&AI vs Patlytics: how they compare in 2026
&AI and Patlytics both sell AI to patent litigators for prior art search, infringement and invalidity claim charts, and contentions. Patlytics sits in the top two bands on ten of fifteen axes and &AI on eight of fifteen, identical on four. Patlytics leads on what it publishes about security and professional duty. It holds SOC 2 Type 2, ISO 27001 and ISO 42001, names NCC Group and A-LIGN as penetration testers, and keeps user history for 90 days with deletion. Its practitioner guidance ties Model Rule 1.6 and prior disclosure bars to the choice of AI tools. &AI's lead is control over what leaves the platform and what it costs. Every citation in a claim chart must be approved by the user before export, and exported charts contain only exact quotations. It publishes credits at $0.40 and a Pro plan at $625 a month. Its terms grant it a perpetual license to use uploaded content and share it with other users, while its security page says it never trains on customer data.
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 models are the product, which is the A band. Every capability the vendor sells is model work: agentic prior art search across patents, non-patent literature and product materials, claim-chart evidence proposed element by element with a relevance rationale, claim constructions generated from the patent and its history, drafted contentions and expert reports, structured analysis across thousands of documents, and the assistant Andy working across all of it. The pricing page says credits pay for the AI and compute-intensive actions, which is every action a customer buys. Take the models away and what is left is a document store and an export formatter. Verified 18 September 2026.
AI native by construction and by the vendor's own framing, with a proprietary corpus underneath. Models are stated to be fine tuned on more than 50 million global public patents, which is model work on a domain corpus rather than a general assistant pointed at patent documents, and the vendor positions the platform explicitly against point tools. Every capability is model output: claim element decomposition, instances and evidence of use surfacing from large data sets, automatic claim chart generation mapping evidence to claim limitations, technology area classification and grouping ahead of portfolio analysis, read strength aggregation into High Medium Low risk scores, and application drafting. Remove the models and there is no product. First A on this axis outside the plaintiff category since Descrybe, and it opens ip-and-patents at the same level.
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 documented to the passage, with one internal measurement short of a published evaluation, which is the B band. The claim charts page says every citation points to a specific passage, figure or timestamp in the source and that exported charts contain only exact quotations and references, and the Charts post says each proposed piece of evidence carries a short relevance rationale. The same post reports that in an internal review of exported charts about 83 percent of the evidence in the final export was present in the first pass, for both invalidity and evidence-of-use charts. That is a real figure, but the test set is described only as exported charts, the measure is closer to recall than to accuracy, no failure mode is named, and nothing states what the system does when no support is found. Drafted contentions and reports, which do contain generated text, carry no published measure. The terms say AI output can be incorrect and place responsibility for it on the user. Verified 18 September 2026.
Three distinct grounding mechanisms are named and the hallucination question is addressed directly, short of any published measurement. The vendor states citation backed outputs, auditing of sources for relevance, and color coded confidence indicators, and describes its models as designed to safeguard against hallucinations, which is naming the failure mode rather than avoiding it. Color coded confidence is the notable element: it surfaces the model's own certainty to the user at the point of reading rather than leaving a confident wrong answer indistinguishable from a confident right one, and almost nothing on this index does that. The claim charting architecture reinforces it structurally, since a chart maps a specific claim limitation to a specific piece of product evidence and is falsifiable on its face. Held at B because nothing is quantified: no accuracy figure, no precision or recall on evidence of use detection, no false positive rate on infringement scoring, no evaluation and no benchmark. A vendor with ISO 42001 certification has an AI management system and still publishes no measured output from 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.
A named mode carries a categorical constraint and a mandatory review gate, which meets the A band on the categorical-constraint reading. For claim charts, the product page states that every citation requires the user's approval before it appears in any exported work product and that no generated content appears in any part of an exported chart, from citations to boilerplate: only exact quotations and references. That tells a buyer before purchase what the output of that mode can and cannot contain, and puts a human decision at a defined point before anything leaves the platform. The product separates modes clearly (search, charts, drafts, tables, pitch and the Andy assistant), and the vendor's own statement of purpose is that the AI handles volume while lawyers make the judgments. The constraint is specific to Charts: drafts and Andy produce generated text under a general expectation of review with no equivalent gate, which is recorded so the grade is read correctly. Verified 18 September 2026.
Oversight is designed into how results are presented rather than asserted as a principle. Published material states that AI automates the repetitive and time intensive parts of infringement analysis while preserving human control over legal judgment, and the product design supports that: risk scores are surfaced as High, Medium or Low for triage rather than as conclusions, color coded confidence indicators travel with the output, sources are audited for relevance and shown, and claim charts present evidence mapped to limitations for an attorney to accept or reject. Portfolio Heatmaps are framed as narrowing a broad portfolio to a smaller set of actionable targets, which is triage feeding human judgment. Held at B because no boundary is published: nothing states what the system does unattended, whether any output can be relied on without review, what the confidence bands actually mean numerically, or what threshold separates a High from a Medium. Color coding a confidence level without publishing what the colors represent is a signal without a scale.
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.
Figures are published without a named customer, which is the B band. The vendor's February 2025 release says early customers at large law firms used the platform to build thousands of claim charts and reported average time savings of 70 to 90 percent, and the product pages say its charts are used in active IPR and district court proceedings. The pricing page publishes average credits per claim chart and per search over the 90 days to July 2026, which evidences production use at volume. No customer is named anywhere; the named practitioners on the home page are an advisory board, not customers, and are not counted here. Verified 18 September 2026.
Named penetration, a named practitioner at a named firm, and dated funding, all independently checkable. Adoption is stated as more than 40 percent of Am Law 100 IP practices, which is a specific and falsifiable market penetration claim rather than a logo wall, alongside Fortune 500 companies and firms ranked by Chambers, IAM Patent 1000 and Legal 500. Attributed comment: Yury Kapgan, Partner and Co-Chair of the Artificial Intelligence Practice Group at Quinn Emanuel Urquhart and Sullivan, quoted by name and title in the dated 8 April 2026 Series B announcement describing the specific workflows the platform covers. A reader can identify the firm, the partner and the date without contacting the vendor. Funding is dated and specific: approximately $65m total, a $14m Series A led by Next47 in February 2025 and a $40m Series B led by SignalFire on 8 April 2026, with Relativity among named investors. Customer impact figures are published as 80 percent reduction in project time, more than $30,000 saved per claim chart, and up to 15 hours recovered per patent application. Those carry no methodology or baseline and the note records that; the grade rests on the named and checkable elements rather than on the percentages.
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.
The confidentiality commitments live in marketing while the published agreement grants the opposite, which places this at C and the gap is the finding. The home and security pages say data stays private and under the customer's control, that &AI and its subprocessors never train on customer or client data, and that zero-data-retention agreements are in place with model providers. The Terms of Service, effective 18 June 2026, grant &AI a non-exclusive, perpetual, irrevocable, worldwide, sub-licensable and transferable license to reproduce, modify, prepare derivative works of and otherwise use content users upload, and state that &AI may make that content available to other users of the platform. The privacy policy lists artificial intelligence engine providers among the service providers that may receive information and contains no training commitment in its body. Nothing addresses privilege or work product, segregation between matters, or retention of uploaded case material. Verified 18 September 2026.
The second record in 48 to engage professional confidentiality directly, after Eve, and the first to engage the specific rule that governs it. The vendor publishes practitioner facing material built around Model Rule 1.6 on attorney confidentiality, USPTO Rules of Professional Conduct, 37 CFR and 35 U.S.C. section 102 on prior disclosure bars, and states the point that makes this category different from every other on the index: an unpublished patent application exposed to the wrong tool does not merely create a compliance problem, it can permanently extinguish the client's patent rights. Naming that consequence correctly is a demonstration of understanding rather than a marketing claim. The controls behind it are specific: customer data siloed and segregated between customers, encryption in transit and at rest, least privilege access, user history retained 90 days and deletable, and a stated commitment that customer data is never used to train models. Graded A because it is strictly more than the B records elsewhere on the index, which carry comparable controls with no engagement with the professional obligation at all. Held short of a perfect record because the confidentiality controls are asserted rather than documented and no privilege specific attestation exists.
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 general output disclaimer sits in the terms while the audience extends beyond lawyers, which is the C band. The terms state that AI sometimes produces incorrect or inapplicable outcomes and that the user is solely responsible for the accuracy and use of output, and self-serve accounts are limited to verified patent practitioners and other approved users. Those approved users expressly include investment funds and universities, and the product is marketed for trial-ready analysis and invalidity arguments. No statement says the output is not legal advice, nothing addresses competence or supervision duties when relying on a chart or contention, and no jurisdiction limits are stated. Verified 18 September 2026.
The most specific professional responsibility material located in the pull. The vendor publishes a due diligence checklist written explicitly for patent practitioners rather than IT directors, naming USPTO Rules of Professional Conduct, 37 CFR, Model Rule 1.6, 35 U.S.C. section 102 prior disclosure bars, and duty of competence obligations under state bar rules, and setting them against enterprise security concepts including tenant isolation, model training on confidential inputs, vendor employee access to stored work product and breach notification. Naming the specific rules and the specific technical threat vectors together, and explaining why the stakes differ for patent work, is a genuine professional responsibility position and it exceeds ProPlaintiff's ABA Model Rules reference in specificity. Held at B rather than A on the same basis applied to ProPlaintiff: it is published as market education in a blog library rather than as a product commitment or a contractual term, nothing binds the vendor to it, and it addresses how a practitioner should evaluate tools rather than what this vendor undertakes.
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.
No governance position published. Searched the home, security, pricing, terms and privacy pages, the product pages and the blog index on 18 September 2026. No responsible AI policy, testing regime, accountable owner or discussion of uneven output was located. The vendor's post on its name sets out a view that AI handles volume and lawyers handle judgment, and the home page lists an advisory board of patent litigators; neither is a governance framework for model behavior. Verified 18 September 2026.
FIRST ISO 42001 CERTIFICATION ON THE INDEX, across 48 records. ISO/IEC 42001 is the artificial intelligence management system standard, and unlike a security certification it covers governance of the AI itself: risk management for AI systems, defined roles and accountability, impact assessment, and lifecycle controls, independently audited. Every record graded on this axis before now was assessed against published policies, model cards or evaluations, and every plaintiff category record graded D. This vendor holds a third party audited AI governance management system. Supporting disclosures are consistent with it: named hallucination safeguards, source auditing, confidence indicators, stated model training boundaries and named penetration testing partners. Held at B rather than A because certification of a management system is evidence that governance processes exist and were audited, not evidence of what they found: no bias or fairness testing result, no model card, no evaluation output, no accuracy monitoring figures, no drift statement and no named internal governance body were located, and no certificate date, auditor or scope is published for the ISO 42001 itself.
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.
Access control and model-provider retention are addressed while retention, deletion and incident practice for the vendor's own storage are not, which places this at C. The security page states encryption in transit and at rest, single sign-on and role-based access control aligned to least privilege, and zero-data-retention agreements with model providers so inputs are not retained by them. It says the trust center lists subprocessors, but that page returns no content to automated retrieval and could not be read on this date. The privacy policy states only that personal information is kept as long as necessary, says nothing about uploaded case material, and gives no deletion commitment or incident notification; the terms' perpetual content license survives termination. Verified 18 September 2026.
A strong and specific stewardship position, undermined by the vendor contradicting itself on the same page. Published across product pages and repeated: customer data is encrypted in transit and at rest, segregated, and never used to train models, stated as guaranteed. Supporting specifics that few records match: user history stored for 90 days and deletable, customer data siloed between customers, least privilege access, 24/7 monitoring with named on call coverage, and models fine tuned on public patent data rather than customer content. THE CONTRADICTION: the same security page states that information is not utilized for training AI models without explicit consent or necessity. Never and not without consent or necessity are not the same commitment, and necessity is undefined and determined by the vendor. Held at B rather than A on that basis. The signal row records the contradiction with both formulations, and the note here flags it as the single thing a buyer should ask about in diligence.
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 that disclaims the exposure the product creates, which is the C band. The terms provide the service and all output as is, with no warranty that output is free of errors, state that the user is solely responsible for the accuracy, quality and usability of output and that &AI has no liability for it, and cap aggregate liability at the greater of one hundred dollars or fees paid in the prior twelve months. The indemnity runs from the user to &AI. No vendor indemnity, output warranty or insurance position is published. Verified 18 September 2026.
No published position located. Nothing was found on liability for AI output, warranty, service levels or remedy. The exposure in this category is distinctive and severe: an invalidity search that misses prior art, a freedom to operate analysis that misses a blocking claim, or an infringement chart that maps evidence to the wrong limitation all carry consequences measured in litigation exposure or lost patent rights, and the vendor's own published material argues that patent work is uniquely high stakes because errors can permanently extinguish rights. Having made that argument, the vendor takes no public position on what happens when its own output is wrong. Checked the security page, the product pages, the blog library, the funding announcements and the site navigation on 29 Aug 2026. Enterprise agreements govern this and are not public.
Practice Systems Integration Depth
How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.
No integration into practice systems located. Searched the home, product, pricing, security and terms pages on 18 September 2026. The platform draws on patent and product data sources and exports charts with configurable formatting, and the Enterprise plan offers custom integrations, but no document management, Microsoft Word, docketing or matter management connection is named or documented. Verified 18 September 2026.
Nothing located. No IP docketing system, patent management platform, document management system or law firm practice system integration was named, no API is described, and no export or workflow connector was found. The gap is notable in this category specifically because IP practice runs on docketing systems tracking prosecution deadlines and annuities, and a platform covering application drafting and prosecution would ordinarily need to reach them. One connection exists at the corporate level and is deliberately not credited as an integration: Relativity appears as an investor in the Series B, which is a funding relationship rather than a product relationship, and nothing describes any technical connection between the two platforms. Checked the product pages, the security page, the funding announcements, the blog library and the site navigation on 29 Aug 2026. Strongest correction candidate on this record.
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.
Deployment options are stated with partial residency detail, which is the B band. The pricing page offers Enterprise customers single-tenant architecture or self-hosted models, and the terms and privacy policy state that information is transferred to and processed in the United States. No region choice is offered, the self-hosted option is not described further, and processing location for third-party model calls is not separated from storage. Verified 18 September 2026.
Nothing located about where data sits. No hosting provider is named, no region or data residency commitment is published, and no single tenant or dedicated instance option is described. Two published facts touch geography without answering the question and are recorded as not credited: 24/7 monitoring with on call rotation in North America and Asia Pacific describes where staff are, not where data is, and models fine tuned on global public patents describes corpus scope rather than infrastructure. Residency matters in this category because unpublished patent applications are subject to foreign filing license requirements and export controls in several jurisdictions, and a firm would need to know where an unpublished application is processed before uploading one. Checked the security page, the product pages and the site navigation on 29 Aug 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.
A named attestation with a stated request route, short of evidence a reader can see, which is the B band. The security page says &AI holds a SOC 2 Type II attestation and that the report can be requested through its trust center, which also lists certifications and subprocessors. That trust center returns no content to automated retrieval and could not be read on this date, so the auditor, report period and whether the request is self-serve could not be established and the top band is not asserted on it. Verified 18 September 2026.
Three certifications, two named testing partners, and a dedicated security page publishing all of it openly without a gate. Certifications: SOC 2 Type 2, ISO 27001, and ISO 42001, the last being the artificial intelligence management system standard and the first instance of it on this index. Penetration testing partners are named, NCC Group and A-LIGN, which is uncommon and materially better than the unnamed third party testing several records claim, since a reader can assess the testing house. Supporting controls are specific rather than generic: encryption in transit and at rest, customer data siloed and segregated, least privilege access, 24/7 monitoring with named regional on call coverage, and a stated 90 day user history retention with deletion. Published open rather than behind a request, which is the top tier under the three tier test. Held short of a perfect record on the familiar gap: no certificate dates, examination periods or certifying bodies are published for any of the three certifications, so currency cannot be established. Calibration: this sits with Lexis+ AI, Exterro and EvenUp at A, and is distinguished by breadth of framework coverage rather than by a publicly verifiable registry entry as Exterro has.
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.
Third-party model providers are acknowledged without being identified, which is the C band. The security page refers to zero-data-retention agreements with the vendor's model providers, the privacy policy lists artificial intelligence engine providers among its service providers, and the Enterprise plan offers self-hosted models. No model, provider or version is named on any readable surface, and no change notification is described; the terms reserve the right to change the services without notice. The subprocessor list in the trust center, which may name providers, could not be read on this date. Verified 18 September 2026.
Nothing located about the chain. The vendor states that its models are fine tuned on more than 50 million global public patents, which describes what the models were trained on and asserts ownership, and says nothing about what they were fine tuned from: no foundation model provider, model family or version is named, no subprocessor list is published, and nothing states whether any third party model processes customer content. Fine tuning necessarily implies a base model supplied by someone, so the disclosure confirms an upstream relationship exists while identifying no party in it. The gap sits awkwardly against ISO 42001 certification, since an AI management system standard addresses supply chain among its concerns, and against the vendor's own published checklist which tells practitioners to assess subprocessor risk. Checked the security page, the product pages, the blog library and the site navigation on 29 Aug 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.
A buyer can price the work before speaking to anyone, which is the A band. The pricing page publishes pay-as-you-go credits at $0.40 each, a Pro plan at $625 per user per month including 2,500 credits with further credits at $0.25, a 20 percent annual discount, an Opportunities add-on at $175 per user and $50 per non-user attorney, and custom Enterprise terms. It then converts the unit into work: over the 90 days to July 2026 an average claim chart used 241 credits and an average prior art search 176, priced at each rate, so $96.40 or $48.20 for a chart depending on plan. The terms add automatic renewal and a seven-day trial with 1,000 credits. What implementation adds is not stated, and Enterprise rates are withheld. A pricing row is written. Verified 18 September 2026.
No pricing published at any level. No price, no range, no tier structure and no unit of charge, and no indication of whether the platform prices per seat, per matter, per patent analyzed or per claim chart generated, which matters because the published value claim is denominated per claim chart at more than $30,000 saved and a buyer cannot set that against what a chart costs. Independent material describes enterprise subscriptions in this segment running to tens of thousands of dollars annually with cost depending on patents analyzed, frequency of use and integration level, and describes Patlytics as competitively priced, which is third party characterization rather than disclosure and is recorded as context. Every route is a demo request. Checked the product pages, the demo page, the pricing navigation and independent review material on 29 Aug 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.
Segments and workflows are described precisely with one limit stated, short of full boundaries, which is the B band. The vendor names law firms from the Am Law 100 to IP boutiques, plaintiff and defense counsel, and in-house teams for litigation preparation, infringement mining and transactions, and it covers invalidity under sections 101, 102, 103 and 112, evidence of use, IPR and district court work. Its comparison page states where it stops, saying prosecution and drafting-centred work is better served by a named competitor. Patent coverage is described as major jurisdictions without a list, and government use is not addressed. Verified 18 September 2026.
Workflow coverage is the most complete in this category and is enumerated rather than gestured at. The lifecycle spans invention harvesting and disclosure, application drafting, prosecution, prior art search, infringement detection, invalidity contentions, claim construction, claim charting, transactional due diligence, portfolio management and pruning, and the vendor positions this end to end scope explicitly against point tools. Corpus scale is stated at more than 50 million global public patents with regular verification and updates for data accuracy, which addresses currency, an element most records on this index omit entirely. Buyer coverage spans law firm IP practices and in house teams. Held at B rather than A because the corpus is not characterized where a practitioner would check it: no jurisdictional breakdown of which patent offices are covered, no statement of historical depth, no update lag figure despite currency being claimed, and nothing on whether non patent literature is included, which for prior art and invalidity work is a material boundary.
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?
The published Terms of Service grant an unbounded license over customer content that never names training, against marketing that says no training occurs. Section 5 grants &AI a non-exclusive, perpetual, irrevocable, worldwide, sub-licensable, transferable license to reproduce, distribute, prepare derivative works of, modify and otherwise use content users upload, and to make it available to other users; no purpose limit, deletion or opt out is attached.
The home and security pages state that &AI and its subprocessors never train on customer or client data under zero-data-retention agreements with model providers, and the privacy policy's page description refers to no-training commitments that its text does not contain.
Policy never, recorded against a contradiction on the vendor's own security page. The dominant and repeated commitment, appearing across product pages and on the security page itself, is that customer data is encrypted, segregated and never used to train models, stated as guaranteed. The quoted sentence appears on the same page and says something materially weaker: not utilized for training without explicit consent or necessity.
Never and not without consent or necessity are different commitments, and necessity is undefined and determined by the vendor rather than the customer. Recorded at policy never because the unqualified formulation is the dominant published claim and appears in more places, and because calling the record silent would erase a real commitment. Recorded as policy rather than contractual because both formulations sit on marketing and security pages rather than in terms or a data processing agreement.
This is the single item a buyer should put to the vendor in diligence, and the note exists so a later reader asks the question rather than reading only the guarantee. Third contradiction of this shape in the pull, after Jhana.ai and the scoped commitments on LinkSquares and Eve.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
Retention is acknowledged without a period for the vendor's own storage. The privacy policy keeps personal information as long as necessary for its purposes and does not address uploaded case material, prompts or outputs. The security page states that zero-data-retention agreements with model providers mean inputs are not retained by those providers. No customer retention setting, deletion commitment or post-termination period is published, and the terms' content license is perpetual.
Disclosed fixed, with a stated period and a deletion right, which is among the best answers on this signal in the pull. The vendor states that user history is stored for 90 days and can be deleted. A named number of days is a commitment a customer can hold the vendor to and a reader can compare, and pairing it with deletion means retention is both bounded by default and shortenable on request. Only Exterro, at zero for its AI layer, states retention more precisely.
Held at disclosed fixed rather than a configurable value because the 90 days is stated as the period rather than as a default a customer can vary, and because the scope is described as user history: nothing states whether uploaded patents, generated claim charts, detection reports and portfolio analyses are covered by the same period or retained separately, which for a platform holding unpublished applications is the question that matters most.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
Access segregation is asserted through single sign-on and role-based access control, with no published detail on matter-level walls. The security page says access follows least privilege under the customer's administrators; the product works by matter. Nothing documents how access is separated between matters or teams, and the terms state that &AI may make uploaded content available to other users of the platform.
Claimed and not documented, with more substance behind the claim than most records carry. The vendor states that customer data is siloed and segregated from each other to ensure privacy, and separately that a least privilege access approach is used, which together describe tenant isolation and internal access control rather than a generic assurance. The vendor's own practitioner checklist names cross matter data leakage and vendor employee access to stored work product as threat vectors that matter in legal software, so it has identified the right questions.
What is not documented is the answer for its own product: no description of how segregation is enforced, no statement of whether isolation operates at matter level within a single firm as well as between customers, and no account of whether the AI layer respects those boundaries when analyzing across a portfolio. A firm running conflicting matters for different clients in the same instance has no published basis to assess it.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
Disclosure in response to legal process is addressed and customer notice is not. The privacy policy permits sharing information with third parties when &AI believes, in its sole discretion, that doing so is necessary to comply with law, a court order, a subpoena or other legal process. Neither the policy nor the Terms of Service commit to notifying the customer or reserve a position on notice.
Not addressed. No government or law enforcement request clause, no commitment to notify a customer before producing their data, and no transparency report were located. The stakes are specific in this category: the vendor holds unpublished patent applications and pre filing invention disclosures, and disclosure of an unpublished application to a third party can bear on prior art and prior disclosure questions, which the vendor's own published material identifies as capable of extinguishing patent rights permanently. Checked the security page, the product pages, the blog library and the site navigation on 29 Aug 2026.
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 and material type without identifying the underlying databases. The vendor says it searches more than 60 million patent publications from major jurisdictions, plus non-patent literature such as research papers, standards and clinical trials from across the internet, and product listings, manuals, videos and teardowns. The jurisdictions are not listed, no data source or license is named, and no update cadence is published.
Named with scale and currency addressed, and no license basis stated. The vendor states that its models are trained on more than 50 million global public patents, which names the corpus, quantifies it, and identifies it as public patent data, and separately states that it conducts regular verifications and updates processes to ensure data accuracy and real time updates, which addresses currency directly. Currency is the element most records on this index omit entirely and it is credited here.
The licensing question is genuinely lighter for granted patent documents than for case law or clinical records, since patent specifications are published by offices as a condition of the grant, and the vendor never says so or states the basis on which it obtained and redistributes them. Also absent: no jurisdictional breakdown of which patent offices are covered, no historical date range, no update lag figure despite currency being claimed, and no statement on whether non patent literature is included, which is a material boundary for prior art and invalidity work.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
Searched the home, product, security and pricing pages on 18 September 2026. On a patent platform the good-law question is whether a cited patent remains in force; the product surfaces a patent's family, prosecution history and prior cases, but nothing states that legal status is checked or shown for cited patents or authority.
Not addressed, and applicable in a form specific to this category rather than a scope fact. The patent analog of a currency check is legal status: whether a patent is in force, expired, lapsed for unpaid maintenance fees, subject to a terminal disclaimer, amended in reexamination, or invalidated in inter partes review or litigation. That status determines whether an infringement or invalidity analysis is worth anything, and it changes constantly.
Nothing published states whether the platform surfaces legal status, tracks post grant proceedings, or would flag that a patent in a portfolio analysis has been invalidated. The vendor addresses data accuracy and real time updates for the corpus generally without stating that legal status is part of it. Checked the product pages, the infringement analysis material, the security page and the blog library on 29 Aug 2026.
Refusal and Uncertainty Behavior
What does the product do when the answer is not in the corpus?
Evidence is scored and explained without a documented abstention path. The vendor says it retrieves and scores evidence against each claim limitation and proposes each piece with a short relevance rationale for the user to accept or reject. Nothing describes what the system does when no adequate evidence exists for a limitation, or what Andy or the drafting tools do when they cannot support a statement.
Documented, and only the second record in the pull to reach this value after Descrybe. The vendor publishes multiple named safeguards against hallucinations: citations attached to outputs, auditing of sources for relevance, and color coded confidence indicators showing relevance. The color coding is the substantive element, because it surfaces the system's own uncertainty at the point the user reads the result rather than presenting every output with equal apparent authority, which is the failure mode that makes a confidently wrong answer dangerous.
The High, Medium and Low infringement risk scoring works the same way at portfolio level, presenting a graded signal rather than a verdict. Recorded as documented rather than documented and demonstrable because nothing published defines what the confidence bands mean, what threshold separates them, what a low confidence output should prompt a user to do, or whether the system will decline to produce a chart at all where evidence is absent.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
Searched the AI Hallucination Cases database maintained by Damien Charlotin and trade press reporting on 18 September 2026 for court records addressing fabricated or hallucinated legal citations in output from &AI. None located. This signal does not record litigation history of any other kind.
None located, with the instrument named. General web searches combining the vendor and product names with court, order, sanction, fabricated citation and patent terms returned nothing on 29 Aug 2026, and no named docket database or court record tracker was searched. Recorded as a statement about what this search found, not as a clearance. The exposure shape in this category differs from a research product: the analogous failure is an invalidity contention citing prior art that does not disclose what the chart claims, or a claim chart mapping product evidence to a limitation it does not meet, both of which would surface in litigation rather than as a fabricated case citation.
Flagged as worth a proper docket search on a later pass given the platform's stated penetration among Am Law 100 IP practices.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
Searched the home, product, security, pricing, terms and privacy pages and the blog index on 18 September 2026. No bar ethics opinion or court rule on AI is named or engaged in connection with the product. The blog addresses privilege risks of AI on patent work in general terms, which was not located in full.
Recorded at generic reference, and the value name understates this record, which the note exists to correct. The vendor publishes practitioner facing material naming specific authorities rather than gesturing at ethics: USPTO Rules of Professional Conduct, 37 CFR, Model Rule 1.6 on attorney confidentiality, 35 U.S.C. section 102 on prior disclosure bars, and duty of competence obligations under state bar rules. It sets those against named technical threat vectors including cross matter data leakage, model training on confidential inputs, vendor employee access to stored work product and inadequate breach notification, and states why the stakes differ in patent practice, since exposure of an unpublished application can permanently destroy patent rights rather than merely creating a compliance issue.
This is the most specific engagement with professional rules located in the pull, ahead of ProPlaintiff's ABA Model Rules reference. It is recorded at this value because the material is market education published in a blog library rather than a product commitment, no jurisdiction specific ethics opinion is analyzed, and nothing binds the vendor to any of it.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
The vendor's charge is built to follow the matter and reach the client, and client disclosure is not addressed. The Enterprise plan offers passthrough invoicing, the pricing page argues that credits track work performed by matter, and the vendor's blog asks why patent AI is sold by the seat when patent work is billed by the matter. The release announcing its seed round reported customer time savings of 70 to 90 percent. No guidance addresses how a firm should disclose or pass these charges to its client.
Savings claims only, denominated unusually precisely. Published: 80 percent reduction in project time, more than $30,000 saved per claim chart, and up to 15 hours recovered per patent application. The per claim chart figure is the notable one because it is denominated in dollars against a discrete deliverable rather than as a percentage, which makes it the most concrete savings claim in the pull, and it carries no baseline, methodology or sample.
Nothing appears on the client's side of the equation: no position on billing for AI assisted claim charting or drafting, no guidance on how a firm should record machine assisted time on an invoice, and no exportable record showing what portion of a chart or an application was machine generated. The omission is pointed for a product sold into Am Law 100 practices where charting is conventionally billed by the hour.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
Not addressed, and this is the strongest open publication on the index to receive that value, which the note records so the position is not misread. Published openly with no login: SOC 2 Type 2, ISO 27001 and ISO 42001 certifications, named penetration testing partners NCC Group and A-LIGN, encryption in transit and at rest, tenant segregation, least privilege access, a stated 90 day retention with deletion, and 24/7 monitoring.
A firm can cite all of that immediately. What is missing is everything that makes a pack forwardable and complete: no subprocessor list, no named model provider, no data processing agreement, no certificate dates or certifying bodies, and no trust center or request route to the underlying reports. A firm can repeat strong claims and cannot obtain a single document behind them. Compare EvenUp, DigitalOwl, Tavrn and Legalyze at on request, each of which publishes less and offers a route or an executable instrument.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
Some elements of a record travel with exported charts, short of a full disclosure record. Every citation in an exported claim chart points to a specific passage, figure or timestamp and must have been approved by the user, and the vendor says exports contain no generated language. Nothing records which model proposed the evidence, and no disclosure export is described for drafted contentions or expert reports, which do contain generated text.
Partial record, and the output format is unusually well suited to the question. A claim chart is by construction a disclosure artifact: it maps each claim limitation to the specific product evidence said to meet it, element by element, so an opposing party or a court can test every mapping individually, and the vendor states charts are citation backed with sources audited for relevance. Color coded confidence indicators travel with the output, which means a reader can see where the system itself was less certain, and that is closer to a machine generated candour signal than anything else on this index produces.
The familiar two limbs are absent: nothing indicates that output records which model produced it or on what date, and no human verification record is captured showing that an attorney reviewed and adopted a mapping before it was served or filed. For contentions filed in litigation or an inter partes review, that second gap is the one an opposing party would probe.
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.
- Practice Systems Integration Depth
- Good Law Verification
- Outside Counsel Guideline Readiness
Which one fits
Choose &AI if
- You need claim charts that contain nothing the model wrote. &AI requires the user to approve every proposed citation before it appears in an export, points each citation to a specific passage, figure or timestamp, and exports only exact quotations and references.
- You want to price patent work by the matter. &AI publishes credits at $0.40 each, a Pro plan at $625 per user a month with 2,500 credits, and an average of 241 credits per claim chart, and its Enterprise plan offers passthrough invoicing.
- You search products as well as patents. &AI searches more than 60 million patent publications alongside research papers, standards, product manuals, videos and teardowns for evidence of use, and Enterprise customers can choose single tenant or self hosted models.
Choose Patlytics if
- Your security review wants AI governance certified, not just described. Patlytics states SOC 2 Type 2, ISO 27001 and ISO 42001, the AI management system standard, and names NCC Group and A-LIGN as its penetration testing partners.
- You work across the whole patent lifecycle. Patlytics covers invention harvesting, drafting, prosecution, prior art search, infringement detection, invalidity contentions, claim charting, due diligence and portfolio pruning, with Portfolio Heatmaps scoring up to 250 patents against target products.
- You want to see how sure the system is. Patlytics attaches citations and color coded confidence indicators to its outputs, audits sources for relevance, and scores each patent and product pair as high, medium or low infringement risk.
In summary
&AI
&AI, from AndAI, Inc. of New York, is an AI workspace for patent litigation used by law firms from the Am Law 100 to IP boutiques, on plaintiff and defense sides, and by in house teams. It searches patents, literature and product materials for prior art and evidence of use, builds invalidity and infringement claim charts, analyzes validity under sections 101 to 112 and drafts contentions and expert reports, with an assistant called Andy. The AI Legal Index grades it in the top two bands on eight of fifteen capability axes, with A grades on AI centrality, autonomy and oversight, and pricing. It states SOC 2 Type II. As of 18 September 2026 the index located no named customer or named model provider.
Patlytics
Patlytics, based in New York, is an AI native patent platform for law firm and in house IP teams, covering invention harvesting, application drafting, prosecution, prior art search, infringement detection, invalidity contentions, claim charting, due diligence and portfolio pruning, with models fine tuned on more than 50 million public patents. The AI Legal Index grades it in the top two bands on ten of fifteen capability axes, with A grades on AI centrality, operational evidence, privilege and security certifications. It states SOC 2 Type 2, ISO 27001 and ISO 42001, and reports use by more than 40 percent of Am Law 100 IP practices. As of 29 August 2026 the index located no published price, liability position or named model provider.
Questions buyers ask
&AI vs Patlytics: which is better for patent litigation?
The grid places them close: Patlytics sits in the top two bands on ten of fifteen AI Legal Index capability axes and &AI on eight of fifteen, identical on four. Patlytics publishes more on certifications, retention and professional duty, and covers the whole patent lifecycle. &AI publishes its prices and a hard rule that exported claim charts contain only approved exact quotations. Litigators who serve charts on opponents have that rule to read from &AI.
Does &AI train on uploaded case documents?
Its marketing says no: the home and security pages state that &AI and its subprocessors never train on customer or client data. Its terms of service, effective 18 June 2026, grant it a perpetual, irrevocable, sublicensable license to use content users upload and let it make that content available to other users, without naming training. The terms are the binding text. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 27, 2026. No vendor pays for placement.
What is ISO 42001 and does Patlytics hold it?
ISO/IEC 42001 is the international standard for an artificial intelligence management system, covering risk management, accountability and lifecycle controls for AI, independently audited. Patlytics states that it holds ISO 42001 alongside SOC 2 Type 2 and ISO 27001. It publishes no certificate date, auditor or scope for any of the three. &AI states SOC 2 Type II only. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 27, 2026. No vendor pays for placement.
How much do &AI and Patlytics cost?
&AI publishes credits at $0.40 each on pay as you go, a Pro plan at $625 per user a month with 2,500 credits and further credits at $0.25, and custom Enterprise terms; it reports an average claim chart at 241 credits. Patlytics publishes no pricing at any level, although it claims more than $30,000 saved per claim chart. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 27, 2026. No vendor pays for placement.
What do &AI and Patlytics both leave unpublished?
Who stands behind a wrong chart, and whose model draws it. Neither offers an indemnity or warranty on output, neither names the foundation model or provider behind its AI, and neither says whether a cited patent's legal status is checked. Neither names a docketing or document management integration, although both sell into practices that run on them. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 27, 2026. No vendor pays for placement.
Three readings to weigh. The outside counsel guideline readiness signal has not been recorded for &AI; that is a gap in the record, not a finding about the vendor. Patlytics' security page says customer data is never used for training and, elsewhere on the same page, not without explicit consent or necessity; both are the vendor's words. On 17 September 2026 Patlytics announced a ChatGPT plugin built on its MCP server, after its record was graded. &AI was verified on 18 September 2026 and Patlytics on 29 August 2026. Neither vendor reviewed this page.
Neither vendor paid for inclusion, placement or a grade, and neither reviewed this page before it published. Everything above comes from public material on the dates shown. How the index grades.