Amplified
Amplified is an AI patent search and research platform that ranks a global patent corpus by conceptual similarity rather than by keyword match. A user starts a project by describing an invention in plain language or entering a patent number, and the system sorts the whole database against that input in real time; results are then narrowed with keyword filters, field selection across title, abstract, claims and description, wildcards, proximity operators and full Boolean queries, which run alongside the AI rather than instead of it. The search is iterative by design: the researcher marks results relevant, flagged or hidden, and asks the system to learn from what was marked and surface more like it. Two AI ranking modes are offered, a semantic mode and a neural concept-similarity mode, and the platform publishes a principle that matters for defensibility, that the AI will sort results but never exclude them. Keyword counts are exposed so a searcher can see where the model has misread an input. Projects retain what has been read and saved, support annotation and notes, export a search history, and can be shared with unlimited viewers alongside paid creator licences. Coverage runs to more than 140 million records across 97 or more authorities and 22 languages, with AI search native in English and Japanese and further languages in beta. Pricing starts at 500 US dollars a month paid annually, with team and enterprise terms quoted. The terms of use carry an unusually precise position on model training: Amplified will never use customer content to train general AI models, and will train customer-specific models only at the customer's direction. Amplified AI, Inc. runs a Vanta-hosted trust centre and states an information security programme aligned with ISO 27001, built on Fly.io and Amazon Web Services.
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 ranking model is the product. A search begins by describing an invention in plain language or entering a patent number, and the system sorts more than 140 million records against that input in real time by document-level conceptual similarity, in one of two published AI ranking modes. Relevance feedback is the core loop: marked-relevant results are fed back so the system surfaces more like them, and machine learning can semi-automate relevancy review of new publications. Boolean search, proximity operators and field selection are all supported, but as instruments for steering the ranking rather than as the mechanism being bought; the vendor's own framing is that its AI reads the corpus so the user does not have to. Strip the model out and what remains is a patent database with a query box, which is the incumbent product this one is sold against. Checked 4 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 and structural rather than asserted, and no current measurement is published. The product returns records from a defined corpus rather than generating text, so there is no fabrication surface in the ordinary sense, and one published design commitment goes directly to defensibility: the AI will sort results but never exclude them, so a ranking error cannot silently remove art from a searcher's reach. Keyword counts are exposed specifically so a user can identify where the model has misread an input, which is an interpretability affordance most products do not offer, and search history can be exported. Failure is acknowledged in the agreement rather than only in marketing, with section 19 stating outputs may be inaccurate or incomplete and section 14 stating patent legal status is displayed as received from authorities without warranty. What is missing is a current, testable figure. The vendor's blog records a 2019 EPOPIC study reporting automated prior art results comparable to professional search across nearly 1.5 million patent search reports, which is a real evaluation but is seven years old, was read only as a search snippet rather than opened, and has no published republication against the current product. Recorded as a constraint on any future testing: clause 4.2(xxi) prohibits benchmarking or comparing the Service for publication without prior written consent.
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.
A written commitment that the professional stays in control, with real review surfaces behind it and no threshold structure. The strongest element is architectural: the AI sorts but never excludes, so the human decides what to stop reading rather than the model deciding what the human never sees. The working loop documented in the help centre is explicitly human-driven, with the searcher marking results relevant, flagged or hidden, iterating, and running a final quality pass with filters removed to catch anything missed. Keyword counts let the user diagnose the model's misunderstanding. The agreement adds a supervision requirement rather than a suggestion: clause 4.2(xix) prohibits using outputs in or as the basis for legal, medical, financial, safety-critical or other regulated decisions without independent human review by a qualified professional, and 4.2(xx) prohibits relying on outputs in place of the customer's own due diligence. What is absent is the rest of the control structure: no confidence signal is published, nothing states a threshold at which the system defers, and nothing addresses what happens after a search is later shown to have missed art.
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.
On the vendor's own surfaces an unattributed testimonial stands in for evidence. The home page carries a quotation praising the product with no name, role or organisation attached, and the blog records conference activity rather than deployments: an EPO keynote at Search Matters 2019, an EPOPIC 2019 presentation, an invitation to speak at LESI, and selection among five of ninety startups at NIKKEI's AI SUM. Those establish standing in the field, not outcomes at a customer. Named customers do exist in the wider record, with Ricoh, Unicharm, Mizkan Holdings and a European automotive manufacturer reported, and figures of more than fifty large enterprises and use in thirteen countries; **all of that is third-party, from LOT Network, CB Insights, WIPO and a revenue database, and none of it is evidence of capability under the ground rules**, so it is recorded here as context and credited to nothing. No case study, no named customer and no dated deployment figure was located on any first-party surface read.
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.
Substantive contractual commitments across most limbs, with the two gaps that hold it below the top. Training is the strongest element and is drawn more precisely than anywhere else in this corpus: clause 6.3 states Amplified will never use customer content to train general AI models, and permits training customer-specific models or classifiers only at the customer's direction and only to provide those features back to that customer. The content licence is purpose-limited and time-limited on its face, running only for the duration of use and ending when content is deleted from active systems. Retention and deletion are specific under 6.5, 12.9 and 16, with deletion from active systems within thirty days of termination or request, written confirmation on request, and a thirty-day export window. A mutual confidentiality clause at section 13 covers information reasonably understood to be confidential. Access control is contractual under 12.4 with personal, non-shareable credentials and an account administrator model. Two limbs are absent: nothing addresses privilege or attorney work product, which matters because projects are documented as holding invention disclosures and internal ideas, and no model provider is identified anywhere, so what any underlying provider may retain is unstated.
UPL and Professional Responsibility Posture
Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point.
A real published position that is specific to what this product actually does, short of the full treatment. Section 14 is headed No Professional Advice and states that professional information provided through the Service is for informational purposes only and not advice, and then does something more useful than a generic disclaimer: it names the concrete failure case for a patent tool, recording that patent legal status is displayed as received from authorities without warranty and directing the reader to the relevant national patent authority for authoritative status. The supervision dimension is handled in the acceptable use policy rather than left implicit, with clause 4.2(xix) prohibiting use of outputs in or as the basis for regulated legal decisions without independent human review by a qualified professional. Eligibility is limited to those acting in a business or professional capacity rather than as consumers, and to adults. What is missing from the top band is a stated jurisdiction limit for the product's own coverage: the direction to consult a national authority addresses legal status specifically, and the coverage page setting out which authorities are included was not opened in this pass.
AI Governance and Bias Disclosure
Published governance over model behaviour: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
No governance position is published for a system that decides which prior art a professional sees. There is no responsible AI statement, no governance framework, no named owner accountable for model behaviour, no description of an evaluation regime for ranking quality, no published testing results and nothing on bias, which has a specific shape here: a similarity model trained predominantly on English and Japanese material and offering native AI search in those two languages may perform unevenly across the 22 languages and 97 authorities in the corpus, and nothing addresses that. The information security programme described as aligned with ISO 27001 answers a security question rather than a governance one. One published clause runs against external scrutiny rather than toward it: clause 4.2(xxi) prohibits benchmarking, testing or comparing the Service for publication without prior written consent. Searched the home page, how it works, pricing, terms and the trust centre on 4 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.
The full set is published, specific, and sits in the agreement rather than on a policy page. Retention and deletion: clause 12.9 commits to deletion from active systems within thirty days of termination or request with written confirmation available, 6.5 adds deletion from backups within the regular backup cycle, and section 16 gives a thirty-day export window before deletion. Access control: 12.4 requires access controls, encryption in transit and at rest, logging and monitoring, vulnerability management and secure development practices, with confidentiality obligations on authorised personnel, and section 3 makes credentials personal and non-shareable under a named account administrator. Incident practice: 12.8 commits to notification without undue delay with the content of the notice specified, covering the nature of the incident, categories and approximate volume affected, likely consequences and measures taken. Subprocessors: 12.5 commits to publishing a current list at the trust centre with subscribable email updates of changes and flow-down obligations no less protective. International transfers are addressed at 12.6 with named mechanisms, audits at 12.10, and US state privacy service-provider terms at 12.12. One limit is named rather than glossed: the trust centre is Vanta-hosted and returned metadata with no body, so the subprocessor list itself was not read and is credited as a published commitment rather than for its contents.
AI Liability and Recourse
What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.
Liability is addressed only through a standard limitation clause that disclaims the exposure the product creates. Section 18 provides the Service as is and as available with all warranties disclaimed including fitness and non-infringement. Section 19 caps aggregate liability at fees paid in the twelve months before the event, excludes all indirect and consequential damages, and then addresses the product risk head on in a way that removes rather than allocates it: the customer acknowledges outputs may be inaccurate, incomplete or inappropriate, is solely responsible for verifying them, and Amplified disclaims all liability arising from reliance on outputs. The indemnity at section 17 runs in one direction only, from customer to Amplified, with no intellectual property or other indemnity given to the customer, which is a notable omission for a product whose entire function is telling a buyer what prior art exists. No insurance position was located. What keeps this off the floor is that the allocation is published, precise and readable before signing, with carve-outs preserved for death, personal injury, fraud and wilful misconduct.
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 documented interchange with an external research tool, and no practice system named. The help centre documents importing a result list from a complementary search tool, naming CAS and STN, which is a real and specific interoperability path for a professional searcher and is documented well enough to act on. Beyond that the site inventory is thin: the navigation carries How it works, Blog, Pricing and Resources, and there is no integrations page, API reference or developer documentation, although a public GitHub organisation is linked from the footer and was not opened. No IP management system, docketing platform, document management system or patent office filing interface is named on any first-party surface, which is the integration a firm using this alongside an IPMS would ask about first. Single sign-on providers are reported in a third-party software directory but appear nowhere in first-party material and are not credited. Export is well served, with search history export and sharing to unlimited viewers, but export is not integration.
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.
Cloud delivery is explicit and neither the tenancy model nor the region is stated. The trust centre names the infrastructure directly, recording that Amplified builds on Fly.io and Amazon Web Services and relies on their compliance with privacy and information security standards including recurring re-examination by independent auditors, with data encrypted in flight and at rest. Naming the providers is not naming a location: no region, country or data centre appears anywhere, no residency option is offered at any tier, and nothing states whether the platform is single or multi-tenant or whether any dedicated arrangement exists for enterprise customers. The one place the question is engaged is at one remove, in clause 12.6, which addresses international transfers of personal data by committing to a lawful transfer mechanism and naming EU Standard Contractual Clauses Module Two, the UK Addendum and IDTA, a Swiss addendum and Brazilian SCCs; that establishes that cross-border transfer happens without saying between where and where.
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.
A real trust portal exists and no certification of Amplified's own is claimed anywhere. The wording is consistent across two surfaces and is careful: clause 12.4 states an information security programme aligned with ISO 27001 industry standards, and the trust centre states adherence to ISO compliance standards. Alignment is a self-assessment; no certifying body, scope, observation period or report date is published, and on the third-party verifiability test a buyer cannot check anything against a register. The certifications that are named belong to the infrastructure: the trust centre credits Fly.io's and Amazon Web Services' compliance and their recurring re-examination by independent auditors, which is the host's evidence and does not travel to the platform. What lifts this off the floor is a genuine access route rather than a badge: a dedicated trust centre is published at trust.amplified.ai, clause 12.4 commits to making security summaries and independent assessment information available there, and 12.10 commits to providing audit summaries and security documentation on written request. The portal is Vanta-hosted and returned metadata with no body on fetch, so its contents and access tier could not be established and the lower tier is assumed for that reason.
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.
Proprietary models are implied throughout and nothing underneath them is identified. The training clause distinguishes general AI models from customer-specific models and classifiers that Amplified itself trains, and the product describes two AI ranking modes, a semantic mode and a neural concept-similarity mode, built on deep learning over the patent corpus. Taken together those establish that Amplified operates its own models rather than reselling a general assistant, which is more than most records in this band show. What is absent is everything a buyer would need to verify it: no model is named, no version or architecture is given, no third-party model or inference provider is identified, no location is stated for where inference runs, and no commitment to notify customers when a model changes was located. Fly.io and Amazon Web Services are named at the trust centre as infrastructure, which says where the platform runs and not whose models process a search. The subprocessor list that might resolve this is published at the trust centre under clause 12.5 but the portal returned no body on fetch.
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.
Real pricing is published for the entry point with the upper range withheld. The rate is stated as from 500 US dollars a month when paid annually, which is a floor rather than a rate, and what it buys is itemised: unlimited projects, unlimited viewers, search across more than 140 million patents, sharing of up to 50 projects, discussion and annotation, and cancellation at any time. The use cases the tiers are meant to serve are named as novelty, invalidity and freedom to operate, and both AI ranking modes are included rather than gated. Purchase is self-serve, with signup and a trial request route published and no sales conversation required to begin. Team pricing and enterprise options are quoted rather than published, which is what holds this below the top band, and clause 9.1 points to a second pricing page inside the application that was not opened. The agreement adds the commercial terms a buyer needs: subscriptions auto-renew for the then-current term unless cancelled in account settings, all payments are non-refundable except where law requires, and fees may change prospectively on notice.
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 is documented with real precision on the corpus and the use cases, and left open on the buyer. The searchable scope is quantified rather than claimed: more than 140 million records, 97 or more authorities, 22 languages, with roughly 34 million original-language publications and 50 million machine translations in the English index, family data and backward and forward citations. The work the product is sold for is named specifically as novelty, invalidity and freedom to operate, which is the vocabulary of the practice rather than marketing. Buyer types are identified across IP law firms, corporate research and development and IP teams, patent offices, universities and individual inventors. One limit is stated plainly and is the kind most vendors omit: AI search is native in English and Japanese only, with further languages in beta and available on request, so a buyer working primarily in another language is told what they are getting. What is missing is the firm dimension: no firm size or seat threshold is addressed, in-house and government use are named as segments without being described, and the full coverage page was not opened in this pass.
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?
Training occurs only where the customer has affirmatively enabled it.
The clearest drafting on this signal located anywhere in the corpus, because it separates two things most vendors collapse. Clause 6.3 of the terms of use states that Amplified will never use customer content to train general AI models, an unqualified contractual prohibition, and then separately provides that customer-specific models or classifiers may be trained on customer content at the customer's direction, to deliver those features back to that customer and its permitted users. Training therefore happens only where the customer has affirmatively enabled it, which is what this value records, and the general-model route is closed contractually rather than by policy. The surrounding licence is consistent: it is purpose-limited to providing, maintaining, securing, troubleshooting and improving the Service, and time-limited to the duration of use, ending when content is deleted from active systems. Clause 7 keeps Usage Data separate and expressly excludes User Content from it.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
The customer controls the retention window, by product configuration or by contractual instruction, but zero retention is not stated as available.
The customer controls the window by contractual instruction and a default is published. Clause 12.9 commits to deletion or return of customer personal data from active systems within thirty days of termination **or request**, with written confirmation that deletion has been carried out available on request, and backups cycling out on standard retention schedules. Clause 6.5 states the same for user content generally, deleting from active systems on the customer's deletion or on termination and then from backups within the regular backup cycle, with limited retention permitted for law, legal process, audit, security and integrity. Section 16 adds a thirty-day electronic retrieval window after termination before deletion. Zero retention is not offered as a setting, and the qualifiers around lawful retention and backup cycles are recorded rather than glossed.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
The product maintains its own permission model, documented, requiring the firm to keep it aligned.
Amplified operates its own permission model and describes it at feature and contract level. Work is organised into projects with sharing controlled by the customer: clause 6.1 places sharing settings with the customer, 6.4 makes sharing an explicit per-recipient licence, and the product adds unlimited viewers alongside purchased creator licences so read access and authoring rights are separated. Section 3 requires a named account administrator with authority over seats and settings, makes credentials personal and non-shareable, and prohibits sharing logins to give others access to paid features, with two-factor authentication encouraged and capable of being required. So the model is Amplified's own and the customer administers it, which is the alignment burden this value describes. Nothing published addresses separation between individual matters inside one customer account, which for a patent search tool is the question of whether one client's freedom-to-operate project is walled from another's.
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.
Section 13 of the terms provides that a recipient of confidential information may disclose it where required by law, but only with prompt notice to the disclosing party where lawful and with cooperation to limit the scope of disclosure. The clause is mutual, so it binds Amplified in respect of the customer's confidential information, and the where-lawful qualifier is the standard and appropriate carve-out for gag orders rather than a general escape. Confidential information is defined broadly enough to reach the customer's research: non-public business and technical information that should reasonably be understood as confidential, which covers invention disclosures and search strategy held in projects. No transparency report of government or third-party requests was located on any surface read, which is what separates this from the top value.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
Coverage is described by jurisdiction with no identification of the underlying corpus.
The corpus is described by coverage rather than by identified source or licence basis. Published figures are specific: more than 140 million records, over 120 million full-text records, 97 or more patent authorities, 22 languages, with roughly 34 million original-language publications and 50 million machine translations in the English index, plus family data and backward and forward citations, and a stated weekly update cadence. That tells a searcher what jurisdictions and record types are in scope, which is the coverage question. What is not stated anywhere is where the data comes from or on what terms: no data supplier, aggregator or patent office feed is named, no licensing arrangement is described, and nothing addresses the scientific literature the marketing separately refers to. Coverage by authority is a jurisdictional description of the collection rather than an identification of it.
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.
The patent analogue of this question is legal status, and Amplified addresses it directly and negatively. Section 14 states that patent legal status is displayed as received from the authorities without warranty, and directs the reader to the relevant national patent authority for authoritative status. That is an express disclaimer of currency rather than a treatment signal: the product passes through what the registers report and does not verify, annotate or flag whether a patent remains in force. Some currency tooling exists alongside it, with alerts configurable on individual patents to notify on changes in legal status, which monitors change going forward rather than validating status at the point of search. No verification prompt, no treatment indicator and no own-status signal is published.
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.
No abstention path is described, and the product's design makes the question take an unusual shape that is worth recording rather than forcing. The system ranks rather than answers, and its published commitment is that the AI will sort but never exclude results, so there is no point at which it declines to return something. Two related affordances exist without being uncertainty signals: keyword counts are exposed so a searcher can see where the model has misunderstood an input, and the help documentation instructs users to read noisy results as evidence that their description was too vague and to correct the input. Neither is a confidence or grounding score surfaced against a result, and nothing describes what the system does when it cannot rank meaningfully. Searched the home page, how it works, pricing, terms, trust centre and the help centre on 4 September 2026.
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.
The AI Hallucination Cases database maintained by Damien Charlotin was searched on 4 September 2026 on both the product name Amplified and the corporate name Amplified AI. No court order, opinion or disciplinary record naming the product was located. The database tracks fabricated legal citations in court filings, and this product returns records from a patent corpus rather than generating citations, so the exposure is structurally low. This records the state of the public record on that date and is not a finding about the product.
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.
Public materials refer to professional responsibility in general terms without naming guidance. Section 14 states that professional information provided through the Service is informational only and not advice and directs users to seek a qualified professional, and clause 4.2(xix) prohibits using outputs in or as the basis for regulated legal decisions without independent human review by a qualified professional. Both engage the professional's duty as a subject. No bar association, patent bar, regulator, rule of professional conduct or ethics opinion is named anywhere, and nothing addresses the duty of candour to a patent office, which is the professional obligation most directly engaged by a prior art search tool. Searched the home page, how it works, pricing, terms, trust centre and help centre on 4 September 2026.
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.
Public materials claim time savings without addressing billing. The headline efficiency claim is that IP professionals eliminate 50 to 80 per cent of the time previously spent searching, and the marketing frames prior art as findable in an hour or less without search experience. Nothing addresses what happens to a client bill or a fixed search fee when that time compresses, and no per-matter record of AI-assisted work is described as available for fee or disclosure purposes. One adjacent capability exists and does not meet the signal: search history can be exported, which evidences what was searched rather than recording AI-assisted work for billing. The vendor's own charging model is per creator licence per month rather than per search, which is a cost fact rather than an answer.
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 mechanism is committed contractually and the artifacts could not be read. Clause 12.5 commits Amplified to publishing a current subprocessor list at its trust centre, with customers able to subscribe there for email updates when it changes, and to imposing data protection obligations on subprocessors no less protective than the DPA. Clause 12.4 commits to making security summaries and independent assessment information available through the same portal, and 12.10 to providing audit summaries and security documentation on written request. A published DPA is embedded in the terms themselves at section 12, which is a forwardable client-facing artifact reachable without executing anything. What could not be established: the trust centre at trust.amplified.ai is Vanta-hosted and returned metadata with no body, so neither the subprocessor list nor the assessment material was read, and no model provider is identified in any document that did render. Under the gated-is-not-absent rule the lower tier is assumed and the reason is stated.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
Some elements of the record are available, short of a document level export.
A genuine exportable record of the search exists, covering sources and process but not the model. Projects retain every patent read, marked relevant, flagged or hidden, and a search history can be exported covering both AI and Boolean searching, which gives a defensible account of what was looked at and in what order. The published commitment that the AI sorts but never excludes strengthens that record, because it means the export reflects a ranking rather than a silent filter. Annotation and notes are retained per project. What is absent is the model dimension and the verification step: nothing states that the model or ranking mode used for a given result set is recorded in the export, no version is captured, and no human sign-off is documented as recorded. No disclosure guidance or template for a patent office or court was located, and nothing addresses the duty of candour context in which such a record would be produced.