than the patent drafting tools: no legal authority is generated, so the analogous adverse finding would be a proceeding contesting the adequacy of a clearance search or an opinion letter, or a wrongful enforcement action arising from a false positive brand protection match, neither of which surfaces as a sanctions order.
Searched the AI Hallucination Cases database maintained by Damien Charlotin, and reporting that draws on it, on 2 September 2026, for MarqVision and Marq Vision Inc. No court order, opinion or disciplinary record naming the product was located. This is a statement about the public record, not a finding about the product. The search fits this product poorly. MarqVision generates takedown notices and enforcement packages rather than legal citations, so its typical failure is wrongful enforcement against a legitimate seller rather than a fabricated case reference.
The vendor's own material names that risk, citing misrepresentation liability under 17 U.S.C. 512(f), tortious interference and suspension of platform reporting as the consequences of false positives. Those outcomes surface as civil claims or platform sanctions, which no hallucination tracker indexes.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
Opinion letters carry an attorney's name, yet ethics rules go unmentioned. No named ethics opinion, USPTO Rules of Professional Conduct reference, 37 CFR citation, duty of competence discussion or bar guidance was located. The gap stands out because the vendor sells only to legal professionals and generates a customized opinion letter, a document issued under an attorney's name and judgment, which engages the duty of competence more directly than a search result does.
Checked the home page, the FAQ, the solutions pages, the published rankings material and the site navigation on 29 Aug 2026.
An authority is named and no instrument is, and the claim sits awkwardly against the agreement. The anticounterfeit page states that MarqVision's AI driven brand protection solutions comply fully with the ABA's ethics guidelines, ensuring the highest standards of legal integrity in every action. Naming the American Bar Association puts this above a bare gesture at professional standards. But no rule, model rule or formal opinion is identified, nothing states which obligations are said to be met or how compliance is assessed, and ABA Formal Opinion 512 on generative AI is not referenced.
The claim also runs against section 1.2 of the company's own terms, which states that Marq Vision does not and cannot provide legal guidance or advice and that no attorney client relationship arises, since the ABA's rules bind lawyers rather than software vendors. A reader is left with an assertion of compliance with a framework the agreement says does not govern the service.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
Savings claims only, and the savings are stated in money against a named market baseline rather than as a share of time. The vendor says clearance searches that would otherwise cost hundreds of dollars each are included in a $299 per month unlimited subscription. Independent coverage supports the baseline, noting that competing tools commonly charge $100 to $800 per search. That makes the claim testable. It also bears on billing.
Search costs are typically passed to the client in trademark practice, so a shift from per search charges to a flat subscription changes what a firm can pass through and what it must absorb. Nothing published addresses that, or how clearance time with AI should be billed, and no exportable record shows how much of a search or opinion letter the machine generated.
Time and cost savings are published with figures, and nothing addresses the bill. The company says Marq AI saves its expert teams 180 times the time previously needed to enforce, that enforcement runs 50 percent faster than traditional brand protection companies, and that mean time to resolution is fifteen minutes. It publishes a Saturation Rate metric and revenue recovery framing meant to show enforcement paying for itself.
Nothing addresses how AI assisted work is recorded, billed or disclosed, and no per matter record of it was located. Here the buyer is usually the brand owner rather than a firm billing a client, so time saved accrues to the customer rather than changing an invoice. That is not the whole picture: MarqLaw engages partner law firms who bill for litigation, customs and investigation work supported by platform output, and nothing published says how AI assisted preparation shows up in those engagements.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
Client guideline paperwork is missing entirely. No trust center, security page, named certification, subprocessor list, named model provider, data processing agreement or documentation request route was located, so a firm has nothing to forward to a client and nowhere to point one. Checked the home page, the FAQ, the solutions pages, the pricing material and the site navigation on 29 Aug 2026.
Checked the MARQ AI terms in full, the privacy policy in full, the home page and the anti-counterfeit, brand protection, content protection, impersonation and MarqLaw pages on 2 September 2026. No subprocessor list exists in any form. The privacy policy describes third parties only by category, covering agents performing tasks on Marq Vision's behalf, affiliated businesses and third-party law firms, advertisers and partners receiving de-identified data, without naming a single entity.
No model or AI provider is named anywhere on any surface. No data processing agreement is published, and no forwardable client-facing disclosure material was located. A Vanta-hosted trust center is linked from the footer and would be the natural home for a subprocessor list; it renders client-side and returned no readable content, which is recorded as a retrieval limit rather than an absence. A brand asked by a partner or regulator which entities process its catalog and seller data could not answer from anything published.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
A partial record, on a product built around producing evidence. On the enforcement side, the vendor says evidence of infringement can be compiled into a report in one click for use if litigation follows. Independent coverage says the company was founded to cut the manual work of collecting listing screenshots for enforcement cases, so evidence capture is the problem it was built to solve. On the clearance side, a report and a customized opinion letter are generated, and the opinion letter is the document produced years later in infringement litigation on what was known before adoption.
Two elements are missing, and one matters more here. Nothing indicates that output records which model produced a similarity match or when, and no record shows the attorney reviewed it, although the generated opinion letter is issued in the attorney's name.
Evidence capture is a published product function, not a byproduct. The anticounterfeit page says the platform automatically extracts and stores the evidence needed to shut down illicit activity and generates full self documentation reports. MarqLaw is marketed on courtroom ready evidence packages, and the impersonation material describes a report ready enforcement package for each channel. A published article says consistent enforcement backed by a documented record of every action is what drives counterfeiters elsewhere.
The anticounterfeit page also promises full visibility into when and how AI is used to enforce and protect a brand. What is not described is an exportable record of that visibility. Nothing captures which model or agent made a given determination, at what confidence, or which reviewer confirmed it, which is what a party challenging a takedown or bringing a 512(f) claim would seek.
The questions both sides leave open
Derived from the records above rather than written, so it cannot favor either vendor. Take these into both conversations and ask each side the same question.
- Deployment Model and Data Residency
- Ethical Walls and Matter Segregation
- Outside Counsel Guideline Readiness
Which one fits
Choose Huski.ai if
- You want to know the price before the first call. Huski.ai publishes 299 dollars per month with the unit stated as unlimited text and image clearance searches rather than metered per search, discounts for twelve month commitments, no contracts and no credit card required to start, in a market where independent coverage reports competing trademark search tools charging between 100 and 800 dollars per search.
- Your clearance work is visual and design codes get in the way. Huski.ai applies image recognition directly to trademark work, stating that this removes the need for design search codes: a user uploads an image and receives visually similar USPTO results in seconds, and the same engine runs watch against newly filed applications so a conflict surfaces at filing rather than at publication, in time for a letter of protest.
- You want search, docketing, watch and enforcement in one place. Huski.ai covers the trademark lifecycle in a single platform, with one click portfolio upload populating clients and marks from filing records, deadline tracking and office action management, and an integration with United States Customs and Border Protection to track recordation statuses.
Choose MarqVision if
- You want detection accuracy measured and the mechanism explained. MarqVision publishes 97 percent on enforceable item detection using SKU level comparison against the brand's own product data, 99.8 percent across more than 48,000 domain impersonation incidents and 94 percent SKU matching on unauthorized sales, states openly that these are internal benchmarks, and publishes what a false positive costs by naming misrepresentation liability under 17 U.S.C. 512(f), tortious interference and platform reporting suspension.
- Your problem is scale across channels you cannot watch by hand. MarqVision scans more than 1,500 marketplaces, social platforms, websites and app stores across 118 countries, analyzes video frame by frame to catch counterfeits pushed through social clips, tracks gray market resellers across a database of more than ten million sellers, and coordinates customs work, criminal enforcement and field investigations through partner law firms.
- You need control over who gets enforced against. MarqVision runs enforcement on a letter of authorization with a customer maintained whitelist of sellers the brand does not wish to act against, describes a tiered model routing borderline cases to human analysts with a reviewer confirming significant decisions, and returns the reasoning and a risk score with each impersonation classification.
In summary
Huski.ai
Huski.ai is an AI trademark platform for IP attorneys, paralegals and legal staff, covering clearance search, docketing, watch and online brand protection. It is built around image recognition the vendor states removes the need for design search codes, with the same engine applied to newly filed applications and to ecommerce monitoring including blurred or obscured product images. According to the AI Legal Index, its price is public: 299 dollars per month for unlimited text and image clearance searches with no contract required. No security certification, training or retention position, model provider or hosting statement is published.
MarqVision
MarqVision is an AI brand protection platform that finds and removes online infringements. It scans more than 1,500 marketplaces, social platforms, websites and app stores across 118 countries, compares findings against a brand's own product data at SKU level and analyzes video frame by frame, with further lines covering impersonation, content protection, unauthorized sales and coordinated legal enforcement. The AI Legal Index records measured detection figures that the vendor describes as internal benchmarks, and terms that state in section 1.2 that the services include no legal representation and create no attorney client relationship. Its agreement also caps liability and gives the customer an infringement indemnity. No model provider, hosting region or price is published.
Questions buyers ask
Huski.ai vs MarqVision: which is better for brand protection?
MarqVision publishes far more: measured detection figures, a full liability allocation and a clear statement of what the service is not. Huski.ai publishes its price in full and very little else. It publishes no certification, no data handling position and no model provider. From the AI Legal Index, based on each vendor's own published materials as of September 3, 2026. No vendor pays for placement.
How much does each cost?
Huski.ai publishes 299 dollars per month covering unlimited text and image clearance searches, with discounts for a twelve month commitment, no contract and no card required to start. MarqVision publishes no figure at any level: fees sit on an order form, are stated to be based on the services purchased rather than usage and to be non refundable, with automatic twelve month renewals, a candid statement that fees may rise at each renewal, and a seven percent annual cap on multi year agreements. From the AI Legal Index, based on each vendor's own published materials as of September 3, 2026. No vendor pays for placement.
How accurate is the detection?
MarqVision publishes figures and names them as internal benchmarks: 97 percent on enforceable item detection, 99.8 percent across more than 48,000 domain impersonation incidents and 94 percent SKU matching on unauthorized sales, with only one carrying a sample size and none a described test set. Section 2.4 of its terms separately states that the customer shall not rely on the accuracy of platform data. Huski.ai publishes no accuracy figure, although independent trade coverage ran an image search and published a worked result. From the AI Legal Index, based on each vendor's own published materials as of September 3, 2026. No vendor pays for placement.
Do either say what the service is not?
MarqVision states it plainly in section 1.2 of its terms, headed no legal representation: the services include no legal representation, its personnel cannot provide legal advice, no attorney client relationship arises, and soft notices are not a substitute for legal claims. Its anti counterfeit page separately states that its AI driven solutions comply fully with the ABA's ethics guidelines. Huski.ai restricts its users to attorneys, paralegals and legal staff, and nothing states that its generated opinion letters are not legal advice. From the AI Legal Index, based on each vendor's own published materials as of September 3, 2026. No vendor pays for placement.
What do Huski.ai and MarqVision both leave unpublished?
Neither names a model or a provider behind the recognition and classification each is sold on. Neither states a hosting region, a residency option or a tenancy model. Neither publishes a retention period or a deletion commitment for customer content, which matters because both hold unfiled designs, product catalogs or client portfolios. And neither publishes an AI governance mechanism: no accountable owner, no pre release testing and nothing on whether detection quality holds evenly across languages, marketplaces or image conditions. From the AI Legal Index, based on each vendor's own published materials as of September 3, 2026. No vendor pays for placement.
Neither vendor paid for inclusion, placement or a grade, and neither reviewed this page before it published. Everything on it comes from public material on the dates shown. How the index grades.
Two clauses in MarqVision's agreement should be read alongside its accuracy figures. Section 2.4 states that Marq Vision assumes no liability for the quality, accuracy or validity of the data gathered in or by the platform and that in no event shall the customer rely on it, and sections 9.1(d) and 8.4(b) exclude from its indemnity any claim arising from the customer's use of the services to request removal or send notices, and place responsibility on the customer where a party was not properly whitelisted. The agreement governs. Huski.ai publishes no security certification, trust surface, training position, retention period or hosting statement, so most of what is known about it is an absence rather than a set of unfavorable terms. Neither vendor reviewed this page.