Consilio vs Steno: how they compare in 2026

Machine learning is the engine of several capabilities a client pays for, and the businesses underneath it would carry on without it. Nine AI products are named and separately marketed: Guided AI Review and Native AI Review for responsiveness classification, AI PrivDetect and AI PrivGen for privilege work, AI Summarize, AI Investigate, Aurora Verity Review and the narrative analysis acquired with TrueLaw in June 2025. The models are described as custom-built in house and fine-tuned for legal document classification, with AI PrivGen trained on thousands of court-approved privilege log entries, and the scale claimed around them is real: more than 1,300 analytics engagements a year, over 250 million documents analysed, more than 40 dedicated AI team members. Two things hold it at this band rather than higher. Sightline is a complete eDiscovery and legal hold platform without any of it, and Aurora is a data unification and portfolio management layer whose core promise, freedom from platform lock-in and visibility across matters, is not a machine learning promise. And the AI is delivered wrapped in labour: AI Guides validate the prompts, privilege review experts quality-check every generated description, and the company's own framing is a hybrid in which technology enhances rather than replaces judgement. Graded level with Casepoint on the same reasoning.

Transcript Genius is a real generative product and it is given away. The vendor states that it is included in the product suite with no additional charge and no usage limitations, and the home page markets it as free with all Steno depositions. What a customer actually pays for is the services business underneath: court reporting, videography, interpreters, remote and hybrid depositions, equipment rental, conference rooms, and DelayPay deferred payment. Remove the AI and every one of those remains, along with Steno Connect for Zoom and the Firm Dashboard. The terms make the commercial position explicit by reserving the right to change the price of Transcript Genius or limit a person's usage of it at Steno's sole discretion, which is the language of a bundled extra rather than a core product. This is also where the membership screen sits: Steno is on the seed list as the deliberate borderline case, included because it ships a transcript analysis product rather than because it books court reporters, and that reading holds, but the same fact that makes it a software vendor at all keeps it low on this axis. The AI is a feature layer on a product whose value stands entirely without it.

Accuracy is asserted repeatedly and measured nowhere. The claims are specific in language and empty of numbers: descriptions that are accurate, defensible and court-ready, results that are reliable and consistent, documents identified with precision, a system trained on thousands of court-approved privilege log entries so that entries contain all required elements while avoiding common pitfalls. The one figure published is about effort rather than correctness, a reduction in attorney hours of up to seventy per cent on privilege log creation, hedged as what clients report. Searched the Sightline page, the Aurora page, the AI PrivGen page and the privacy policy on 31 Aug 2026 and located no accuracy rate, no recall or precision statistic, no validation protocol, no test set and no error analysis, which is a conspicuous absence for privilege classification, where a false negative discloses a privileged communication and the consequence is not recoverable. Nothing describes grounding either: no citation of the underlying document from a generated log entry, and no retrieval method. The substitute for measurement is people, with privilege review experts providing oversight to ensure accuracy and defensibility, which is a control rather than evidence. One limb does not apply and is neither credited nor penalised: a citator or good-law check is out of scope for a product operating on a client's own collected documents. Eight of the nine AI product pages were not opened and are the rebuttal route.

Grounding is real, documented and unusually well built into the product, and no number is published anywhere. Every generated answer carries a clickable page-line citation into the transcript, the interface presents the answer side by side with the source, and the vendor's own FAQ frames the point in one line: human verification is a simple click away. Summaries carry the same linked page-line citations so a reader can check any assertion against the original. Where a firm has ordered Text to Video Sync, a citation opens the synchronised deposition video at that moment, so a claim about testimony can be checked for tone and credibility rather than wording alone, which is the most complete verification path located in this pull. Retrieval is described at a working level, combining exact-match and semantically related search across the firm's transcripts. What is absent is measurement: no accuracy figure, no test set, no evaluation, no error rate, and no published comparison, against a marketing claim of unparalleled accuracy. The agreement is franker than the marketing and is recorded here for the gap: it states that AI is not perfect and may make mistakes, and disclaims any warranty of accuracy, completeness, truthfulness, timeliness or suitability for outputs. One limb does not apply and is neither credited nor penalised, since a citator is out of scope for a tool reading deposition transcripts rather than law.

Human oversight is not a caveat here, it is the delivery model, and it is described with more specificity than most vendors manage. The framing is the AI Trust Gap, and the answer given is a hybrid in which advanced AI is combined with seasoned professionals and guided by human oversight at every step, so that legal teams retain full control. The people are named as roles rather than gestured at: AI Guides bringing three stated types of expertise, review design, legal domain knowledge, and AI prompting and validation, who refine and validate prompts and ensure outputs are relevant and aligned with the matter. AI PrivGen publishes a four-stage flow ending in a named quality assurance step where privilege review experts check generated descriptions before delivery. Aurora adds customer-side control, with AI adopted at the client's own pace and the client choosing which models and review platforms operate on its data. What is not published, checked across those pages and the privacy policy on 31 Aug 2026: no threshold at which a model declines to produce an answer, no confidence or uncertainty signal exposed, no description of what happens when a generated description is wrong, and no correction route. The oversight model is well staffed and undocumented at the point of failure.

A written commitment that the tool sits under a lawyer's judgement, alongside the most candid statement about unsupervised operation in this pull. The agreement says both things in the same clause: that Transcript Genius uses artificial intelligence with no human oversight to assist with searching and interrogating transcripts, and that its outputs are provided as research aids and do not constitute legal advice, attorney work product, or a substitute for review of the underlying record. The first half tells a buyer that nobody at Steno checks an answer before it appears; the second places the obligation to check on the user and names the thing they must check against. Real review surfaces support it rather than leaving it as an instruction: per-answer page-line citations, a side-by-side transcript view, and video sync to the cited moment. Adoption is opt-in, with the terms requiring the customer to enable AI features through the Dashboard, and a design limit is published, in that only final ordered transcripts can be interrogated and not drafts. What is missing is the failure path: no abstention behaviour, no confidence or uncertainty indicator, no threshold at which the tool declines, and no route to report or correct a wrong answer.

Operational scale is published in quantity and customer outcomes are not attributed to anyone. The scale figures are real and specific: more than 1,300 analytics engagements a year, over 250 million documents analysed, more than 120 review managers trained on the AI suite, over 40 AI team members, upwards of 210 terabytes ingested monthly, more than 7,000 servers, over 30,000 matters hosted and fifteen or more data centres. Those describe the vendor rather than what changed for a client. On the customer side there is one figure, a reduction in attorney hours of up to seventy per cent on privilege log creation, attributed only to what clients report, with no named organisation, no matter, no date and no method. Searched the Sightline page, the Aurora page, the AI PrivGen page and the site navigation on 31 Aug 2026 and located no named customer anywhere. Two artifacts make the gap concrete rather than theoretical. The customer success section of the flagship AI product page still carries its unfilled template instruction, reading as a headline highlighting customer results with a direction to emphasise time saving and use numbers to maximise credibility. And the Sightline page's expert section renders roughly one hundred cards all reading Full Name and Job Title. A case study library exists and was not opened, which is the rebuttal route.

Named customers in quantity, no measurement attached to any of them. Five quotes carry a name, a title and a firm: a partner at Panish Shea Ravipudi, a director at Ivie Mcneill Wyatt Purcell & Diggs, a paralegal at Citron & Citron, an attorney at Gamliel Law, and an attorney at Burneikis Law who is also the subject of a published customer story about Transcript Genius. That is full attribution rather than a logo wall, and the firms are identifiable plaintiff and litigation practices rather than unnamed segments. What is absent is any figure. The customer story is qualitative, describing time reclaimed without saying how much, and none of the quotes carries a number or a date. One figure exists but sits outside the vendor's own current material, in trade coverage of the beta reporting an average saving of two to four hours per case, which is not published on any Steno page located on 31 Aug 2026 and is not credited here. No case study with a stated method was located. Checked the Transcript Genius page, the home page and the terms of use.

Confidentiality is asserted clearly and in general terms, and the document that would bind it is not published. What is published is not nothing. The privacy policy states that all client data is treated as confidential whether it includes personally identifiable information, protected health information or simply non-public information, and that Consilio's handling policies keep data and devices protected from unauthorised access while in its possession, supported by strict access control policies, network segregation between internet-facing and sensitive storage segments, and background screening of all employees and contractors. The AI PrivGen page goes directly at privilege, stating that privileged content never leaves Consilio's secure environment and that attorney-client confidentiality is maintained, with processing on private data centres rather than public cloud. What is missing is the whole contractual layer. No terms of service, master agreement or data processing addendum is published anywhere on the property, so there is no readable commitment on training with client material, no matter or client segregation model, no ethical wall description, and no work product treatment. The privacy policy itself narrows the point, stating that Consilio acts primarily as data processor for client services, which places the operative terms in client contracts a buyer cannot read in advance. Checked the full footer inventory on 31 Aug 2026.

Substantive published commitments on both confidentiality and training use, with one disclosure a buyer needs to read carefully. The agreement limits vendor access directly: Steno will not access or view Content except as authorised or instructed by the customer, as required to perform its obligations, as provided in the agreement and privacy policy, or as required by law, and undertakes physical and technical safeguards consistent with SOC 2. The customer retains copyright in its Content and grants only a revocable licence to access it to provide and operate the Services, with a further carve-out permitting aggregated and de-identified use to improve the Services. Training is addressed twice and in the customer's favour, with the terms stating that data shared with Anthropic through Steno is not used as broader training data and the product page stating that transcripts and conversations are never used to train the AI and that AI partners do not retain data after a session. The disclosure that needs reading is about recordings: the terms state that Steno may record the entirety of proceedings for backup purposes, including portions considered off the record, which in a deposition can include exchanges counsel would treat as privileged, and commit only that a backup recording will not be shared absent a valid subpoena without objection or a court order. What is missing is segregation: nothing describes a wall between matters, and the Firm Dashboard is designed for search across all of a firm's depositions at once.

Nothing published addresses the advice line, and there is no document on the property that would carry it. The footer inventory is privacy policy, privacy rights requests, anti-slavery statement, cookies policy, cookies preference manager and a PAIA manual. There is no terms of service, no disclaimer page and no professional responsibility statement, so nothing states that output is not legal advice, that no attorney-client relationship arises with Consilio, or where responsibility sits when a generated privilege description proves wrong. No bar or ethics authority is named anywhere, including ABA Formal Opinion 512. The exposure is sharper here than for a pure software vendor, because Consilio's own reviewers and privilege experts exercise legal judgement on client matters as part of the service, and the published material describes that arrangement warmly without addressing the professional responsibility questions it raises. One thing does exist and is recorded rather than credited: a supervision and competence posture is published, in that human oversight is promised at every step and technology is said to enhance rather than replace critical legal judgement. That is a statement about how the work is done, not about who answers for it. Checked the Sightline, Aurora and AI PrivGen pages and the privacy policy on 31 Aug 2026.

The single best-drafted advice-line clause located in this pull, and no engagement with the guidance behind it. The AI section of the terms states that AI tools are provided as research aids and do not constitute legal advice, attorney work product, or a substitute for review of the underlying record. That one sentence does three things most vendors leave to three separate clauses or omit entirely: it separates tooling from advice, it disclaims work product status rather than implying it, and it names the supervision obligation in concrete terms by identifying what the lawyer must go back and read. The general disclaimer reinforces it, stating that content on the Services is informational only, is not advice of any kind, is not a substitute for professional advice including legal advice, and that the user must seek independent professional advice before acting. Eligibility is stated at eighteen and above with a competence representation. What holds it below the top band is the surrounding framework rather than the clause. No bar or ethics authority is named on any product or policy surface, including ABA Formal Opinion 512, and no jurisdictional limit is stated. The vendor's blog carries an item titled as covering ABA and state legal ethics guidance on artificial intelligence, which was not opened on 31 Aug 2026 and is the rebuttal route.

Principles and a staffing model are published; a governance framework a buyer could audit is not. The published material is organised around the AI Trust Gap, and it names ethical concerns and the absence of clear governance frameworks as an industry problem Aurora is meant to answer, with the answer given as expert oversight, client choice over which models run, and processing on private infrastructure. Two compliance statements sit in the privacy policy and are the closest thing to a mechanism: the right not to be subject to a decision based solely on automated processing is recited, and Consilio states that it complies with applicable CCPA and CPPA requirements including those relating to risk assessments, cybersecurity and automated decision-making technologies. Beyond that, searched the Sightline, Aurora and AI PrivGen pages and the privacy policy on 31 Aug 2026 and found no responsible AI page, no accountable owner for model behaviour, no pre-release testing regime, no model documentation and nothing at all on bias or uneven output. The gap has an edge on this product: models that classify responsiveness and privilege determine which documents a reviewing lawyer ever sees, and uneven performance across custodians, languages or document types is the question, unaddressed. A chief information security officer is named as a role, for security rather than for AI.

What exists is disclosure rather than governance, and one published term runs directly against a governance reading. The agreement is unusually forthcoming: it names the model, states that the tool runs without human oversight, commits that the model provider does not train on the data, and separately discloses that Steno uses AI internally for drafting communications, support triage, software development and routine operations under agreements that prevent those providers training on the data. Disclosing internal AI use at all is rare. But none of that is a governance framework. Searched the Transcript Genius page, the home page and the complete terms of use on 31 Aug 2026 and located no responsible AI page or principles, no accountable owner for model behaviour, no pre-release testing or evaluation regime, no published result, and no model documentation. The term that cuts against a governance reading is explicit: Steno may change the generative AI platform on which Transcript Genius is built at any time and in its sole discretion, which is a reserved right rather than a change-management commitment. Nothing anywhere addresses bias or uneven output, and the gap has a specific edge here, since the product analyses spoken testimony and the same company books interpreters, so performance across accents, dialects and interpreted evidence is the live question and is unasked.

The privacy policy carries a substantial safeguarding section that covers most of this ground, which matters more than usual here because it is the only substantive published document on the property. Certifications are stated with the holding entity distinguished: Consilio is certified to ISO/IEC 27001:2022 and HITRUST CSF, its collocated data centres are certified to either ISO 27001 or SOC 2 Type 2, and its UK data centre and offices hold Cyber Essentials Plus. Named controls follow: a chief information security officer heading the security team, enterprise policies reviewed by management, a vendor management programme requiring vetting and periodic security audits of suppliers, network segregation using a DMZ and multiple firewalls between internet-facing segments and sensitive storage, intrusion detection and prevention, third-party monitoring, background screening of all employees and contractors, an incident management policy, disaster recovery and business continuity policies, and a backup policy specified as data replication plus daily incremental and weekly full backups. A list of current third-party suppliers is published at a stated public URL. What is absent: no retention period for anything, since the Data Retention and Disposal Policy is named but not published and is obtainable only by email; no deletion commitment for client data; and no breach notification timeframe, the incident policy being described rather than published. Policy updated 17 Aug 2026.

Most of the ground is covered across the agreement and the product page, with the usual two gaps. Retention and deletion are customer-controlled and stated plainly: uploaded documents, recordings and other materials are stored until the customer instructs Steno to delete or destroy them, which places the lifetime in the customer's hands rather than behind a schedule. Model-side retention is addressed separately and tightly, with AI partners stated not to retain data after a Transcript Genius session and the conversation thread kept only to give follow-up questions context and not shared to others. Access is limited contractually, with Steno undertaking not to access or view Content outside four named circumstances. Certification is claimed as SOC 2 Type II and HIPAA, and a data processing addendum is published and incorporated into the terms where applicable law requires. Third parties are named where they touch data: Anthropic as the model provider, Stripe and Bill.com for payments. What is missing is a maintained subprocessor list beyond those named in passing, and any incident response or breach notification practice, neither of which was located on the readable surfaces on 31 Aug 2026. The Vanta-hosted trust centre and the published DPA were not read and are the rebuttal route for both.

Nothing is published on who bears the loss when the system is wrong, because no customer agreement of any kind is published. The footer legal inventory was read in full on 31 Aug 2026 and consists of a privacy policy, a privacy rights request portal, an anti-slavery statement, a cookies policy, a cookies preference manager and a PAIA manual. There is no terms of service, no master services agreement, no service level agreement, no data processing addendum and no acceptable use policy. In consequence nothing states a liability cap, an indemnity in either direction, a warranty on the software or on generated output, an insurance position, a limitation period, a governing law or a forum. A buyer evaluating this vendor cannot read the allocation of risk in advance at any level, and the only route to it is a negotiated contract. The contrast within this category is the useful measure rather than the criticism: Casepoint, read the same day, publishes a complete standard agreement including a dedicated clause on AI inputs and outputs, a stated cap, named carve-outs and a seventy-two hour breach notification. Consilio hosts more than thirty thousand matters across fifteen or more data centres and publishes none of it.

The exposure the AI creates is disclaimed precisely and nothing is allocated back to the buyer. The terms cap Steno's liability for any damages arising from or related to the agreement, for any cause and regardless of the form of action, at one hundred US dollars, and state that more than one claim will not enlarge the limit. No carve-out from that cap is named anywhere, which distinguishes this from the other capped agreements in this category that at least exclude fraud, wilful misconduct or confidentiality breaches. Indirect, consequential, exemplary, incidental, special and punitive damages are excluded, suppliers are given no liability of any kind, and a broad release with a California Civil Code 1542 waiver runs from the customer to Steno. Disputes go to mandatory individual arbitration before the AAA in Los Angeles with a class action waiver and a thirty-day opt-out. A dedicated AI disclaimer states that outputs are provided as is and with all faults, with no representation or warranty of accuracy, completeness, truthfulness, timeliness or suitability, and that the customer assumes all risks associated with using them. The only indemnity in the agreement runs from the customer to Steno. No warranty attaches to any output, and no insurance position appears.

Integration is the product rather than a feature, and the named connections are to competitors, which is unusual and creditable. Aurora is described as platform-agnostic and deliberately upstream of review, integrating with multiple review platforms and naming Relativity, Relativity Server, Everlaw and Sightline, with the stated purpose of freeing an organisation from single-vendor lock-in so it can select the optimal platform for each matter. Supporting mechanics are named: enhanced auto-loaders and data mapping capabilities that reduce processing volumes, and a version of AI Investigate built for Relativity Server so the AI reaches data held in a third-party platform. On collection, Aurora is said to connect with an organisation's data sources from email servers and cloud storage to collaboration platforms and enterprise systems. What is missing is depth and specificity. No individual data source is named, only categories; no document management system, matter management, e-billing or filing integration is named; no API or developer documentation was located; and nothing describes what moves in which direction or what a client must configure. Checked the Sightline, Aurora and AI PrivGen pages and the navigation on 31 Aug 2026; the technology partners page was not opened and is the rebuttal route.

The first named practice management integrations in this category. Steno publishes dedicated pages for a Clio integration and a Litify integration, both of which are legal practice systems in daily use rather than generic cloud storage, and both are surfaced in the primary navigation under Technology and Integrations rather than buried. Steno Connect for Zoom is a deeper integration than the name suggests, running the deposition inside Zoom with exhibit upload, sharing, numbering and annotation handled in the meeting rather than alongside it. The terms add third-party account linking for Google and Microsoft credentials and a general provision on connected accounts governing export of information to third-party services. What is not established is depth. The Clio and Litify pages were not opened on 31 Aug 2026, so what those integrations actually move, in which direction, and what a firm must configure is unverified, and no application programming interface, developer documentation or integration index was located. Nothing names a document management system, with nothing found for iManage or NetDocuments, and nothing for e-billing or court filing.

The deployment model is stated plainly and repeatedly, and it is the company's central differentiator rather than a technical footnote: everything runs on Consilio's own private hosted infrastructure and not on the public cloud, a point made on the Sightline page, the Aurora page and the AI PrivGen page alike, with the argument that privileged content never reaches a public cloud provider. The estate is quantified at more than fifteen global data centres, over seven thousand servers and more than thirty thousand matters hosted, with offices and data centres across North America, Europe and Asia. Residency is addressed in part: Aurora promises complete data sovereignty in the client's chosen jurisdiction, its UK and European launch is stated to be hosted in Consilio's UK data centre with GDPR and UK data protection compliance, ITAR compliance is claimed, and cross-border transfer is handled through an intra-company data transfer agreement incorporating the EU Standard Contractual Clauses and the UK Addendum. What is not published: no list of the data centre locations, so a buyer cannot check which jurisdictions are actually available; no tenancy model; and no distinction between where data is stored and where processing or model inference happens. Read 31 Aug 2026.

Cloud delivery is implied throughout and the residency question is never reached on any readable surface. The product is a browser-based dashboard with a Zoom application, and the terms describe materials being stored on Steno's systems and housed securely pursuant to its security protocols, which establishes that the vendor holds the data without saying where. Searched the Transcript Genius page, the home page and the complete terms of use on 31 Aug 2026 and located no cloud provider, no region, no data centre location, no residency option or commitment, no tenancy model, and no statement distinguishing where transcripts are stored from where model inference runs, which matters here because the terms name a third-party model provider that processes the content. One retrieval limit belongs on the record rather than in the grade: the vendor operates a Vanta-hosted trust centre at trust.steno.com, linked from the primary navigation and every page footer, whose body renders client-side and returned no content on fetch on 31 Aug 2026. That is a limit on this index rather than an absence on the vendor's side, and nothing in it is inferred here, but it is the most likely home for infrastructure detail and should be read before this grade is relied on.

The certifications are real, current and stated with unusual care about which entity holds what, and there is no security page or trust centre anywhere to put them in. What is claimed: Consilio itself certified to ISO/IEC 27001:2022 and HITRUST CSF; its collocated data centres certified to either ISO 27001 or SOC 2 Type 2, which is a meaningful distinction most vendors blur, since the SOC 2 attaches to the facilities rather than to Consilio; UK data centre and offices certified under Cyber Essentials Plus; ITAR compliance; and certification to the EU-US Data Privacy Framework with the UK Extension and the Swiss-US Framework, lodged with the US Department of Commerce and independently verifiable on the public Data Privacy Framework register, with JAMS named as the recourse provider and the Federal Trade Commission named as having jurisdiction. That last item is the one attestation a buyer can confirm without asking. Against it: no security page exists, so all of this sits inside a privacy policy; no auditor is named; no scope, coverage period or report date is given for the ISO or HITRUST certifications; and no route is published for obtaining any report. Footer inventory checked in full on 31 Aug 2026.

Certification is real and stated in two places, with none of the surrounding detail that would let a buyer test it. SOC 2 Type II and HIPAA compliance are asserted on the Transcript Genius page as covering all data in the platform, and the agreement reinforces the first by undertaking safeguards in accordance with the policies guiding SOC 2 compliance. A dedicated trust centre exists at trust.steno.com, hosted on Vanta, and is linked from the primary navigation and from every page footer, which satisfies the connector this axis looks for and puts it a long way ahead of the vendors in this pull whose attestations sit on an unlinked parent portal. The vendor also extends the point outward rather than stopping at itself, stating that its AI partners adhere to the same standards and linking directly to Anthropic's own trust centre, terms and privacy policy from the agreement. What is absent: no auditor is named, no report date or coverage period is given, no scope statement appears, and no route to obtain a report is published on any readable surface. The trust centre body renders client-side and did not return content on 31 Aug 2026, which is recorded as a retrieval limit rather than graded, and nothing inside it is credited.

The vendor says the models are its own and stops there. What is published: AI algorithms described as custom-built and developed in house, hosted on Consilio's private cloud; models fine-tuned for legal document classification; AI PrivGen models described as specifically trained on legal privilege concepts and on thousands of court-approved privilege log entries, which is more than most vendors say about training data; and TrueLaw's technology integrated following the June 2025 acquisition. What a buyer cannot establish: whether any third-party foundation model sits beneath the fine-tuning, and if so whose; no model name, version or provider anywhere; no inference location stated as distinct from the data residency claims; and no notice commitment on model change. One unreconciled tension belongs in the record. The AI product pages say the models are custom-built in house, while Aurora is marketed on letting clients leverage AI models they choose and avoiding lock-in to proprietary models, which implies third-party models in the stack. Nothing published reconciles the two. A list of current third-party suppliers is published at a stated URL and was not opened on 31 Aug 2026; it is the most likely place for a provider name and is the rebuttal route.

The most specific model disclosure in this pull, and it is in the agreement rather than the marketing. The terms state that Transcript Genius is built primarily on the large language model of Anthropic's Claude technology, name the provider, and link directly to that provider's trust centre, commercial terms and privacy policy so a buyer can read the downstream commitments themselves. No other vendor across pull 2 names the actual model family its product runs on. The data position is stated alongside it, with data shared with that provider not used as broader training data, and a separate paragraph disclosing that Steno's internal AI use runs under agreements preventing those providers from training on data. What keeps it out of the top band is the change term, which is published and adverse: Steno may change the generative AI platform on which Transcript Genius is built at any time and in its sole discretion. That is a disclosure rather than a notification commitment, and it means the model behind an answer given today may not be identifiable tomorrow. Also absent: no model version, no statement of where inference runs, and no subprocessor list beyond the providers named in passing.

No pricing information is published at any level, including the unit of charge, and no pricing page exists. The full primary navigation covering solutions, expertise, technology, resources and about, and the complete footer, were read on 31 Aug 2026: there is no pricing entry in either. No rate, no unit, no tier or package names, no volume band, no term length and no statement of what implementation or managed review adds. Every call to action across the Sightline, Aurora and AI PrivGen pages is a demo request or a contact form. The one place pricing appears at all is as an argument about other vendors: Aurora's marketing states that public-cloud platforms trap legal teams in rigid contracts with annual price hikes of eight to twenty per cent, and offers complete cost visibility across a portfolio as the remedy. A platform sold substantially on controlling and making visible the buyer's legal technology spend publishes nothing about its own.

Real figures are published and none of them is the price of the thing a firm actually buys. What a buyer can learn without asking: Transcript Genius carries no additional charge and no usage limitation, though Steno reserves the right to change that at its sole discretion; cancellation of a court reporter or videographer after 4pm on the business day before a job attracts a late fee of up to 400 dollars; interpreter cancellation terms run from 48 hours to 30 days in advance depending on dialect and the fee may be the interpreter's full charge; invoices more than 30 days late accrue interest at the lesser of 1.5 per cent per month or the legal maximum; and DelayPay defers case costs until settlement. Those are specific, unusual to publish, and all ancillary. The core rate is not published: the terms state that Steno makes its rates available on a rate card provided upon request and that rates may be separately negotiated, and that any quotation is a non-binding estimate for which the firm remains liable if the invoice exceeds it. There is no pricing page, no tier structure and no per-page, per-hour or per-witness figure. A buyer can price a cancellation to the dollar and cannot price a deposition at all.

Coverage is documented across several dimensions with real substance, and nowhere states a limit. Practice areas are enumerated as litigation and arbitration, global investigations, antitrust and competition, cyber incident response, corporate transactions, and risk and compliance. Industries are enumerated separately as financial services, pharmaceutical and life sciences and healthcare, insurance, transportation and mobility, consumer brands, and technology. Service lines run to eight, including eDiscovery, analytics and AI, document review, risk management and compliance, talent, cyber incident response, data forensics and investigations, and legal transformation. Product-level coverage is more specific still, with Sightline published against regulatory response, cyber incident response, data subject access requests, early case assessment, PII and PHI detection and review, multimedia review, small matters and investigations, and AI PrivGen against complex litigation, multi-party matters, regulatory investigations and mergers and acquisitions. The buyer types are named as multinational law firms, corporations and government organisations, and the site is published in nine locales. What is absent is any boundary: no practice area or data type is identified as unsupported, no jurisdictional limit is stated, and no matter size threshold appears, though a distinct Small Matters module implies the range is deliberate. Checked 31 Aug 2026.

Who the product is for is clear and where it reaches is not stated, which for a deposition services business is the question that matters most. The buyer is identified without ambiguity across the material: litigation teams at law firms, with the deferred payment offering aimed squarely at contingency and plaintiff-side practices, and the terms contemplating firm administrators managing end users such as legal assistants and paralegals. Product coverage is stated in useful detail, with Transcript Genius accepting third-party transcripts in text and PDF alongside Steno's own so a firm can analyse depositions taken by opposing counsel, and with proceedings of any scale claimed from single depositions to complex multi-party matters. What is absent is geography and practice. No list of states or jurisdictions in which Steno can staff a proceeding was located, which is the decisive coverage question for court reporting, and the only geographic signal on the readable surfaces is a navigation entry for litigation support services specific to California and a Nevada firm registration in the footer. No practice area is named as served or unsupported. The services pages were not opened on 31 Aug 2026 and are the rebuttal route.

No located document states whether client material is used to train or refine any model, either way, and the document that would say so is not published. The privacy policy lists improving the Products and Services among the purposes for every category of Personal Data it covers, which is the usual improvement language, but it governs Personal Data such as contact and technical information rather than the collected corpus a client uploads, and it states that Consilio acts primarily as data processor for client services, placing those terms in contracts that are not public. No terms of service, master agreement or data processing addendum exists on the property. What the vendor does say about training data points elsewhere: AI PrivGen models are described as trained on thousands of court-approved privilege log entries and on legal privilege concepts, with no statement about whether client documents contribute. Checked the full footer inventory, the Sightline, Aurora and AI PrivGen pages on 31 Aug 2026.

The commitment sits in the agreement rather than in a policy page, which is what lifts it to this value. The AI section of the terms of use, effective 13 May 2025, names the model provider and states that data shared with it through Steno is not used as broader training data, and a separate paragraph on Steno's internal AI use states that its agreements with language model providers ensure they do not train on data provided to them. The product page states the same thing more broadly, that transcripts and conversations are never used to train the AI, and the FAQ adds that the conversation is saved within a context window or thread only to allow follow-up questions and is not shared to others. One carve-out belongs in the record and a buyer should read it alongside: the Content licence in the terms permits Steno to use customer Content on an aggregated, de-identified basis for the purposes of improving the Services, which is narrower than model training but is not nothing.

Retention is acknowledged, deferred to an unpublished document, and never quantified. The privacy policy states that Personal Data is processed for as long as is reasonably necessary to comply with contractual or legal obligations and to pursue legitimate interests, and that it is retained in accordance with a named internal document, the Data Retention and Disposal Policy, which is not published and is obtainable only by emailing the privacy address. That is better than silence, because the instrument is named and a route to it exists, and it is short of a period a buyer could rely on. Nothing separately addresses prompts submitted to the AI products or the descriptions and summaries they generate: no retention period, no customer-configurable setting, no zero-retention option, and no statement of what happens to generated output at the end of a matter. Policy updated 17 Aug 2026 and read the same day.

Retention is placed in the customer's hands and stated in the agreement. The terms provide that materials a customer uploads through the Dashboard, Steno Connect for Zoom or a third-party integration are stored on Steno's systems until the customer instructs Steno to delete or destroy them, so the lifetime is a customer decision rather than a schedule the vendor sets. Model-side retention is separately closed off, with the product page stating that AI partners do not retain data after a Transcript Genius session. One element is disclosed without a period and should be read as the gap: the FAQ states that a conversation is saved within a context window or thread to enable follow-up questions with prior context, and that the thread is not shared to others, but nothing says how long Steno keeps that thread or whether deleting the underlying transcript removes it. Checked the terms, the Transcript Genius page and the home page on 31 Aug 2026; the data processing addendum was not opened.

Separation is claimed at the level of security controls and never described as a permission model. The privacy policy states that all client data is treated as confidential whether or not it contains personal or health information, that handling policies keep data protected from unauthorised access while in Consilio's possession, and that networks use a DMZ and multiple firewalls with strict access control policies to separate internet-facing segments from sensitive data storage. The AI PrivGen page adds that privileged content stays inside Consilio's environment and that attorney-client confidentiality is maintained. None of that is a segregation mechanism. Nothing published describes separation between one client's matters and another's, between two matters for the same client, or between review teams inside a matter, and nothing states whether the AI products can retrieve across matter boundaries. The question carries unusual weight for a provider that also staffs managed review, since reviewers move between engagements. Checked the privacy policy and the Sightline, Aurora and AI PrivGen pages on 31 Aug 2026.

Nothing addresses separation between matters, and the product is designed to run the other way. Transcript Genius is marketed on searching across all the depositions in a firm's Dashboard at once, with filtering by case, attorney or witness offered as a convenience rather than a boundary, and multi-transcript interrogation across a case is the headline capability. Access control is described only at firm level: the terms make a designated Administrator responsible for maintaining the list of End Users on the Firm Dashboard and for removing people who should no longer have access, and disclaim liability for anything arising from that list being wrong. Searched the Transcript Genius page, the home page and the complete terms of use on 31 Aug 2026 and located no ethical wall concept, no matter-level permission model, no statement that a user can be restricted to particular cases, and nothing about whether the AI respects any such restriction. For a firm running two matters with adverse interests, or screening a lateral hire, nothing published answers the question.

Disclosure to authorities is contemplated and notice to the customer is never reached. The privacy policy states, in its Data Privacy Framework section, that there may be a requirement for Consilio to disclose personal information in response to lawful requests by public authorities including to meet national security or law enforcement requirements, and its recipients table names government officials, law enforcement or others where permitted by the notice or required by law. Nothing commits Consilio to notifying the affected client before or after disclosure, to seeking a protective order, to resisting or narrowing a demand, or to seeking a waiver where notice is prohibited, and no transparency report or disclosure statistics were located. No terms of service exists that might carry a stronger commitment. Checked 31 Aug 2026.

The situation is addressed for one category of material and no notice is committed for any of it. The terms state that in certain jurisdictions Steno may record remote depositions for backup purposes, and that it will not share such a recording absent the receipt of a valid subpoena without objection from any other party or entity, or a valid court order. That contemplates a party objecting, which implies someone would know, but nothing commits Steno to notifying the customer that a demand has arrived, to resisting or narrowing it, or to seeking a protective order. The provision covers backup recordings only; for Content generally the terms say Steno will not access or view it except, among other grounds, as required by law, with no notice attached. A related disclosure belongs alongside it, because it enlarges what could be demanded: Steno states it may record the entirety of proceedings for backup purposes, including portions considered off the record. No transparency report was located on 31 Aug 2026.

The question does not arise in its usual form and a related one does. Consilio operates on documents its clients collect and upload, so there is no vendor-assembled corpus of primary law, no third-party content licence and no upstream supplier of legal material to identify. Recorded as not addressed on that basis, with the reason stated rather than left to look like neglect. The adjacent question that does bite is training-set provenance: AI PrivGen is described as trained on thousands of court-approved privilege log entries, and nothing published states where those entries came from, whether they originated in Consilio's own client work, whether any consent or licence underpins their use, or how they were selected. For a model whose output is offered as court-ready, the provenance of the material it learned from is a live question and it is unanswered. Searched the Sightline, Aurora and AI PrivGen pages and the privacy policy on 31 Aug 2026.

The question does not arise in its usual form for this product. Transcript Genius operates on the deposition transcripts of the customer's own matters, whether produced by Steno's court reporters or uploaded by the firm from another source, so there is no vendor-assembled corpus of primary law, no third-party content licence and no upstream data supplier to identify. The terms confirm the direction of ownership, with the customer retaining copyright and any other rights it already holds in Content it submits, and Steno taking only a revocable licence to access that Content to provide and operate the Services. Recorded as not addressed because that is the honest value, with the reason stated so it does not read as a gap. One product limit sits close to provenance and is worth recording: only final, ordered transcripts can be interrogated, not drafts or un-ordered ones, which the vendor frames as preserving the work court reporters provide and which also means the analysed text is a certified record.

Not applicable to this product class, and neither credited nor penalised. Consilio's platforms search, classify and summarise a client's own collected documents; they do not retrieve primary law or assert propositions whose continued validity would need checking, and no citator or treatment signal is claimed anywhere. Searched the Sightline page, the Aurora page, the AI PrivGen page and the privacy policy on 31 Aug 2026. The accuracy question that does apply to this product is validation of classification and privilege calls, and it is recorded on the Citation Accuracy axis, where no recall, precision or error figure was located.

Not applicable to this product class, and neither credited nor penalised. Transcript Genius reads deposition testimony rather than legal authority, so there is no proposition of law whose continued validity would need checking and no citator is claimed anywhere. Searched the Transcript Genius page, the home page and the complete terms of use on 31 Aug 2026. The verification question that does apply to this product is whether a generated answer or summary faithfully reflects the transcript, and the vendor answers it structurally rather than through a citator, with every response carrying a clickable page-line citation and a side-by-side view of the source; that is recorded on the Citation Accuracy axis, where no measurement of correctness was located.

Nothing published describes what the models do when they are unsure. Searched the Sightline, Aurora and AI PrivGen pages and the privacy policy on 31 Aug 2026 and located no abstention path, no confidence score or uncertainty indicator surfaced to a reviewer, no statement of behaviour where a document is ambiguous or the basis for privilege is unclear, and no threshold at which a document is routed to a human rather than described automatically. The vendor's answer to uncertainty is organisational rather than technical and is recorded here because it is the substitute: privilege review experts provide quality assurance over generated descriptions, and AI Guides validate prompts, so the design assumes a person catches what the model gets wrong rather than the model flagging it. Nothing states how that review is sampled or scoped.

Nothing published describes what the tool does when the transcript does not support the question. Searched the Transcript Genius page, the home page and the complete terms of use on 31 Aug 2026 and located no abstention path, no confidence or relevance indicator surfaced to the user, no threshold at which the tool declines to answer, and no statement of behaviour where testimony is ambiguous or the requested contradiction does not exist. The terms acknowledge the risk in general terms, stating that AI is not perfect and may make mistakes and that outputs carry no warranty of accuracy or completeness, and separately that the tool runs with no human oversight. The vendor's answer to uncertainty is verification rather than signalling, and is recorded because it is the substitute: every answer is cited to a page and line with the transcript shown alongside, so a user can check any individual claim, without being told which claims the system was least sure of.

No court order, opinion or disciplinary record naming this company or its products has been located as of 31 Aug 2026. Instrument searched: the AI Hallucination Cases database maintained by Damien Charlotin, which tracks decisions worldwide where a court addressed hallucinated AI content and records the tool implicated where known, together with several independent 2026 sanctions trackers and law library guides, searched on the company name and on the product names. This is a statement about the public record on the date shown rather than a clearance. The failure mode this signal tracks also sits at an angle to the product: Consilio's AI classifies and describes a client's own documents rather than citing legal authority, so a fabricated citation reaching a filing would originate elsewhere, while an inaccurate privilege log description, which is the risk this product actually carries, would surface as a privilege dispute rather than as a hallucination case.

No court order, opinion or disciplinary record naming this product has been located as of 31 Aug 2026. Instrument searched: the AI Hallucination Cases database maintained by Damien Charlotin, which tracks decisions worldwide where a court addressed hallucinated AI content and records the tool implicated where known, alongside several independent 2026 sanctions trackers and trade coverage, searched on the company name and the product name. This is a statement about the public record on the date shown rather than a clearance. The failure mode also sits at an angle to what this product does: Transcript Genius cites page and line locations inside a transcript the customer already holds rather than legal authority, so a fabricated case citation reaching a filing would originate elsewhere, while a citation to testimony that does not say what the summary claims would be a different failure, one this signal does not capture and one the product's side-by-side verification design is built to catch.

Defensibility and court-readiness are claimed throughout and no authority is ever named behind them. AI PrivGen is marketed as producing court-ready privilege log descriptions that withstand scrutiny, containing all required elements while avoiding common pitfalls that might compromise privilege claims, and trained on thousands of court-approved entries. Searched the Sightline, Aurora and AI PrivGen pages and the privacy policy on 31 Aug 2026 and located no citation to any rule, standard or opinion: nothing names Federal Rule of Civil Procedure 26(b)(5), which governs what a privilege log must contain, no Sedona Conference commentary, no ABA Formal Opinion 512 and no state bar guidance. For a product whose entire output is a document filed to sustain a privilege claim, the absence of the rule that defines sufficiency is the finding. Nothing maps any feature to a professional obligation a supervising lawyer could evidence.

No professional conduct authority is named on any product or policy surface. Searched the Transcript Genius page, the home page and the complete terms of use, effective 13 May 2025, on 31 Aug 2026 and located nothing citing ABA Formal Opinion 512, any state bar guidance on generative AI, or any rule governing the use of AI-assisted work in filings. The agreement does state a clear professional position without sourcing it, that AI outputs are research aids and do not constitute legal advice, attorney work product, or a substitute for review of the underlying record, which addresses the substance of the competence and supervision duties those authorities describe while citing none of them. One rebuttal route is recorded rather than credited: the vendor's blog carries an item titled as covering ABA and state legal ethics guidance on artificial intelligence, which was not opened. Editorial coverage of guidance is not the same as measuring the product against it, but it should be read before this value is relied on.

Savings are the headline and the client's bill is not addressed. The published claims are a reduction in attorney hours of up to seventy per cent on privilege log creation, lower overall review costs through automation of repetitive tasks, and substantial cost reductions without sacrificing thoroughness, alongside Aurora's portfolio-level cost visibility and real-time matter cost alerts. Searched the same pages and the privacy policy on 31 Aug 2026 and located no per matter record of AI-assisted work intended for fee purposes and no guidance on billing, fee or client disclosure treatment where AI-generated work product is delivered. The question is unusually direct for this vendor, because Consilio's own charges are frequently a disbursement passed through to a client and its reviewers' output is billed work, so an hours reduction of the size claimed changes what a client pays and nothing published addresses how that is recorded or disclosed.

Time savings are claimed and the client's side of the bill is addressed only for the underlying service, not for the AI. The efficiency claims run through the material: hours of transcript review reduced to minutes, summaries generated in minutes instead of hours, and hours saved on every case. Nothing published gives a per matter record of AI-assisted work for fee purposes, and no guidance exists on how a firm should treat or disclose AI-assisted analysis when billing a client. Two adjacent facts are recorded because they change the shape of the question for this vendor. Transcript Genius carries no additional charge, so there is no separate technology cost to pass through in the first place, which removes the disbursement question that metered AI tools create. And DelayPay defers case costs until settlement, which is a real published position on when a client bears litigation expense, though it concerns court reporting fees rather than AI. Checked the Transcript Genius page, the home page and the terms of use on 31 Aug 2026.

A current third-party supplier list is published, which is the core of what this signal asks for. The privacy policy states that Consilio may use third-party sub-processors to deliver services, that engaged processors work under a specified contract with appropriate technical safeguards, and links a list of current third-party suppliers at a public URL requiring no agreement or request. Alongside it a firm can point a client to certifications stated in the same document, ISO/IEC 27001:2022 and HITRUST CSF for Consilio, ISO 27001 or SOC 2 Type 2 for its data centres, Cyber Essentials Plus in the UK, and a Data Privacy Framework certification independently verifiable on the US Department of Commerce register. Two qualifications: the supplier list itself was not opened on 31 Aug 2026, so whether it names AI model providers is unestablished; and no data processing addendum, security page or AI-specific disclosure pack is published, so anything a client's AI clause asks about model handling would need to be negotiated.

A firm can answer the three questions a client's AI clause usually asks straight from the public agreement. The model provider is named, Anthropic's Claude, with direct links to that provider's trust centre, commercial terms and privacy policy so the client can read the downstream commitments itself. The training position is stated contractually, that data shared with the provider is not used as broader training data. Payment processors are named as Stripe and Bill.com. A data processing addendum is published and incorporated into the terms where applicable law requires, and a Vanta-hosted trust centre is linked from the primary navigation and every page footer, with SOC 2 Type II and HIPAA claimed. Two qualifications: this is a set of providers named in passing across an agreement rather than a maintained subprocessor list, so a firm cannot be sure it is complete or current; and the data processing addendum and the trust centre contents were not read on 31 Aug 2026, the latter because the portal renders client-side.

Nothing published would help a lawyer disclose or certify AI use to a court. Searched the Sightline, Aurora and AI PrivGen pages and the privacy policy on 31 Aug 2026: no model is identified or versioned, so which system generated a given privilege log description cannot be established; no audit trail or transparency log of AI decisions is described; nothing states that the expert quality assurance step is recorded, sampled at a stated rate, or producible; and no export, template or guidance exists for an AI-use disclosure or a certification of verification. Aurora publishes dashboards and standardised metrics, but they track matter progress and cost rather than AI provenance. The absence bites hardest on the flagship product, since a privilege log is filed with the court and an opponent challenging it would ask how each description was produced and who checked it.

More of the record than most, and no artifact designed to produce it. What a lawyer can establish: which model generated an answer, since the agreement names Anthropic's Claude as the platform Transcript Genius is built on, which is uniquely identifiable in this pull; and what the answer rests on, since every response carries a clickable page-line citation into the transcript, is shown side by side with the source, and where Text to Video Sync is ordered opens the synchronised video at the cited moment. That is a strong verification trail. What is missing is the record itself: nothing states that queries, answers or the acceptance of an answer are logged, no export exists for an AI-use disclosure or a certification of verification, and no template or guidance is published for a court's standing order. One published term undercuts retrospective identification and belongs here: Steno may change the generative AI platform at any time in its sole discretion, so the model behind an answer given months earlier may not be establishable later.