E
Everchron
Everchron is collaborative litigation management software for case teams, operated by Icebox Inc. of Los Angeles and built by founders who practised as litigators at Irell & Manella before entering the market more widely in 2016. The platform centres on a shared workspace for a matter: a dynamic case chronology linking facts, events, documents, issues and people; a master file built for large document sets including multi-district litigation with numerous parties; witness profiles; deposition transcript management and designations; and full-text search across the record.
It integrates with RelativityOne and Relativity Server so that documents, metadata, coding and work product move between platforms. The AI layer is EC:AI, described by the vendor as its proprietary AI and offered as an optional, opt-in component under the customer agreement. It covers agentic chat that plans multi-step work across the matter and returns cited answers; fact extraction that drafts structured facts with dates, headlines, summaries and linked sources for the team to accept or reject; plain-language assisted search; comprehensive page-line transcript summaries; metadata and profile extraction; AI extraction into customer-defined custom fields; topic and custom transcript summaries; and chronology summaries.
Two design commitments sit alongside those features. The scope and control of EC:AI are configurable, with the suite enabled at organisation level, switched on or off matter by matter, and gated by per-user permissions, and a user can see the plan, the sources reviewed and the steps taken for any request. And AI provenance is explicit: content generated by EC:AI is identified as such throughout the matter and the vendor states the distinction carries through when content is exported.
The customer agreement carries a dedicated AI Products section which states in terms that output may contain hallucinations and that the customer must evaluate accuracy including by human review.
Capability grades
All 15 axes, graded from public sources on the date shown. Hover a grade to see what the letter means on that axis.
AI Centrality
How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.
The AI layer is extensive and the agreement itself establishes that it is severable. EC:AI spans eight named capabilities across two groups, Matter Intelligence and Document Intelligence, and the vendor describes it as its proprietary AI suite built for litigation, investigations and arbitration. Against that, the customer agreement defines AI Products as the optional, opt-in components of the Services that utilise artificial intelligence technologies, and section 7.1 begins by stating the customer may opt in.
A subscriber who never opts in still receives the product Everchron built and sold from 2016: the case chronology, the master file, witness profiles, transcript management and designations, full-text search and the Relativity integration. That is the same severability finding the pricing table produced on another record in this pull, arriving here through the contract rather than the price list, and it is a stronger form of the evidence because opt-in status is a term rather than a packaging decision. The AI is a substantial, well-developed layer on a platform that stands without it. Checked 7 September 2026.
Citation Accuracy and Hallucination Disclosure
Whether the vendor publishes measured accuracy on citations and assertions, grounds output to primary sources, and says plainly what its system does when it does not know. Legal has a documented public record of fabricated citations reaching filed briefs, so an untested claim of accuracy is not evidence.
Grounding is documented as a design principle and the limitation is disclosed in the contract, with no measurement anywhere. The vendor devotes a named section to it: EC:AI keeps its work connected to the materials behind it, and answers, extracted facts, search results and summaries link back to the relevant documents and transcript passages so the output is easier to verify. Agentic chat returns clickable citations linking every response to the supporting material, drafted facts carry links to their supporting sources, and transcript summaries are tied to their page-line excerpts and to the transcripts themselves.
The hallucination disclosure is unusually direct because it sits in the agreement rather than a help page: section 7.3 states in terms that use of the AI Products may result in incorrect output that does not accurately reflect reality, that the customer must evaluate the accuracy of any output including by using human review, and that output may contain hallucinations and may be inaccurate. Naming the failure mode in a contract is rarer than naming it in marketing.
What is absent is any figure a reader could test: no accuracy rate, no test set, no evaluation, no error taxonomy and no described retrieval method beyond the citation link. Grounding to primary legal authority does not bite on a fact management product and is counted neither way. Checked 7 September 2026.
Autonomy and Oversight Model
What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.
All four limbs of the top band are published separately, and the record is stated limb by limb because the grade is rare. What the system runs alone: agentic chat plans the work, gathers the right context, reviews the record and carries out as many steps as the question requires before assembling a cited response, and extraction and summarisation run in bulk across uploads and imports. What constrains it: EC:AI is opt-in at the organisation level, configurable matter by matter with administrators able to enable or disable it for each matter, and gated by granular permissions determining which users can access its features; scope is directed per request to the full matter, a filtered set, selected items, specific documents, or related materials such as families and exhibits.
How a lawyer checks it: the user can see the plan, the sources reviewed and the steps taken, with clickable citations linking every response to the supporting material, and generated content is labelled as AI-generated throughout the matter. The route back to human judgement: drafted facts are presented for accept or reject with the vendor stating that the team decides what belongs in the chronology, the user can refine scope and continue with follow-ups while remaining in control of the analysis, and the agreement requires the customer to evaluate accuracy including by human review.
Visible intermediate reasoning is the limb most vendors omit, and publishing the plan and the steps taken is what separates this from a review gate alone. Checked 7 September 2026.
Operational and Outcome Evidence
Named, dated evidence that the product works in production at real firms or legal departments. Case studies with figures and identified customers count. Unattributed testimonials and launch announcements do not.
A named customer without figures, which is the middle band precisely. The vendor publishes a case study on its own blog for Bennett Bigelow and Leedom, a named law firm, and states that since entering the market more widely in 2016 the platform has been adopted by numerous AmLaw firms and litigation boutiques. Vendor material describes its customer base as including AmLaw 100 firms, top litigation boutiques and corporate counsel.
Founder provenance is stated and is relevant to credibility on this axis: the founders practised as litigators at Irell and Manella, an AmLaw 200 firm, before building the product. What is missing is measurement. No figure of any kind is published: no customer count, no adoption number, no time saved, no matter volume and no outcome. The case study is the only named deployment located, and no dated result with a method a reader could assess accompanies it.
Aggregator and directory profiles carrying employee counts and funding status were seen and are excluded as sources under the evidence rule, so nothing from them is graded, and the seed's supplier robustness flag S-R3 concerning undisclosed funding is a supplier screening test rather than an index criterion and bears on nothing here. Checked 7 September 2026.
Privilege and Confidentiality Posture
How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.
Privilege is named expressly in the agreement, which most records fail, and one clause cuts hard the other way. Section 4.1 defines Confidential Information to include information a reasonable person would consider confidential, attorney-client privileged communications, or proprietary information, whether or not marked, and expressly brings customer content, protected health information and personally identifiable information inside it.
Section 4.2 binds both parties not to disclose or use it outside the delivery of the Services and requires protection at least as protective as a party's own confidential information and never less than commercially reasonable care. Section 1.5 leaves the customer owning its content. Access controls are real, with single sign-on via SAML, multi-factor authentication by TOTP application, an organisation settings panel, and matter-level provisioning.
The clause that holds this below the top band is section 4.5, and it belongs on the record rather than in a footnote: Everchron states it acts as a neutral vendor with no duty to review cases for conflicts of interest, that it shall at no time be required to perform a conflict of interest review or seek any waiver, and, in the same clause, that the customer agrees Everchron may have access to its confidential information and the confidential information of other parties using the Services.
On a litigation platform where opposing parties may both be customers, that is a disclosed limit a buyer should weigh. Work product is not named anywhere. Checked 7 September 2026.
UPL and Professional Responsibility Posture
Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point.
A real published position on advice versus tooling, short of the full treatment. Section 4.4 is direct: while the Services support customers in the legal industry, at no time does Everchron work under or at the direction of any counsel using the Services, at no time is there an attorney-client relationship between Everchron and any party, and no information provided in or through the Services is legal advice. Section 4.5 adds a framing that is unusual and useful to a litigator: Everchron provides the Services as an independent contractor and neutral vendor, does not represent any party in any matter supported by the Services, and shall at no time be deemed an expert witness or consultant of any subscriber or party.
That last point matters in litigation, where a vendor's status can itself become discoverable. The supervision dimension is present through the AI section, which requires the customer to evaluate the accuracy of any output including by using human review. What is absent is jurisdiction. Nothing names where the product is intended to be used or what limits apply, no bar or conduct authority is referenced, and no guidance is offered on what a lawyer must still do to discharge competence and supervision duties when an agent runs multi-step work across a matter. Checked 7 September 2026.
AI Governance and Bias Disclosure
Published governance over model behaviour: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
No governance disclosure was located. There is no AI policy, no responsible AI or ethics statement, no governance framework, no bias or fairness discussion, no model evaluation or testing description, no named internal owner for AI decisions, and no alignment claim to any framework such as ISO 42001 or the NIST AI Risk Management Framework. The gap should be read against what the vendor does publish, so that it is not mistaken for general silence.
Everchron publishes a detailed account of how a user controls EC:AI, covering organisation-level opt-in, matter-level enablement, per-user permissions, scope selection, visible plans and steps, and provenance labelling. That is governance of use, and it is credited on autonomy and on data stewardship where it belongs. None of it is governance of the models: nothing describes how they are chosen, assessed before release, monitored in production, or tested for differential performance across the kinds of material a litigation record contains.
Surfaces read on 7 September 2026: the EC:AI page in full, the Terms of Use in full, the privacy notice, the published business associate agreement reference, the homepage and the complete site navigation.
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.
Control over exposure is the strong half and the handling of content inside the AI is the weak one. On the credit side, the AI is off unless chosen: EC:AI is opt-in at organisation level under the product material and, under the agreement, AI Products are defined as optional opt-in components, with administrators able to enable or disable them per matter and permissions determining which users reach them. That is a real containment architecture rather than a promise, and it lets a firm keep a sensitive matter outside the AI entirely.
The content licence in section 1.4.1 is granted for the sole purpose of providing the Services, section 4.6 confines the vendor's own improvement use to de-identified data about usage such as document types and tag counts, and section 1.5 leaves ownership with the customer. Security is described with an independent element: the web application and network infrastructure undergo security audits and testing by an independent security firm, alongside SAML single sign-on and TOTP multi-factor authentication.
What is missing is the AI-specific data account. Nothing states what happens to content while EC:AI processes it, whether it leaves the vendor's environment, whether any provider outside Everchron receives it, or how long anything derived from a request persists. Section 12.1 acknowledges contractors, third-party vendors and hosting partners generically and names none. And the licence in 1.4.1 extends to improving the Services over content itself, which is broader than a pure purpose limitation. Checked 7 September 2026.
AI Liability and Recourse
What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.
Nothing establishing recourse survives the agreement's own drafting. Section 8.2 disclaims all warranties expressly, including any warranty that results obtained from the Services will be accurate or reliable, and provides the Services and output on an as-is and as-available basis. Section 7.3 states that output is provided as is and that Everchron shall not be liable for any damages a customer or third party alleges to incur as a result of or relating to any output.
Section 9.1 goes further than the usual formulation by excluding direct damages, not merely indirect, incidental, special and consequential ones, before capping total cumulative liability at fees paid in the six months preceding the event. There is no uptime commitment, no service credit and no cure period, and the trial offering is stated to carry no service level agreement at all. The one apparent counterweight does not hold on reading: section 10.2 has Everchron indemnify the subscriber for claims arising out of Everchron's breach of its warranties under section 8.2, but section 8.2 is the disclaimer clause and contains no warranties to breach.
That is recorded as what the document says rather than as an inference about intent, and a buyer should read it before relying on the indemnity. Disputes go to mandatory individual arbitration before the American Arbitration Association in Los Angeles with a class action and jury trial waiver, subject to a thirty-day opt-out by email. Checked 7 September 2026.
Practice Systems Integration Depth
How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.
One integration, and it is the one that matters most for this product class. Relativity has a dedicated page in the product navigation and the vendor describes integration with both RelativityOne and Relativity Server, positioning the platform as bridging the gap between e-discovery and litigation management. For a fact management tool that sits downstream of review, Relativity is the practice system a buyer would ask about first, and naming both the cloud and server editions is a meaningful distinction rather than a logo.
An application programming interface exists, established from the agreement rather than from documentation: section 6.3 reserves the right to suspend access for excess use of an API by a user. Single sign-on via SAML supports integration with a firm's identity management. What holds this below the top band is breadth and documentation. No document management system, practice management system, billing system or court filing system is named anywhere, the Relativity integration page was not opened in this pass so the depth of the connection is not established first-party, and no public API documentation was located in the site navigation. The row is named as amendable on that page. Checked 7 September 2026.
Deployment Model and Data Residency
Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.
Neither limb is stated, which is what the band requires, and the one residency-adjacent statement runs against the buyer. Amazon Web Services is named as the infrastructure, with the vendor pointing to AWS's own risk management and recurring compliance assessments. That identifies the host and nothing more. No region, data centre location or residency option is published, no commitment restricts where content sits, and no tenancy model is stated: nothing describes whether the service is single-tenant, pooled or configurable, and the matter-level access provisioning that does exist is an access control rather than a separation architecture.
The privacy notice addresses location directly and in the permissive direction, stating that unless otherwise instructed by the customer, Everchron and its subprocessors may process personal information globally, and may transfer it from the European Economic Area to another country outside it. That is a disclosure, and it tells a buyer that global processing is contemplated rather than constrained. No self-hosted or private deployment option was located.
A status page is published for service availability, which is operational transparency rather than a residency or tenancy commitment. Checked 7 September 2026.
Security Certifications and Trust Center
Independent attestation a buyer can pull without a sales call: SOC 2, ISO 27001, penetration test summaries, a trust center with current reports and named scope rather than a badge image.
Real controls including an independent testing claim, and no certification or portal of any kind. The site has no security page: the full navigation was established across several pages and covers features, solution, in-house, blog, about, careers, press, support and status, with terms, privacy and cookies in the footer. What security material exists sits on the homepage and is specific in one respect that matters: the web application and network infrastructure undergo security audits and testing by an independent security firm.
The firm is not named, no report is published and no date is given, so the claim is real but unverifiable from outside. Alongside it are single sign-on via SAML, multi-factor authentication using a TOTP application, and an organisation settings panel for security configuration. A business associate agreement is published in full for customers handling protected health information, which is a genuine artifact rather than an offer to negotiate one.
What does not exist is attestation: no SOC 2, no ISO 27001, no certification of any kind, no auditor named, no penetration test report and no trust centre or document portal. The AWS compliance material the vendor points to is the hosting provider's and is not credited here. Nothing unsupported is badged, which is why this does not sit lower. Checked 7 September 2026.
Model Supply Chain Disclosure
Which models sit underneath, whose they are, where they run, and whether the vendor commits to telling customers when that changes. A legal buyer inherits every dependency it cannot see.
Nothing about the models is disclosed. No model is named, no version or family is given, no provider is identified, no inference route is described, and no subprocessor list, data processing agreement or trust centre exists where such a list might sit. The only characterisation anywhere is the phrase proprietary AI, used once on the product page. That asserts ownership rather than describing provenance: it does not say whether the models were built in house, adapted from open weights, or licensed and run on Everchron infrastructure, and an unsupported adjective is not a supply chain disclosure.
The agreement acknowledges third parties only in the most general terms, with section 12.1 noting that Everchron may use contractors, third-party vendors and hosting partners to provide the Services and the necessary hardware, software, networking and storage, which is not AI-specific and names no one. Amazon Web Services is named as the host and has been credited on deployment; naming an infrastructure provider says where software runs rather than whose model reads a case file, and counting it twice is the error this axis most often invites.
The contrast within this record is worth stating: the vendor publishes an unusually complete account of how a user controls the AI and nothing at all about what the AI is. Checked 7 September 2026.
Commercial Transparency
Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.
No pricing information is published at any level, including the unit of charge. The absence is a finding about the site rather than a limit on the researcher: the full page inventory was established across several fetches and the navigation carries features, solution, in-house, blog, about, careers, press, support and status, with terms, privacy and cookies in the footer. There is no pricing page, and every commercial route on the site resolves to Schedule a Demo.
The agreement confirms the position rather than relieving it. Section 3.1 conditions access on payment of any applicable fee as further described in Everchron's fee schedule, and the definition of Documentation includes applicable fee schedules, so a fee schedule exists as a contractual artifact and is not published. What the agreement does publish is billing mechanics rather than price: subscriptions are charged automatically to a card on file, monthly fees are payable in advance and renew automatically, annual fees are payable in advance on the anniversary, fees and pricing are subject to change on notice, all fees are non-refundable with no credit for partial periods or downgrades, and a free trial period exists whose length is set by the vendor and is not stated.
None of that tells a buyer what the product costs or how it is metered, and no unit is named anywhere: not per user, not per matter, not per volume. Checked 7 September 2026.
Firm and Practice Coverage
Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.
Two buyer surfaces and real depth in one kind of matter, short of practice-area coverage. The product navigation carries a Solution page and a separate In-House page, so law firms and corporate legal departments are addressed distinctly rather than lumped together, and the vendor describes rolling the platform out to an entire firm or to specific practice areas, which acknowledges the practice-group dimension of a firm sale.
Depth is claimed where it counts for this product: the master file is described as built to handle the largest case files on the most complex matters including multi-district litigation with numerous parties, and the platform is positioned for cases of all sizes in the United States and worldwide. The AI page widens the frame beyond litigation proper, naming litigation, investigations and arbitration as the matter types EC:AI was built for, and collaboration is described as extending to co-counsel, local counsel and experts as well as the firm's own team.
What holds this at B is that coverage is by matter type and buyer type rather than by practice area: no employment, criminal, personal injury or other subject-matter page exists, no jurisdiction breakdown is published, the interface is English only, and the governing law and venue are Californian. The Solution and In-House pages were not opened in this pass and are named as the artifacts that would refine this row. Checked 7 September 2026.
Legal Signals
What each signal meansA signal records what public sources say on the date shown. It is not a grade and it is not a recommendation. Where a signal reads Not addressed, it means the index did not locate the material in public sources on that date, which is a statement about disclosure rather than about the product.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
A published agreement or policy exists and none of it addresses the question either way, or the document that would answer it could not be read and the summary names the retrieval limit. The summary states which shape the silence takes: an improvement right granted that never names training, or no improvement right granted at all.
The agreement is silent on whether Everchron trains on customer content, and the silence is conspicuous because the same agreement addresses training in the other direction. Section 7.2 prohibits the customer from using output to train their own machine learning models. Having raised training as a concept, the agreement says nothing about Everchron's own use of customer content for that purpose: no permission, no prohibition and no opt-out appears anywhere.
Two clauses were tested and neither reaches a permissive value. Section 1.4.1 licenses confidential information and content for the sole purpose of providing the Services, expressly including the right to create, offer, monitor, troubleshoot and improve the Services; improvement of a service is broader than pure purpose limitation but the clause does not name machine learning, training or models, so under the settled test of whether the clause names the thing it does not reach contractual-permitted.
Section 4.6 is narrower and confines the vendor's improvement and development use to de-identified data about usage, giving document types and tag counts as the examples, with that de-identified data owned by Everchron. The product page states that team review decisions help EC:AI better understand the case and deliver more relevant results over time, which describes in-matter learning from accept and reject decisions rather than model training and is recorded here rather than graded as one.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
Retention is acknowledged in public materials with no stated period.
Retention is addressed and no standing period is stated. What is unusually good here is scope: the agreement's definition of Content expressly includes, if applicable, any input prompts provided to the AI Products, so prompts are contractually treated as the customer's content rather than as vendor telemetry, and they inherit the ownership, confidentiality and deletion provisions that attach to everything else the customer uploads.
That is a real disclosure and most records in this corpus do not make it. What follows is bounded only at exit. Section 6.5 provides that on termination Everchron will for a period of not more than sixty days make user data available to the subscriber, that it may maintain user data for a reasonable period afterwards, and that it may at its discretion delete all user data ninety days following termination without notice.
During the life of a subscription no retention period is published for prompts, for outputs, or for the intermediate products of an agentic request such as the plan and the steps taken. Nothing states whether a rejected draft fact persists, whether a discarded summary is retained, or whether anything derived from a request survives the request.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
Segregation is asserted in public materials with no published detail on how it is enforced.
Access control is claimed at the right granularity and the segregation model behind it is not documented, with one clause running directly against the concern this signal exists for. What is claimed: access is provisioned and managed at the matter level, EC:AI can be enabled or disabled per matter, granular permissions determine which users reach its features, an organisation settings panel holds security configuration, and single sign-on and multi-factor authentication govern entry.
Those are the controls a firm would use to keep a conflicted fee earner out of a matter, and they are described rather than merely asserted. What is absent is the model: nothing describes how one customer's data is separated from another's, no tenancy architecture is published, and nothing states whether the AI's retrieval respects the requesting user's permissions when it searches across a matter. Section 4.5 then addresses the adjacent question and answers it in the negative, and a buyer should see it here: Everchron acts as a neutral vendor with no duty to review cases for conflicts, states it shall at no time be required to perform a conflict of interest review or seek any waiver, and has the customer agree that Everchron may have access to its confidential information and to that of other parties using the Services.
On a litigation platform where opposing parties may both be customers, that is disclosed plainly and is the reason this value sits where it does.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
Terms commit to notice where lawfully permitted. No transparency report located.
The commitment is contractual, mutual and specific as to purpose. Section 4.3 provides that where a party is required by law to make a disclosure of confidential information that is otherwise prohibited or constrained by the agreement, that party will provide the owner of the confidential information with prompt written notice, to the extent permitted by law, prior to such disclosure, so that the owner may seek a protective order or other appropriate relief.
Three things make it substantive: notice is due before disclosure rather than after, the stated purpose is to preserve the owner's opportunity to resist, and confidential information is defined in section 4.1 to include customer content and attorney-client privileged communications, so the undertaking reaches the material a litigator cares about. The qualifier to the extent permitted by law is honest rather than evasive, since a sealing order can bar notice outright.
It stops short of the top value because Everchron publishes no transparency report and no statistics on demands received. One counterweight is recorded: the privacy notice, which governs personal information rather than the matter record, describes sharing with law enforcement, regulators and other parties where required by law or subpoena or where Everchron reasonably believes it necessary, without a matching notice undertaking.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
No located public material identifies the corpus behind the product’s answers.
Two questions sit here and only one applies. EC:AI operates on the customer's own matter, being the documents, transcripts, filings and exhibits a case team uploads, and the vendor is explicit that answers are grounded in that record and linked to it. There is no external legal corpus, case law library or licensed third-party content set behind the product, so the sourcing and licensing limb this signal was written for does not bite and is not counted against the vendor.
The limb that does apply is the provenance of the models themselves, and nothing addresses it: the product page calls EC:AI proprietary AI and says no more, so a buyer cannot establish what any underlying model was trained on, whether it was built in house or adapted, or on what rights it rests. That gap is recorded on the model supply chain row where it is graded, and noted here so a reader does not take the floor value as covering both questions.
Surfaces read on 7 September 2026: the EC:AI page in full, the Terms of Use in full, the privacy notice and the site navigation.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
No located public material addresses whether authority is checked for subsequent history.
The product ships no citator and makes no good-law claim, which is the expected position for litigation fact management software. The citations EC:AI produces point to the customer's own documents and transcript passages, not to legal authority, so there is no published case to validate, no treatment signal and no subsequent history to check. Nothing in the product purports to tell a user whether a decision remains good law.
Recorded at the floor because the value set requires a value, with the reason stated here so that a reader does not take the value as a finding against the vendor. Nothing else in this record depends on it.
Refusal and Uncertainty Behaviour
What does the product do when the answer is not in the corpus?
No located public material addresses what the product does when it cannot ground an answer.
Uncertainty is acknowledged in the agreement and no behaviour is described. Section 7.3 is candid about the condition, stating that output may not be unique across users, that use may result in incorrect output that does not accurately reflect reality, and that output may contain hallucinations and may be inaccurate, objectionable or otherwise unsuited to purpose. That tells a user to expect error; it does not say what the system does when it encounters one.
No confidence score, likelihood indicator or reliability signal is described as surfaced anywhere in EC:AI, nothing states when the assistant declines to answer or flags a weak result, and nothing addresses a document it cannot parse or a question the matter record does not support. Two adjacent features are named and set aside because they answer different questions. The visible plan, sources reviewed and steps taken make the process auditable rather than describing behaviour under uncertainty, and are graded on autonomy.
The accept and reject gate on drafted facts applies uniformly rather than varying with the model's own confidence. Surfaces read on 7 September 2026: the EC:AI page in full, the Terms of Use in full and the homepage.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
No court order, opinion or disciplinary record addressing fabricated or hallucinated legal citations produced by this product has been located as of the date shown. This is a statement about the public record on that one subject, not a finding about the product, and this signal is not a litigation history.
No matter naming Everchron or Icebox Inc. was located in the hallucination case tracking maintained by Damien Charlotin or in the sanctions reporting drawn from it, searched on 7 September 2026 on both the trading name and the registered corporate name. The tracked corpus is large and the reporting reviewed names the products involved where they are known, including the rare instances tied to purpose-built legal AI tools rather than to general chatbots.
Everchron appears in none of it. The product class explains why the exposure is structurally low: EC:AI cites to the customer's own uploaded documents and transcript passages rather than generating citations to legal authority, drafted facts pass through an accept or reject gate before entering a chronology, and AI-generated content is labelled as such throughout the matter. Recorded as none located rather than as a positive finding about the vendor.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
No located public material engages with bar or ethics guidance.
No bar or regulator guidance is named, referenced or mapped anywhere, and the agreement's one compliance obligation points somewhere else. Section 8.1 requires the customer to use the Services in accordance with all applicable Regulations, but Regulations is a defined term and the definition is confined to health information and data protection law, naming HIPAA, HITECH and the rules promulgated by the Department of Health and Human Services in the United States, and the GDPR where it applies.
Professional conduct is not within it. The agreement engages the lawyer's position in other ways, disclaiming any attorney-client relationship, stating that Everchron never works under or at the direction of counsel, and describing itself as a neutral vendor that is at no time an expert witness or consultant, but none of that is alignment with guidance on the use of artificial intelligence in practice. Nothing addresses what a firm's own obligations are when an agent plans and executes multi-step work across a matter record, and no jurisdiction-specific treatment exists.
Surfaces read on 7 September 2026: the Terms of Use in full, the EC:AI page in full, the privacy notice and the site navigation.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
Public materials claim time savings without addressing billing or disclosure, and the product sits inside a fee relationship between a lawyer and a client where those savings would change the bill.
Efficiency claims are published and the fee consequence is not addressed. The claims are qualitative rather than quantified: the vendor states its goal is to enable lawyers, paralegals and practice support to spend less time on manual tasks and focus more on high-level substantive work, describes making the path from input to results as quick and simple as possible, and frames EC:AI as solving the blank page problem so teams spend less time assembling a first draft and more time evaluating the record.
No figure accompanies any of them. Nothing addresses what happens to a client bill when transcript review that took a day takes an hour, no per matter record of AI-assisted work is described, and no guidance on fee or disclosure treatment is offered. Two features come closer to the underlying question than most records manage and still do not answer it. EC:AI can be enabled or disabled matter by matter, so a firm could in principle know which matters used AI at all, and AI-generated content is labelled throughout the matter, so the artifacts are distinguishable. Neither is presented as a billing or fee artifact, and no surface connects either to what the client is charged.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
No located public material supports a client side disclosure obligation.
A firm cannot answer a client's AI clause from anything Everchron publishes. No subprocessor list exists anywhere on the site, and there is no data processing agreement, trust centre or document portal where one might be obtained. No model provider is named, and no statement identifies whether any third party processes matter content through the AI at all. Section 12.1 of the agreement acknowledges only that Everchron may use contractors, third-party vendors and hosting partners to provide the Services and the necessary hardware, software, networking and storage, naming none of them and not distinguishing the AI layer from the rest.
The privacy notice states that Everchron and its subprocessors may process personal information globally and may transfer it outside the European Economic Area, which tells a firm that subprocessors exist and global transfer is contemplated without identifying either. Amazon Web Services is named as the host, but infrastructure identifies where software runs rather than whose model reads a client's privileged file and does not satisfy this signal.
No forwardable client-facing material was located. One adjacent artifact is named and set aside because it answers a different question: a business associate agreement is published in full for protected health information, which is a genuine published instrument but addresses HIPAA rather than AI disclosure.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
The product exports a per document record covering model used, sources retrieved and human verification.
This is the strongest position on this signal located in the corpus so far, because the vendor treats provenance as a product feature rather than as a by-product of logging. Under a named heading, Clear AI Provenance, Everchron states that a user will always know which content was generated by EC:AI and which was created by their team, that AI-generated summaries, facts, analysis and other work product are clearly identified throughout the matter, and, decisively for this signal, that the distinction carries through when content is exported.
An exportable record that preserves the mark is what this signal asks for: a firm producing a chronology or an analysis can show which entries a person authored and which the model drafted. Two further features support it. Every AI output links back to the document or transcript passage behind it, so the provenance mark travels with a verifiable source rather than alone. And the agentic workflow exposes the plan, the sources reviewed and the steps taken, so the route to an answer is inspectable rather than opaque.
The agreement adds an obligation pointing the same way: section 7.2 provides that the customer shall not represent that output was human-generated. What is not published is guidance on disclosing AI use to a court, a certification template, or any statement about what an export contains in a form a tribunal would expect, so the vendor supplies the record without telling a firm how to use it.