Ankar
Ankar is an AI platform for patent professionals covering the work from invention capture through to enforcement. Its drafting assistant lets attorneys upload prior applications and templates to build a personalised style guide, then works through an application section by section with consistency checks that keep claims and descriptions aligned; every suggestion is non-destructive and every output editable. A patentability engine searches more than 150 million patents and publications, returning prior art with structured context on why it matters and how it compares, with the similarity criteria and result volume set by the user rather than fixed. Separate modules handle office action responses, claim charts, small molecule analysis, sequence analysis, traceable citations, and guidelines and case law, alongside invention disclosure and research intelligence tools aimed at R&D teams rather than attorneys. An infringement detection suite lets an IP team define what to flag and returns product-level matches with visual evidence, source links and technical context. Drafting works natively with Microsoft Word. The platform runs either as a hosted service in a customer-selected AWS region or inside the customer's own virtual private cloud, and the subscription terms state that customer content is not used to train, retrain or fine-tune any AI models, by Ankar or by third parties, and that the customer owns all output. Named customers include the global law firm Vorys, Valeo, Nexans, L'Oreal and Motorway UK. Ankar is operated by Ankar AI Ltd.
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 mechanism throughout and there is no conventional product beneath them. Ankar describes itself as an AI-native platform and every module sold is a generative or machine learning capability: drafting that works through an application section by section against a learned style guide, patentability analysis over more than 150 million patents and publications, office action responses, claim charts, small molecule and sequence analysis, and infringement detection that slices claims and returns product-level matches. The corpus is public patent and publication data, which is a commodity substrate rather than a licensed content asset, so removing the models leaves a searchable bibliographic database rather than a product a firm would buy. The vendor's own framing is consistent, positioning the platform as an operating system for patents spanning generation, prosecution and protection rather than as conventional software with an AI layer added. Verified 2 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 sold as a named capability: Traceable Citations is one of the platform's listed modules. Prior art is returned with structured context rather than as a list, described as explaining why a reference matters, how it compares and what to do next, and the vendor states explicitly that the tools are not a black box and offer transparent reasoning with adjustable criteria. Infringement results carry visual evidence, source links and technical context. Hallucination is disclosed concretely and in the operative agreement rather than in marketing: clause 7.2 of the terms states that outputs may contain inaccuracies and omissions and that the legal status of generative output is uncertain, and it requires the customer to review, validate and approve all output before filing or submitting it. What is absent is measurement. No accuracy figure, benchmark, evaluation or test set appears on any surface read, so a buyer gets a described method and a candid warning without a number to test either against. Verified 2 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.
The oversight model is the most completely documented in this pull, and it is stated in three places that reinforce each other. Contractually, clause 7.2 of the terms makes the customer solely responsible for reviewing, validating and approving all output before using, filing, submitting or otherwise relying on it, and it names the work products this covers: patent applications, office action responses, litigation, prosecution strategy, legal analysis and client advice. The same clause states that Ankar does not file, prosecute, maintain or monitor intellectual property rights and is not responsible for filing deadlines, prosecution decisions or legal positions, which draws the line between what the system does and what the attorney owns. At product level the review surface is described concretely: all outputs are editable, every suggestion is non-destructive, and a user can revise, adapt or override at any time. And the thresholds are the user's rather than the vendor's, with the infringement module letting an IP team set similarity criteria, technical focus areas and result volume, and the prior art engine offering adjustable criteria. Nothing in the platform is described as acting unattended, so the limb asking what runs alone does not bite and is recorded as inapplicable rather than failed. Verified 2 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.
Three customers are named with outcomes described rather than asserted, and two carry attributed quotes from identified executives. Valeo, described as one of the world's top patent filers, partnered with Ankar to modernise drafting by automating formatting and aligning output to internal templates, with Murielle Khairallah, VP Intellectual Property, quoted on the record. Nexans used the infringement module to isolate 30 highly specific product-level matches after years of generic tools, with Boris Welzer, Director IP, quoted saying he had been looking for such a tool for fifteen years. Motorway UK is named for patentability review under a surge in disclosures. Figures are published: a 40 to 60 per cent reduction in time spent on drafting and office actions, and over 90 per cent adoption of the prosecution tools after trial. The global law firm Vorys and L'Oreal are named elsewhere in the company's own materials. What keeps this below A is method and dating: the 40 to 60 per cent and 90 per cent figures carry no measurement basis, sample or date, and no figure is tied to a named customer. A dedicated case studies page exists and was not opened this pass. Verified 2 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.
Four limbs are met and one is not. The training prohibition is contractual and unusually complete: clause 2.3 of the terms bars Ankar from using customer content to train, retrain, fine-tune or otherwise improve any artificial intelligence models or to develop products or services, and it extends the bar to third parties Ankar might authorise. Ownership runs the right way, with clause 2.2 confirming the customer retains all rights in inputs and owns all outputs, and Ankar assigning any rights it might acquire by operation of law. Confidentiality is structural rather than incidental: clause 2.1 defines customer content as the customer's confidential information by default, so no marking is required, and clause 5.2 permits disclosure only to personnel and subcontractors under equivalent written obligations. Deletion is stated, with a thirty day retrieval window and destruction thereafter, and the data processing addendum adding deletion at the customer's choice with a certificate on request. Isolation is available in the strongest form through customer virtual private cloud deployment. The limb that fails is the express one: neither privilege nor work product is named anywhere, and for a product handling unfiled invention disclosures and prosecution strategy that is the doctrine a firm would want addressed. Verified 2 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.
Clause 7.2 of the terms is the most complete professional responsibility provision located in this corpus, and all of it sits in the operative agreement rather than a footer. It states what the product is and is not: outputs may contain inaccuracies and omissions, the customer should not rely on the services or any output for advice of any kind including legal advice, and output is not a substitute for advice from a qualified professional. It states the supervision obligation and names the work products it attaches to, making the customer solely responsible for reviewing, validating and approving all output before using, filing, submitting or relying on it in connection with any patent application, office action response, litigation, prosecution strategy, legal analysis, client advice or other professional work product. It draws the boundary of the vendor's role expressly, stating that Ankar does not file, prosecute, maintain or monitor intellectual property rights on the customer's behalf and is not responsible for filing deadlines, prosecution decisions, legal positions or the professional obligations of the customer or its users. And it addresses a limitation specific to this practice area that few vendors mention at all, warning that the legal status of generative output is uncertain and that outputs may not attract intellectual property protection. Jurisdictional coverage of legal content does not bite on a drafting tool operating on the customer's own inventions and is recorded as inapplicable. No bar or patent office guidance is named, which is recorded on the signal. Verified 2 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.
A design philosophy is published and no mechanism stands behind it. The company states a consistent position across its blog and product material: tools should empower experts rather than replace them, reasoning should be transparent rather than a black box, criteria should be adjustable by the professional, and speed is meaningless if a practitioner cannot stand behind the output. Those are governance principles in substance and they are reflected in the product through non-destructive suggestions and user-set thresholds. What no published material supplies is the machinery. No responsible AI page or governance framework exists, no individual or function is named as accountable for model behaviour, nothing describes what is evaluated before a model or module change ships, and there is no ISO 42001 or equivalent certification; the trust centre's 106 controls and ISO 27001 are information security rather than AI governance. Nothing anywhere addresses uneven output, which matters on a platform with separate small molecule and sequence modules where performance across technical domains is precisely the question. Checked the terms, data processing addendum, trust centre and blog on 2 September 2026. Verified 2 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.
Four of the five elements are strong and one has a specific, documented hole. Retention and deletion are precise: a thirty day post-termination retrieval window under clause 9.6, destruction thereafter, and under the data processing addendum deletion or return at the customer's choice with backups cleared on the standard cycle and a deletion certificate available on request. Incident practice is stated properly, with notification without undue delay and where feasible within seventy-two hours, and the notice content enumerated across five items including categories and approximate numbers of records affected. Access control is contractual as well as technical, with per-user accounts that may not be transferred or shared, a customer obligation to report suspected unauthorised access, and a Security Commitment at clause 9.2 binding Ankar to the safeguards in its SOC 2 Type II report and the requirements of its ISO 27001 certification. Subprocessor governance is excellent on paper: ten days prior written notice of any addition or replacement, an objection right, materially similar written obligations, and Ankar remaining liable for subprocessor performance. The hole is the list itself. Clause 6(a) states that current subprocessors are set out in Schedule 2, and the published addendum ends after Schedule 1, so no subprocessor is actually named. AWS is identified in the transfers clause, which tells a reader the infrastructure and nothing about the AI layer. Verified 2 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.
The allocation is published in full, tiered, and mutual. Clause 7.3 caps each party's aggregate liability at the amounts paid in the twelve months before the first event giving rise to it, then sets a separate cap at two times that figure for breach of the confidentiality clause, breach of the security commitment, or claims arising under the data processing addendum, which is an uplift aimed precisely at the failures that would matter most to a firm handing over unfiled inventions. Sitting outside the caps entirely are gross negligence, fraud, wilful misconduct, the indemnification obligations, and material breach of the acceptable use policy. Indirect and consequential damages are excluded symmetrically. Clause 8.2 gives the customer a real indemnity running the right way, with Ankar defending third-party claims that its own intellectual property infringes, and remedies ordered as procure, replace or modify, or terminate with a pro-rated refund. There is an affirmative warranty at clause 6.2 that the services will operate in substantial conformity with the documentation, with a stated remedy of termination and a pro-rated refund if a nonconformity is not remedied within thirty days. What is not offered is any warranty on output, expressly disclaimed at 7.1 and 7.2, and no insurance position is published. Beta services and free trials are carved out of the indemnity, which is stated plainly. Verified 2 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.
One integration is named and real, and the rest is contractual scaffolding without content. Ankar has published a redesigned editing experience for native Microsoft Word compatibility, which matters on this product because patent drafting happens in Word and a tool that cannot round-trip a specification is a tool a drafter leaves. Clause 1.6 of the terms contemplates third-party integrations that a customer connects by linking authorised user accounts, and describes the risk allocation for them, but names not one such integration and states they are optional. Beyond that nothing was located: no IP management or docketing system is named, no document management system, no patent office filing interface, no connector list and no API or developer documentation. For a platform selling across the full patent lifecycle, the absence of any named link to the docketing and IP management systems a firm already runs is the gap, and a contract clause anticipating integrations is not the same as a documented one. Checked the terms, data processing addendum, trust centre, blog and site navigation on 2 September 2026. Verified 2 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.
Two deployment models are defined in the agreement and residency is customer-selected and named. Clause 9 governs hosted delivery and clause 9.1 states that customer content is processed and stored solely in the data hosting location set out in the order form. The data processing addendum makes that concrete rather than abstract: the hosting region is selected by the customer at onboarding, with UK and EU tenants stored and processed in UK and EU AWS regions and Paris and London named as examples, and US tenants in US AWS regions. Naming both the cloud provider and specific regions is more than most records in this corpus manage. Clause 10 sets out the alternative in detail, licensing the platform for installation and operation inside the customer's own virtual private cloud, with the number of concurrent installations set on the order form, deployment assistance, an express allocation of responsibility for the local environment to the customer, minimum system and configuration requirements covering GPU architectures, runtime libraries, container orchestration, network topology and identity and access management, an obligation to apply updates within a commercially reasonable period, and a signed officer attestation of deletion within thirty days of termination. Clause 10.3 is candid about the trade-off, stating that local deployment may limit Ankar's ability to maintain and support the platform and that the functionality warranty is correspondingly narrowed. Verified 2 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.
Two standards are named on a genuinely open trust surface, and the evidence behind them is not published. The trust centre at trust.ankar.ai is Bastion-hosted and renders with no authentication of any kind: it displays ISO/IEC 27001:2022 and SOC 2 Type 2 by name, a count of 106 controls across business operations, human resources, cloud security and application security, four available documents, and a last-refresh timestamp updated daily. Penetration testing appears among the listed controls. The certifications are therefore publicly disclosed and named rather than sitting behind a request flow, which is the access tier that can still reach the top band. Security is contractual here rather than only asserted: clause 9.2 of the terms makes the security commitments a defined obligation whose breach triggers the enhanced liability cap, so a buyer has a remedy attached to the claim. What holds this below A is the attestation itself. No auditor or certifying body is named for either standard, no report period, issue date or expiry is published, and no scope statement identifies which services the certification covers. A control count and a daily refresh timestamp evidence a live compliance programme rather than an audited opinion a reader can weigh. Whether the four listed documents download freely or require a request was not tested, because testing it would mean submitting one, and that distinction is not what decides the grade here. Verified 2 September 2026; access tier established on re-check the same day, grade unchanged.
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 architecture is described and the models behind it are not identified at all. What a reader can establish is the shape of the deployment: hosted service on AWS with the region selected by the customer, or the platform installed and operated inside the customer's own virtual private cloud, with the virtual private cloud clause naming GPU architectures, runtime libraries and container orchestration among the dependencies a customer must provision, which tells a technical reader that inference runs on infrastructure the customer controls under that option. Clause 2.3 of the terms acknowledges third-party involvement obliquely by extending the training prohibition to third parties Ankar might authorise, which confirms that external providers exist without saying who they are. Beyond that, nothing. No model is named, no version, no provider entity, no default, and no statement of what any provider may retain. The subprocessor list that would ordinarily carry it is promised at Schedule 2 of the data processing addendum and does not appear in the published document. No commitment to notify customers of a model change was located, though the ten day subprocessor change notice would catch a change of provider. Verified 2 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 the terms in real detail and no figure appears anywhere. Charging is per seat: clause 1.2 ties authorised user accounts to the number of seats on the order form, and clause 3.3 sets out mid-term seat changes, with additions billed immediately and pro-rated and reductions taking effect at the next invoice without a refund for the current period. Fees are invoiced annually in advance, due within thirty days, with late payment attracting 1.5 per cent per month. Renewal pricing carries a genuinely buyer-favourable published cap that few vendors offer at all: clause 3.4 limits any renewal increase to the lesser of five per cent or the increase in the UK Consumer Prices Index over the preceding twelve months, with at least thirty days notice. Consumption is addressed openly at clause 9.4, where excess cloud usage relative to spend triggers a pass-through of hosting costs only after ten days written notice and a good-faith discussion. Free trials exist under clause 4.4. What is absent is any number, any tier, and any pricing page: the site's only commercial route is Book a demo, and the currency itself is set on the order form. Verified 2 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.
The buyer split is documented structurally rather than claimed, with the site navigation dividing modules between patent professionals, who get drafting, office action responses, claim charts, traceable citations and guidelines and case law, and R&D professionals, who get invention disclosure and research intelligence. That is a clear statement that the product is sold into both the law firm or IP department and the innovation function upstream of it, and the named customers bear it out, running from a global law firm in Vorys to corporate IP teams at Valeo, Nexans and L'Oreal. Technical coverage is stated with unusual precision for this axis through dedicated small molecule analysis and sequence analysis modules, which name chemistry and biotechnology as supported subject matter rather than leaving a buyer to assume. What is missing is the boundary. No patent office or jurisdiction is named anywhere, so nothing states whether the drafting and office action modules are built for USPTO practice, EPO practice or both, which is the first question a prosecution team would ask. No firm size band is given and nothing states where the product stops. Verified 2 September 2026.
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?
The published terms prohibit training on customer content. Not a policy page, the agreement.
Clause 2.3 of the terms of service is headed No Training and is among the most complete prohibitions located in this corpus. It bars Ankar from using the quoted content for the quoted purposes and adds the words or to otherwise develop products or services, closing the route by which improvement language usually swallows a no-training promise. It also extends the bar to third parties, providing that Ankar shall not authorise third parties to do what it may not do itself, which matters on a platform that does not name its model providers. The affirmative side is drawn just as tightly: the licence Ankar receives over customer content runs solely to providing the services including technical support, enforcing the agreement and complying with law. Two adjacent permissions are narrow rather than open. Clause 2.6 permits use of Usage Data for any lawful purpose but defines it to exclude customer content and to exclude anything identifying the customer, an authorised user or any client of the customer. The data processing addendum permits deidentified or aggregated data only where no individual can reasonably be re-identified. Clause 2.3 survives termination.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
A specific retention period is published and the customer cannot change it.
A specific period is published and the customer cannot vary it. Clause 9.6 gives a customer who asks before termination the quoted thirty day window to retrieve customer content, after which Ankar destroys it under clause 4.5, which requires each party to destroy or return the other's confidential information with backup copies deleted on the receiving party's standard deletion cycle. Because clause 2.1 defines Input and Output together as Customer Content and clause 2.2 confirms the customer owns the output, prompts and generated drafts fall inside both the retrieval right and the destruction obligation rather than needing to be argued in. The data processing addendum adds a customer choice between deletion and return at clause 8, with written confirmation or a deletion certificate on request. What is not published is any in-term position: nothing states how long prompts, generated drafts or uploaded prior art persist while the subscription runs, and no retention setting is offered. Customer virtual private cloud deployment moves the question entirely, since content then sits in the customer's own environment.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
No located public material addresses walls or matter level segregation.
Checked the terms of service in full, the data processing addendum in full, the trust centre and the product navigation on 2 September 2026. Nothing describes segregation between matters, between clients of a firm, or between users inside a customer account. The access model published is per-user rather than per-matter: clause 1.2 requires an individual account for each authorised user, bars transfer or sharing of accounts, and obliges the customer to report unauthorised access. Tenant separation is answered only in its strongest and most expensive form, since clause 10 allows the whole platform to be installed inside the customer's own virtual private cloud, which removes the question rather than answering it for hosted customers. The gap is worth naming on this product class: a patent firm routinely holds unfiled invention disclosures for competing clients in the same technical field, and conflicts walls are the mechanism that keeps them apart. Nothing published states whether one can be configured.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
Terms commit to notice where lawfully permitted. No transparency report located.
The commitment is express, and it goes a step beyond notice. Clause 5.3 of the terms provides that a party required by law or legal process to disclose confidential information shall, other than to the extent prohibited by law, give the quoted prior written notice and cooperate at the disclosing party's expense with efforts to contest or limit the disclosure requirement, naming a protective order as the example. Prior notice plus assistance in resisting is stronger than the bare notification most records in this corpus offer. It reaches customer material because clause 2.1 defines customer content as the customer's confidential information without any marking requirement. The obligation is reciprocal and survives termination. The data processing addendum points the same way, requiring Ankar to inform the customer before processing personal data on a legal requirement rather than on instructions, except where the law prohibits telling them. It is not the top value because no transparency report was located and nothing records how many demands have been received or answered.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
No located public material identifies the corpus behind the product’s answers.
Checked the terms of service, the data processing addendum, the trust centre, the blog and the product navigation on 2 September 2026. The corpus is described by volume and material type and never identified. The published figure is access to over 150 million patents and publications for patentability analysis, and a separate module is named Guidelines and Case Law, so at least three kinds of material are in scope: patent documents, scientific publications and legal or office guidance. Not one source is named. No patent office, database, publisher or data supplier is identified, no jurisdictions are listed, no update cadence is stated, and no licence, subscription or permission basis is described for any of it. The distinction matters unevenly across those categories: patent documents are public records in most jurisdictions, whereas scientific publications and case law reporting are commonly licensed, and the published material does not distinguish between them or say how the non-public portion is obtained. The Guidelines and Case Law product page was not opened this pass.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
No located public material addresses whether authority is checked for subsequent history.
Checked the terms of service, the data processing addendum, the trust centre, the blog and the product navigation on 2 September 2026. Nothing addresses whether authority surfaced by the platform remains current. The product class shifts what the question means: patent work turns less on whether a case has been overruled than on whether a reference is still prior art, whether a patent remains in force, and whether office guidance has been superseded, and none of those is addressed either. A module named Guidelines and Case Law exists in the navigation and its product page was not opened this pass, so it is named here rather than credited; nothing located elsewhere describes a treatment, status or currency signal attached to what it returns. The company blog does track and summarise Federal Circuit decisions and revised USPTO guidance, which shows currency is understood as an editorial matter without establishing that the product flags it.
Refusal and Uncertainty Behaviour
What does the product do when the answer is not in the corpus?
No located public material addresses what the product does when it cannot ground an answer.
Checked the terms of service, the data processing addendum, the trust centre and the blog on 2 September 2026. Nothing describes what the platform does when it cannot ground an answer, and no confidence, coverage or grounding indicator is described as shown to the user. The published answer to the same underlying risk is transparency and user control rather than abstention: the vendor states that its tools are not a black box, that reasoning is transparent, that prior art is returned with structured context explaining why it matters and how it compares, and that similarity criteria and result volume are set by the user. Clause 7.2 of the terms places the burden of judging sufficiency on the customer, requiring review, validation and approval of all output before use. That is an allocation of responsibility and a set of controls rather than an account of system behaviour when retrieval returns little.
Fabricated Citation Record
Does a public court record exist involving output from this product?
No court order, opinion or disciplinary record naming this product has been located as of the date shown. This is a statement about the public record, not a finding about the product.
Searched the AI Hallucination Cases database maintained by Damien Charlotin, and reporting drawing on it, on 2 September 2026 on the product and corporate name Ankar and Ankar AI Ltd. No court order, opinion or disciplinary record naming the product was located. This is a statement about the public record rather than a finding about the product. One structural caveat bounds it and cuts differently from most records in this index. The tracker records fabricated citations in court filings, and the primary output here is a patent application or an office action response filed at a patent office rather than a brief filed in court, so the forum that would catch an equivalent failure is the USPTO or EPO and its responses are examiner objections and practitioner discipline rather than published sanctions orders. A published third-party practitioner assessment did identify non-compliant claim numbering in generated drafts, which is a compliance defect rather than a fabricated citation and is not recorded on this signal.
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 obligations are addressed directly and no authority or instrument is named. Clause 7.2 of the terms is explicit that the customer should not rely on output for legal advice, that output is not a substitute for a qualified professional, and that Ankar is not responsible for the professional obligations of the customer or its authorised users, and it names the specific work products the customer must review and approve before filing. That is a substantive engagement with the duty, and more than most records in this corpus offer. What is absent is the source. No patent office rule is cited, with neither the USPTO Rules of Professional Conduct nor its guidance on the use of artificial intelligence in filings named, nothing from the EPO or the UK IPO appears, and no bar or professional body material is referenced. Nothing states what a practitioner must do to satisfy a specific duty when using the platform, as distinct from stating that the duty remains theirs. The company blog does report on revised USPTO guidance as legal news rather than mapping the product to it.
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.
Time savings are published with figures and nothing addresses the bill. The company states that customers report a 40 to 60 per cent reduction in time spent on drafting and office actions, and frames the benefit as more time with inventors, faster filings and greater team capacity. Nothing addresses how AI-assisted work is recorded, billed or disclosed to a client, and no per matter record of AI-assisted work was located. The gap has a specific shape in patent practice, where a substantial share of prosecution work is billed at fixed or capped fees per application rather than hourly, so the question a client would ask is less about compressed hours than about whether a fixed fee negotiated for attorney drafting still reflects the work performed. Nothing published engages that. The platform does carry audit functionality according to the data processing addendum's processing description, but it is presented as access and activity logging rather than as a billing record.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
The material exists behind a sales conversation or an executed agreement.
The material exists and is not published. Clause 6(a) of the data processing addendum states that Ankar's current subprocessors are set out in Schedule 2, described as the Subprocessor List, and the addendum as published on the website ends after Schedule 1, so the list is referenced as an existing artifact a contracting customer receives rather than one a prospect can read. The governance around it is strong and is published: at least ten days prior written notice of any addition or replacement, a customer right to object on reasonable data protection grounds within ten days, written agreements imposing materially similar obligations on each subprocessor, and Ankar remaining liable for their performance. Audit rights at clause 5 offer SOC 2 Type II and ISO 27001 reports or summaries. AWS is named in the international transfers clause, but that is infrastructure and does not tell a firm whose model saw its client's unfiled invention. No model provider is named anywhere on any surface, so a firm answering a client's AI clause today would have to request the schedule rather than forward a published page.
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.
Checked the terms of service, the data processing addendum, the trust centre and the product navigation on 2 September 2026. Nothing addresses disclosure of AI use to a patent office or tribunal, and no exportable per document record of which model produced which passage, what was retrieved and who reviewed it is described. The product does hold most of the raw material: Traceable Citations is a named module, prior art is returned with source links and structured context, the processing description in the data processing addendum names audit functionality, and clause 7.2 of the terms creates a contractual review-and-approve step that would be the natural thing to evidence. None of it is presented as a disclosure record. The question is live in this practice area, since patent offices have begun addressing the use of artificial intelligence in prepared filings and the duty of candour attaches to what is submitted, and a platform that already logs activity and traces citations is closer than most to being able to emit exactly this record.