Augmetec vs Nuix Neo: how they compare in 2026
Augmetec and Nuix Neo both sell to teams running internal and regulatory investigations, from different starting points. Augmetec's LEIAA is a case command center built around the line of inquiry, the question an investigation is pursuing, with AI that surfaces evidence, flags altered documents, identifies witnesses and drafts interview questions. Nuix Neo starts from processing, turning more than a thousand file types into searchable data and classifying it for review. Nuix Neo sits in the top two bands on ten of fifteen axes and Augmetec on six of fifteen. Nuix leads on outside evidence: a named customer in the Los Angeles County District Attorney's office, ISO 27001 and SOC 2 stated for its review product, a published pricing unit and a bring your own AI option. Augmetec names no customer, displays certification marks without claiming a certification, and names no model. Its counterweight is paperwork. Its agreement and addendum give each customer a dedicated instance, restrict customer content to support unless the customer authorizes more in writing, exclude privileged material from usage data, commit to breach notice within 72 hours, and set out security controls in detail.
At a glance
All 15 axes, side by side
The same grid applied to every vendor in the index, graded from public sources. Hover a grade to see what the letter means on that axis.
AI Centrality
How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.
The models drive several core capabilities and sit on an investigations platform that would work without any of them. The AI functions are named and specific on the platform page as modified in November 2025: identifying relevant evidence and building fact patterns against a Line of Enquiry using what the vendor calls accurate AI algorithms, document fraud detection, automatic identification of witnesses and preparation of interview questions, chronology building, and AI-driven triage, scoping and categorisation of cases. Underneath sits a full case management system that predates the models and delivers a great deal on its own: a centralised datalake, evidence management, reusable templates, customisable workflows, dashboards, collaboration, single sign-on, enterprise integrations and automated evidence bundle creation. Strip the models out and an investigations team still has a working command centre, which is what keeps this off the top band. The master services agreement reinforces the point by never mentioning artificial intelligence in its operative terms at all. Checked 5 September 2026.
The models are the engine of a core capability layered on a product that would still function without them. Nuix Neo's AI classifies and contextualises data, applies language models described as trained for the customer's use case to cut review volume, and drives automated redaction; remove them and the Nuix Engine, processing more than a thousand file types into searchable form, plus early case assessment, review in Nuix Neo Discover and investigation workflows remain, which is the product the company has sold for years. The EULA restricts use to internal ediscovery, investigations and information governance projects and describes the licence by data volume processed, not by AI. Home page, products page, Neo Legal and Discover pages and EULA read 6 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.
Accuracy and defensibility are the central promise and neither is measured. The platform page states that a team can answer questions defensibly, that investigations are conducted accurately and defensibly, and that evidence identification runs on accurate AI algorithms, alongside industry-leading document fraud detection. Not one of those claims carries a figure, a test set, an evaluation, an error rate or a statement of failure modes, and no accuracy or benchmark material appears anywhere on the estate. Nor is grounding described: nothing states whether a finding surfaced against a Line of Enquiry carries a link back to the document and page it came from, which is the control a reader would look for on a product whose output feeds regulatory reporting. The agreement is markedly more candid than the marketing and the gap is the finding: clause 2.5 has the customer acknowledge that the software may contain bugs, errors and other defects, may contain design flaws, and may produce unexpected results from its functionality. A product sold on defensibility and contracted for on an acknowledgement of unexpected results is exactly the marketing-against-agreement shape.
Accuracy is asserted without measurement, and the primary-authority limbs do not apply. The vendor states that its language models improve accuracy and reduce the data to review, and the telecommunications page claims elimination of up to ninety-five per cent of non-relevant data early, with no test set, method or recall figure published; EULA section 10.1 disclaims any warranty of truth or accuracy of data or information generated and makes the licensee responsible for determining whether generated information is accurate. The product classifies and culls documents rather than citing law, so the citation limbs are not held against it; what the band asks of a review tool, a measured statement of how often its classification is wrong, is not published on the surfaces read. Product pages and EULA read 6 September 2026.
Autonomy and Oversight Model
What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.
The level of automation is stated with unusual precision for one feature and no control structure is described anywhere. The platform page says case scoping is semi-automated, that smart actions suggest team actions to progress cases, and that witnesses are identified and interview questions prepared automatically in real time. Semi-automated and suggest are real words about where the machine stops, and they are why this is not the floor band. What is absent is everything around them: nothing states which outputs require review before use, what a reviewer sees, at what point the system acts without a person, what happens when a fraud detection flag is wrong, or whether generated material is marked as generated. No abstention behaviour or confidence signal is described. The agreement allocates the consequence without describing the mechanism, clause 10.2 providing that the customer assumes sole responsibility for results obtained from use of the software and for conclusions drawn from such use. On a product that prepares interview questions and identifies witnesses in a fraud or bribery investigation, the unstated review point is the live question.
A written commitment that the models work alongside reviewing people, with real review surfaces, short of the full control structure. The review product is described as responsive review with AI-enabled workflows that support faster and more precise human review, the bring-your-own-AI option is described by a named customer as running within its own environment under its own compliance controls, and automated workflow templates run culling and redaction that a reviewer then works from. What is not published is the threshold at which a document is excluded from review, what executes without a person, or a stated route back after a misclassification beyond ordinary review. Discover, Neo Legal and Neo pages read 6 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.
No production evidence was located. No customer is named anywhere on the estate, no logo wall appears, no case study exists, and no deployment is described or dated. The single customer voice is one unattributed sentence on the About page, carrying no name, role or organisation, saying that LEIAA is clearly built by practitioners for practitioners. Two figures are published on the platform page and neither is attributed or explained: administrative effort alleviated by up to 45 per cent, and the platform being up to 80 per cent cheaper than other platforms, the second being a competitive pricing claim rather than an outcome and naming no comparator. The founding team's own experience is set out in detail, including more than 150 investigations conducted by the chief executive before founding the company, which is credential rather than deployment evidence and is recorded as such. Searched the home, about, platform and security pages, the master services agreement and the site navigation on 5 September 2026.
A named customer without figures, and figures without a named customer. Donn Hoffman, Chief Privacy Officer and Deputy District Attorney at the Los Angeles County District Attorney's office, is quoted on the on-premises Discover page about implementing bring-your-own-AI within the office's own environment, with no measured outcome; IAV GmbH is named on the same page as an ediscovery services customer. The up-to-ninety-five-per-cent data reduction figure on the telecommunications page is unattributed and undated. Nothing joins a named customer to a figure. Discover on-premises page and telecommunications page read 6 September 2026; customer stories were not opened.
Privilege and Confidentiality Posture
How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.
Substantive contractual commitments, a rare express reference to privilege, and no privilege treatment as such. Clause 8.3 of the published master services agreement is the core: the customer retains full title to Customer Content, and without the customer's consent Augmetec will not access, use or disclose it except as reasonably necessary to support the customer's use of the software, respond to support requests, or for any other purpose authorised by the customer in writing. That is a closed purpose list rather than an open licence. Clause 8.4 permits collection of usage and performance data expressly excluding anything that could identify the customer or its own customers or that is confidential or privileged in nature, which names privilege directly. Clause 6 makes Customer Content the customer's confidential information under a five-year regime, and the addendum requires everyone processing the data to be under a strict duty of confidentiality. Separation is documented rather than claimed, with each customer on a dedicated instance and Annex II stating that segregation procedures prevent other customers from accessing customer data. What is missing is treatment rather than acknowledgement: nothing describes how privileged material is handled once inside, no matter-level walls within a customer are described, and no model provider is named.
Substantive published commitments, structural and contractual, short of the full picture. The deployment is on-premises under the EULA: one copy installed in the licensee's own production environment, so matter data stays with the licensee, and EULA section 3.4 has the licensee own Licensee Data and license it to Nuix solely to provide support. Section 9 treats licensee data as confidential information and sections 9.3 and 9.4 commit, on any legally compelled disclosure, to prompt notice where permitted, assistance in seeking a protective order, and disclosure of no more than counsel advises is required. The bring-your-own-AI option keeps third-party models inside the customer environment. Not located: any statement on training use, since the agreement is silent on machine learning, any description of matter-level segregation inside the review product, and any treatment of privilege or work product. EULA read in full; product pages read 6 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. Where the advice line is not the duty a product raises, the axis is read through the nearest professional duty it does raise: judicial conduct rules and the reviewing duty for products sold only to courts, and the duty to bill for time actually spent for products that draft time entries.
Nothing published addresses the advice line. No statement that Augmetec is not a law firm, no disclaimer that output is not legal advice, no description of the professional judgement required before a finding is acted on, no jurisdiction limit and nothing on supervision or competence was located on the platform page, the about page, the security page or in the master services agreement, which is the natural home for such a clause and contains none. The nearest provision is clause 10.2, under which the customer assumes sole responsibility for results obtained and conclusions drawn, which allocates liability rather than describing what the product is not. The exposure is specific on this product class: the platform builds fact patterns, identifies witnesses, prepares interview questions and produces investigation reports in matters concerning fraud, bribery and improper conduct, where a conclusion carries consequences for named individuals. The terms of use page governing the website was not opened in this pass and is the rebuttal route.
The advice line is not drawn and the audience is broad. The EULA confines use to the licensee's internal ediscovery, investigations and information governance projects and makes the licensee responsible for the accuracy of generated information, but no surface read states that the product does not provide legal advice or how it supports a supervising lawyer's duties; the buyers named span law firms, corporate legal, law enforcement, financial regulators and telecommunications companies. No jurisdiction limit is named beyond export control and the territory clause. EULA and product pages read 6 September 2026.
AI Governance and Bias Disclosure
Published governance over model behavior: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
No governance position of any kind was located. There is no responsible AI page, no principles statement, no accountable owner or function named for model behaviour, no pre-release evaluation regime, no AI management system and no certification such as ISO 42001. Nothing addresses uneven output, and the omission has a sharp edge on this product: a system that identifies which individuals are witnesses, flags documents as fraudulent and builds fact patterns about alleged improper conduct is making determinations about named people, and nothing published describes how those determinations are tested or governed. The governance material that does exist is security governance and is credited on the stewardship row rather than counted twice, comprising an information governance committee and ISMS group, a dedicated person overseeing information security and compliance, and certified security staff. The chief security officer holds a doctorate in machine learning, which is a credential rather than a governance framework. Navigation and footer inventoried 5 September 2026.
Responsible-use language without a published mechanism, testing regime or accountable owner. The review product is marketed as powered by ethical AI, the Neo page describes language models trained to the customer's use case, and the vendor writes about capturing AI productivity responsibly; no governance framework, ISO 42001 or equivalent, pre-release testing description or statement about uneven classification across document types or languages is published on the surfaces read. Product pages read 6 September 2026.
AI Safety and Data Stewardship
Retention, deletion, access control, and what happens to prompts and documents after they are processed. Whether the vendor states its subprocessors and its incident practice, or leaves the buyer to assume.
Among the most completely documented stewardship positions in this corpus, held off the top band by one absence. The published addendum's Annex II runs to several pages of specifics: AES-256 at storage system and storage device level with separately keyed backups, TLS 1.2 or greater in transit, VPN and multi-factor authentication, access control lists on a need-to-know basis with differentiated access profiles, automatic account locking and idle logout, network separation through a DMZ, a web application firewall, logging of access, modification, entry and deletion, intrusion detection, disaster recovery and continuity planning, physical security of data centres described down to visitor logging and video surveillance, an ISMS group and a named accountable security function, internal and external audits, penetration testing with triage and remediation, and a responsible disclosure programme. Incident practice is contractual rather than asserted: clause 2.13 commits to informing the customer of a security incident without undue delay and within 72 hours, with cooperation on the customer's own reporting duties. Deletion is addressed at clause 2.14, destroy or return on written instruction at termination. What is missing is a retention period: nothing states how long data is held while the agreement runs, only that it is retained as required to perform obligations.
Some of the ground is covered, much of it by the deployment model rather than by published policy. Under the EULA the licensee installs the software in its own environment and is solely responsible for backing up and securing Licensee Data, Nuix processes personal information only with prior written consent and then under a DPA that prevails over the licence, and Usage Data sent to the licensing server excludes Licensee Data. No retention period, deletion commitment, sub-processor list or incident-notification practice for hosted deployments was located; the DPA at the URL the EULA names and the 2020 Discover SaaS terms of use were not opened and are the rebuttal route for the hosted product. EULA read in full 6 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.
A real two-sided position published in full, with a cap a buyer should read before signing. Clause 9.2 commits Augmetec to indemnify the customer against direct damages arising from third-party intellectual property infringement claims connected to use of the software, with a remedy ladder at 9.3 covering procurement of continued rights, replacement or modification on stated equivalence conditions. Clause 10.3 preserves liability for death or personal injury, fraud and wilful misconduct outside all caps. Against that, clause 10.1 provides the software as is with the customer assuming all associated liability, and clause 10.4 sets the ceiling at the lesser of one hundred per cent of a contract year's subscription fees or twenty thousand US dollars, with the intellectual property indemnity capped at the lesser of two hundred per cent or fifty thousand dollars. On an investigations platform sold to enterprises, twenty thousand dollars is a low absolute ceiling and it is the operative number for everything except the IP indemnity. The exclusions at 10.4 are also unusually pointed for this product, expressly excluding loss of privilege and loss of confidentiality. Nothing warrants the accuracy of any output.
A real published position on liability, short of the full picture because the vendor gives no indemnity. EULA section 11.2 caps Nuix's aggregate liability at the licence fee paid under the applicable order form in the preceding twelve months, section 11.1 excludes indirect, consequential, and lost-profit claims and bars claims not brought within a year, both carving out intentional misconduct, gross negligence and bodily injury; section 10.3 gives repair, replacement or a prorated refund as the sole remedy for a warranty that cannot be excluded, section 10.1 disclaims accuracy of generated information, and section 11.5 states that Nuix is not an insurer and the licensee looks to its own insurance. Only the licensee indemnifies, under 11.6. A signed direct agreement supersedes the EULA. EULA read in full 6 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.
Integration is claimed as a core proposition and no counterparty is named. The platform page promises core enterprise integrations, single sign-on and secure integrations with existing platforms, document management systems, and Microsoft and Google, which identifies the productivity and identity estates by name and the document management layer only as a category. The agreement contemplates the mechanism at clause 4, covering third-party services the customer connects and single sign-on authentication, and confirms that connecting a service authorises Augmetec to access and store information from it. Nothing published states what data moves between LEIAA and any external system, in which direction, on what trigger, or what a firm must configure, and no API or developer documentation was located; the addendum's data description mentions material from Microsoft Teams, Slack, WhatsApp, Telegram, Amazon Chime and Google Meet, but as categories of personal data that may be uploaded rather than as connectors. A knowledge base exists on a third-party support platform and was not opened in this pass; it is named here as the limit.
Integrations are referred to without documentation an implementer could use on the surfaces read. The Engine's ingestion of more than a thousand file types and connection to data sources is the product's core, the EULA contemplates APIs as ancillary software licensed for use with the software, and the Neo Legal page describes orchestrating business applications across the enterprise, but no integrations page or documentation was opened and no document, matter or review system connection is described with what moves and in which direction. Product pages and EULA read 6 September 2026; the documentation is the rebuttal route.
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.
The tenancy limb is answered more completely than on any other record in this pull, and the region limb is not answered at all. The security page states it plainly under its own heading, your LEIAA, your environment: each customer's instance is hosted on a separate, dedicated environment available only to and accessible by that customer, on a virtual private cloud architecture segregated for data isolation. The agreement repeats it at clause 10.1, that the instance allocated to a customer will be held on separate, dedicated infrastructure, and the software is defined as reached at a customer-specific subdomain of leiaa.com. Annex II adds that customer data is separated on their own instances with segregation procedures preventing access by other customers, and names Google Cloud Platform as the hosting infrastructure. Against that, no region is named anywhere. The virtual private cloud is described as a global network spanning multiple regions, which describes reach rather than residency, no region choice is offered, and no residency commitment appears in the agreement, though the addendum does carry full restricted-transfer machinery through the EU standard contractual clauses and the UK addendum.
The deployment model is stated clearly and residency for hosted deployments is not. Nuix Neo Discover is offered as a native on-premises deployment for organisations with data sovereignty and chain-of-custody requirements, the EULA licenses one copy installed in the licensee's production environment, and the vendor's own description is that the customer keeps complete control over its sensitive information; a hosted route exists through service provider partners and third-party hosting providers under separate terms. For on-premises use residency is wherever the licensee puts it, which answers the question fully; for the hosted route no region, tenancy model or processing location is stated on the surfaces read. Discover on-premises page and EULA read 6 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.
Four certification marks displayed as images, no scope, no date, no auditor, no report, and an agreement that describes something weaker than the badges imply. The security page carries marks for ISO/IEC 27001:2022, HIPAA, GDPR compliance and an AICPA SOC logo, each an image with no accompanying text claiming certification, naming a certifying body or stating a period or scope, and one of the image filenames misspells the statute it refers to. Nothing in the page text asserts that any certification is held. The published agreement is more precise and points lower: Annex II lists regular benchmarking and testing with industry standards, giving ISO 27001, the SANS Top 20 controls and NIST guidelines as examples, and qualifies it as applying on certain systems. Benchmarking against a standard on certain systems is not certification to it, and a buyer reading the badge and then the annex would find two different propositions. A trust page exists at a published subdomain, named twice in the agreement as the location for backup and disaster recovery detail and for subprocessor change notices, and was not opened in this pass; it is the rebuttal route and the cheapest upgrade on this record.
Certification is stated on the product's own surface, short of a report reachable without asking. The Nuix Discover page states that the product holds globally recognised certifications including ISO 27001 and SOC 2. No auditor, coverage period, report type or route to the report is published, and no trust centre was located on the surfaces read; the EULA adds no security warranty beyond the licensee's own obligations. Discover page and EULA read 6 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.
The vendor refers to its algorithms and identifies nothing underneath them. The platform page speaks of accurate AI algorithms and AI-powered analytics, and the addendum's own definition of the services includes the development of future functions, features, tools, algorithms and models, so a model layer is acknowledged as existing. No model is named, no version, no provider and no hosting arrangement for the model layer, and no commitment to notify customers when any of it changes was located. Clause 2.4 concedes the shape of the gap without filling it, stating that the services may use technology and data licensed by Augmetec from third parties, sublicensed to the customer, and naming none of them. One infrastructure provider is named and is not a model: Google Cloud Platform, identified in Annex II as the hosting environment and as the source of the secret management service. The approved subprocessor list at Annex III is referenced in the agreement and fell outside the extraction window of the published PDF, so its contents are unread and nothing is inferred from them in either direction.
The supply chain is partly disclosed. The vendor describes proprietary AI in the Engine and bespoke language models trained for the customer's use case, and a bring-your-own-AI option that lets the customer run its chosen models inside its own environment, which places the third-party model, where it runs and who controls it with the customer; a named customer describes using the latest AI advancements within its own environment on that basis. What is not published is the name of any model Nuix supplies, any provider behind its own models, or a change-notification commitment. Neo page, Discover on-premises page and EULA read 6 September 2026.
Commercial Transparency
Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.
The unit and the commercial structure are established from the published agreement, and no figure appears anywhere. There is no pricing page in the site navigation and every route is a demo request. The agreement supplies the shape: the software is licensed under an Order Form as a Package with stated features and pricing, User Rights itemise the number of user licences and the software instance subscribed to, each authorised user holds a single non-shareable account, and the commitment runs across an Initial Term with defined Renewal Terms. A buyer can therefore establish that this is per-user licensing on a committed term with the package defined in an order form, and cannot establish any rate, band, minimum or implementation cost. One published number is a comparative claim rather than a price and is recorded as such: the platform page states that the product is up to eighty per cent cheaper than other platforms, naming no comparator and no basis. The floor band does not fit, because the unit of charge is published even though the price is not.
The unit and structure are stated in the agreement without the figure. EULA section 5.2 defines a consumption-based licence with a unit of one terabyte of uncompressed data processed per annum, aggregated across instances, expiring unused at the end of each twelve-month term, with quarterly usage reporting and true-up at order-form overage rates; section 7.2 sets automatic twelve-month renewals with ninety days' notice and section 4.1 folds standard support into fixed-term licence fees while premium tiers cost extra. No figure, tier name or price page was located. EULA read in full 6 September 2026.
Firm and Practice Coverage
Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.
The practice is defined with real precision and the buyer boundary is left open. What the product covers is stated repeatedly and consistently: internal and regulatory investigations end to end, from inception through scoping, evidence management, witness interviews and analysis to reporting, extending to ethics, compliance and regulatory reviews. The subject matter is evidenced rather than asserted, with the addendum's own data description naming identity documents, invoices, expense receipts, financial payment data and material from corporate and ephemeral messaging applications, and the founder's stated background covering fraud, money laundering, bribery and corruption. That is a clearer account of the work than most records in this lane manage. Users are addressed as law firms and in-house professionals, and beyond that nothing is stated: no firm or organisation size band, no jurisdiction despite an English-law agreement and a UK base, no statement of which investigation types or regulated sectors the product does not suit, and nothing on languages for a platform ingesting multilingual messaging data.
Segment and coverage are described with substance; the boundaries are partly stated. Primary users are listed on the Neo Legal page as law firms managing large-scale ediscovery, corporate legal teams handling regulatory matters and disputes, financial services firms, compliance officers, investigation teams and security leaders, with separate solution lines for investigations, data privacy and industry pages for telecommunications, integrity agencies and financial regulators; the Engine's coverage of more than a thousand file types is stated. The EULA's territory clause and the on-premises option are stated boundaries; no matter type or practice area is named as unsupported. Product pages and EULA read 6 September 2026.
The 12 legal signals, side by side
Recorded rather than graded. These are the questions a practitioner has to answer before a tool touches a client matter, and the answers are taken from public material only.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
Use of customer content is contractually closed except with written authorization, which is the opt-in shape rather than an outright prohibition. Clause 8.3 of the published master services agreement provides that without the customer's consent Augmetec will not access, use or disclose Customer Content except as reasonably necessary to support use of the software, respond to support requests, or for any other purpose authorized by the customer in writing.
Training falls outside the first two limbs and would therefore require written authorization. Clause 8.4 reinforces it, permitting collection of usage and performance data only where that information could not identify the customer or its customers and is not confidential or privileged in nature, and the addendum adds at clause 2.3 that in no event shall Augmetec process the data for its own purposes or those of any third party.
One tension inside the same document is recorded rather than smoothed, because a careful reader will find it: Annex I lists the activities relevant to the transferred data as including development and enhancement of the services, expressly naming the research and development of additional algorithms and models, and the addendum defines the services to include continuous development of future models.
The published agreement does not name training either way. EULA section 3.4 has the licensee own Licensee Data and grants Nuix a license to use, reproduce, store and process it solely to provide support, and section 3.3 grants a broad license over Usage Data, which is defined to exclude Licensee Data, for development and improvement of the software. A license confined to support does not permit training, but no clause names machine learning or model training, and the vendor's product pages describe language models trained for the customer's use case without saying on what. The deployment is on-premises, so Licensee Data does not ordinarily reach Nuix. Surfaces checked 6 September 2026.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
Retention is addressed in the published addendum and no period is stated. Clause 2.14 provides that Augmetec retains data as required to perform its obligations under the agreement, and that on termination or expiry it shall, on the customer's written instruction, destroy or make available for retrieval all data including all copies in its possession or control and any subcontracted for processing, with an exception where law requires retention, in which case the data is isolated and protected from further processing until deletion is possible.
That is a clear end-state with a customer instruction behind it and no clock attached to the relationship itself. Annex I states the period as the duration of the agreement and refers back to clause 2.14. Nothing separately addresses how long generated material persists, including chronologies, fact patterns, prepared interview questions or fraud detection outputs. Annex II records that published retention policies exist providing guidance on how long data should be kept accessible; those policies were not located on the estate.
No located public material addresses how long prompts or outputs are retained, and under the EULA the question mostly does not arise: the software is installed in the licensee's own environment, the licensee is solely responsible for backing up and securing Licensee Data, and Nuix receives only Usage Data, which excludes Licensee Data, through the licensing server. For hosted deployments through partners, the 2020 Discover SaaS terms of use and the DPA were not opened and are the rebuttal route. EULA read 6 September 2026.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
A separation model is documented at architecture level rather than asserted as a claim. Annex II to the published addendum states that customer data is separated on their own instances, that access is available only to authorized users with strong segregation procedures in place, and that these prevent other customers from having access to customer data. The security page describes the same arrangement in product terms, each customer's instance hosted on a separate dedicated environment accessible only to that customer under a virtual private cloud architecture, and clause 10.1 of the agreement confirms dedicated infrastructure, with the software reached at a customer-specific subdomain.
Inside a customer, Annex II adds differentiated rights by security group and access control list on a need-to-know basis, with logging of access, modification, entry and deletion. What is not addressed is the level a conflicted matter would need: nothing describes walls between investigation teams within one organization, which is a live question where an investigation concerns the conduct of colleagues who may also be users.
No located public material describes how matters or cases are segregated inside the review product. The on-premises deployment gives each licensee its own installation, which is separation between organizations rather than between matters, and the product documentation that would describe case-level security in Nuix Neo Discover was not opened. Product pages and EULA checked 6 September 2026.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
A notice commitment exists in the published addendum, and the main agreement's compelled-disclosure clause runs the other way, which a buyer should see. Clause 2.10 of the addendum requires Augmetec to provide reasonable and timely assistance in responding to any request from a data subject and to any other correspondence, inquiry or complaint received from a data subject, regulator or other third party in connection with the processing, and provides that where any such request is made directly to Augmetec it shall promptly inform the customer with full details.
That reaches requests from regulators and third parties about customer data. The asymmetry is recorded because it is striking: clause 6.2 of the main agreement, headed mandatory disclosure, imposes the notice obligation on the customer, requiring the customer to give Augmetec as much notice as possible before disclosing Augmetec's confidential information under compulsion, and no reciprocal clause obliges Augmetec to notify before disclosing the customer's. No transparency report is published.
The published agreement commits to notice. EULA section 9.3 requires a party legally compelled to disclose the other's confidential information to give prompt written notice if legally permitted so that a protective order or other remedy can be sought, and to provide reasonable assistance in opposing the disclosure; section 9.4 limits any disclosure that remains required to the portion counsel advises is legally required, with commercially reasonable efforts to obtain confidential treatment.
Licensee Data is confidential information under section 1.5. No transparency report is published. Surfaces checked 6 September 2026.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
No corpus stands behind the product's output and the question does not bite on this product class. The material the models work on is the organization's own investigation dataset, described in the addendum as documents, contracts, invoices, expense receipts, identity documents, financial payment data and messages from corporate and ephemeral messaging applications, all uploaded by the customer. There is no case law database, statutory source, publisher or licensed reference set behind a finding, and the historical analysis feature compares against the customer's own previous investigations rather than any external body.
One clause is recorded because it is the only reference to licensed external material and names nothing: clause 2.4 states that the services may use technology and data licensed by Augmetec from third parties, sublicensed to the customer, without identifying any provider or the nature of the data. Searched the platform page, the about and security pages and the master services agreement including both annexes reached on 5 September 2026.
No located public material identifies a legal corpus behind the product's answers, and the product is not built on one: it processes and classifies the customer's own collected data and cites no law. The vendor describes language models trained for the customer's use case without stating the training material. Product pages checked 6 September 2026.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
Nothing addresses checking authority for subsequent history, and the product neither retrieves nor cites primary law. Its outputs are evidence identifications, fact patterns, chronologies, interview questions, fraud detection flags, investigation plans and reports, all drawn from the customer's own material. The nearest adjacent question is regulatory currency rather than case law, since the platform is sold for regulatory investigations and reviews and a regulatory obligation can change, and nothing published describes whether any regulatory content is maintained or checked; that is recorded here so a reader sees it was considered rather than missed.
The value is the honest absence rather than a finding against the vendor. Searched the platform, about and security pages and the published agreement on 5 September 2026.
No located public material addresses whether authority is checked for subsequent history, and the product does not retrieve or cite primary law; its output is processed, classified and redacted evidence for review. Recorded as the honest value for a product without a citator function. Surfaces checked 6 September 2026.
Refusal and Uncertainty Behavior
What does the product do when the answer is not in the corpus?
No material describes what the system does when it cannot reach a reliable output, and the one candid statement on the subject is a liability acknowledgment rather than a behavior. Clause 2.5 of the agreement has the customer acknowledge that the software may contain bugs, errors and other defects affecting performance, may contain design flaws or other issues, and may result in unexpected results generated from its functionality.
That is an unusually direct admission for a published agreement and it describes the possibility of failure rather than how the system signals or handles it. Nothing states whether a fraud detection flag carries a confidence level, whether an evidence identification against a Line of Inquiry can return nothing, what happens when a document is unreadable or a fact pattern cannot be built, or whether generated interview questions are marked as machine-suggested.
No abstention path or confidence indicator is described anywhere. Searched the platform page, the security page and the published agreement on 5 September 2026.
No located public material describes what the classification or review models do when they cannot classify a document with confidence. The vendor describes AI that zeroes in on relevant data and identifies gaps or inconsistencies in collections, which is a detection claim rather than an abstention path, and no confidence signal or no-answer behavior is described. Product pages checked 6 September 2026.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
The AI Hallucination Cases database maintained by Damien Charlotin was searched on 5 September 2026 on the product name LEIAA and on the corporate name Augmetec. No court order, opinion or disciplinary record naming the product or the company was located. One point of context is recorded rather than left implicit: this tracker records hallucinated content in court filings, and the product's primary output is an internal or regulatory investigation report rather than a filing, so a negative result here covers less of this product's risk surface than it would for a drafting or research tool. This records the state of the public record on that date and is not a finding about the product.
No court order, opinion or disciplinary record naming Nuix, Nuix Neo or Nuix Discover was located as of 6 September 2026. The AI Hallucination Cases database maintained by Damien Charlotin was searched on the company and product names together with a general search for court findings; results returned sanctions involving general-purpose chatbots and industry commentary, none of which names this product. This is a statement about the public record, not a finding about the product; a processing and review platform that cites no authority carries a remote exposure on this signal.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
No located material engages with bar or ethics guidance at any level. No bar association, law society, rule of professional conduct, ethics opinion or regulator guidance for lawyers is named or referred to in general terms, and nothing addresses the professional obligations of a solicitor or in-house counsel conducting an investigation through the platform. Nor is there any adjacent statement of the sort that usually accompanies one, since no advice disclaimer and no statement that the company is not a law firm was located either.
The regulatory material the vendor does engage with is extensive and sits in a different field, covering UK and EU data protection law, standard contractual clauses and the UK international data transfer addendum. That is notable on a product built and sold by an investigations lawyer to law firms and in-house legal functions. Searched the platform, about and security pages and the published agreement on 5 September 2026.
No located public material engages with bar or ethics guidance. The EULA confines use to internal ediscovery and investigations projects and the marketing describes ethical AI, but no ethics opinion, bar rule or professional responsibility framework is named on any surface read. EULA and product pages checked 6 September 2026.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
Efficiency and cost claims are published with figures and nothing addresses the billing consequence. The platform page states that administrative effort is alleviated by up to 45 percent, that fact-finding timescales are reduced, that time-consuming tasks are automated to free time for analysis and strategy, and that the platform is up to 80 percent cheaper than other platforms. All of that is directed at the buyer's own cost and capacity.
Nothing reaches this signal: no per-matter record of AI-assisted work is described as available, no guidance on fee or disclosure treatment is published, and nothing addresses what a client is told when an investigation report supporting a regulatory response was produced with automated evidence identification and machine-prepared interview questions. The direction is worth recording because one of the two named buyer types is law firms conducting investigations for clients, so the compression falls directly on billable investigative work.
Law firms and legal service providers are named buyer segments, and the published position on the bill is a savings claim: AI-driven data reduction is said to eliminate up to ninety-five percent of non-relevant data early, saving time and review resources, and the consumption license is priced per terabyte processed. Nothing addresses how AI-assisted review is recorded on a client's bill or what a firm has to disclose when AI culls the review set; the product keeps processing and review audit records for defensibility rather than fee purposes. Surfaces checked 6 September 2026.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
A published subprocessor regime with real customer rights, and no model provider named in anything read. Clause 2.9 of the addendum requires that Augmetec not subcontract processing without consent, gives at least thirty days prior notice of any addition or removal with details of the processing, publishes those changes at a stated trust subdomain, imposes equivalent data protection terms on subprocessors with the customer as third-party beneficiary, keeps Augmetec fully liable for subprocessor acts, and gives the customer a right to object on reasonable grounds with termination without penalty if no solution is reached.
A list of approved subprocessors is stated to be attached at Annex III. Two limits belong on the record. The Annex III list fell outside the extraction window of the published PDF on 5 September 2026, so no subprocessor is named in this note and none is credited. And no artificial intelligence or model provider is identified anywhere in the material read, so a firm could forward a strong governance regime and still not tell a client whose model processed its investigation.
No sub-processor list, model provider list or client-facing AI disclosure material was located. The EULA names a DPA that prevails over the license and applies where Nuix processes personal information with consent, but that document was not opened; the bring-your-own-AI option places the model choice with the customer, and Nuix's own models are not attributed to any provider. For on-premises use a firm can answer its client's AI clause from its own configuration rather than from the vendor's disclosure. Surfaces checked 6 September 2026; the DPA is the rebuttal route.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
Evidence provenance is well handled and model provenance is not addressed at all. The product is sold on defensibility and the supporting mechanisms are real: automated evidence bundle creation, a centralized datalake holding the whole investigative dataset, notes tagged and stored, chronologies and timelines built from the record, and Annex II committing to logging of data access, modification, entry and deletion together with authentication logging and monitored system access.
A team could therefore show what evidence was held, who touched it and when. None of that reaches this signal, which asks for a record of the AI's own contribution: nothing identifies which model or version produced an evidence identification, a fact pattern or a fraud detection flag, nothing marks generated interview questions or report content as machine-produced, and no export is described for producing such a record to a regulator, a court or a client. No disclosure template or guidance is published, and the agreement does not address it.
Some elements of a disclosure record are available and no export of an AI verification record is described. The Neo Legal page states that chain of custody is designed in for litigation and regulatory matters, the on-premises Discover page describes defensible review and reporting, and the platform's processing produces forensic audit records; that is a record of what was collected and processed and by whom. Nothing states that a per-document record of the model used, its classification decision and the human verification can be exported for a court or opposing party. Product pages checked 6 September 2026.
The questions both sides leave open
Derived from the records above rather than written, so it cannot favor either vendor. Take these into both conversations and ask each side the same question.
- Primary Law Corpus Provenance
- Good Law Verification
- Refusal and Uncertainty Behavior
- Bar Guidance Alignment
Which one fits
Choose Augmetec if
- You run investigations, not only document review. Augmetec's LEIAA organizes a matter around each line of inquiry, surfaces relevant evidence and fact patterns against it, flags altered documents, identifies witnesses, prepares interview questions, builds chronologies and assembles evidence bundles, from scoping through the final report.
- You want your investigation data on infrastructure no other customer shares. Augmetec gives each customer a dedicated instance on its own subdomain inside a virtual private cloud on Google Cloud Platform, and its addendum lists AES 256 encryption, TLS 1.2 or greater, multi factor access, logging and a 72 hour breach notice commitment.
- You want your content used only to serve you. Augmetec's agreement bars it from accessing, using or disclosing customer content except to support your use, answer support requests, or for purposes you authorize in writing, and excludes anything confidential or privileged from the usage data it collects.
Choose Nuix Neo if
- Your investigations start with messy data in many formats. The Nuix Engine processes more than a thousand file types of unstructured data into searchable form, and Nuix Neo classifies and contextualizes it, culls non relevant material and automates redaction before review in Nuix Neo Discover.
- You want evidence that public bodies run it. The Los Angeles County District Attorney's office is named as a customer, with its chief privacy officer describing running its own AI inside its own environment, and Nuix sells separate solution lines to financial regulators, integrity agencies and law enforcement.
- You want to budget by data volume rather than by seat. Nuix defines its license unit as one terabyte of uncompressed data processed per year, reported quarterly with true up at order form rates. Augmetec licenses per named user on a committed term, and neither publishes a figure.
In summary
Augmetec
Augmetec, based in London, builds LEIAA, a command center for internal and regulatory investigations sold to law firms and in house legal, ethics and compliance teams. It holds a whole investigation in one place, from scoping and evidence management through interviews and analysis to the final report, with AI that surfaces evidence against each line of inquiry, detects document fraud, identifies witnesses, drafts interview questions and builds chronologies. The AI Legal Index grades it in the top two bands on six of fifteen capability axes. Its published agreement and data processing addendum give each customer a dedicated instance, restrict use of content to support, and commit to breach notice within 72 hours. As of 5 September 2026 the index located no named customer, no named model and no AI governance position.
Nuix Neo
Nuix Neo is the investigative analytics and ediscovery platform of Nuix Limited, an ASX listed company in Sydney, sold for investigations, fraud, data privacy and ediscovery to regulators, government agencies, law enforcement, law firms and corporate legal teams. The Nuix Engine processes more than a thousand file types, and the AI layer classifies data, culls non relevant material and automates redaction, with customers able to run their own models on premises. The AI Legal Index grades it in the top two bands on ten of fifteen capability axes. Its license agreement leaves data with the licensee and caps Nuix's liability at twelve months of fees without a vendor indemnity. As of 6 September 2026 the index located no accuracy figure, no AI governance framework and no named model of its own.
Questions buyers ask
Augmetec vs Nuix Neo: which is better for internal investigations?
On published evidence Nuix Neo sits in the top two bands on ten of fifteen AI Legal Index capability axes and Augmetec on six of fifteen, because Nuix publishes a named customer, stated certifications and a pricing unit. The products fit different jobs, though. Augmetec manages the investigation itself, from lines of inquiry to interviews and the final report. Nuix processes and classifies large volumes of evidence for review. Augmetec's published agreement and security annex say more about how customer data is held.
Does Augmetec use investigation data to train AI?
Augmetec's agreement says it will not access, use or disclose customer content except to support the customer's use of the software, answer support requests, or for a purpose the customer authorizes in writing, so training would need written authorization. Usage data it collects must exclude anything confidential or privileged. The same addendum lists research and development of additional algorithms and models among the activities relevant to customer data, without reconciling that with the use restriction. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 25, 2026. No vendor pays for placement.
Does Nuix Neo have named government customers?
Yes. The Los Angeles County District Attorney's office is named on Nuix's on premises Discover page, where its chief privacy officer and deputy district attorney describes running AI within the office's own environment. Nuix sells separate solution lines for financial regulators, integrity agencies and law enforcement. No figure is attached to the named customer. Augmetec names no customer at all, and its only customer voice is an unattributed quote. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 25, 2026. No vendor pays for placement.
How is Augmetec priced compared with Nuix?
Neither publishes a figure. Augmetec's agreement licenses the software per named user on a committed initial term with renewals, with features and price set in an order form, and its site claims it is up to 80 percent cheaper than other platforms without naming a comparison. Nuix licenses by consumption, one terabyte of uncompressed data processed per year, with quarterly reporting, true up at order form rates and twelve month renewals. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 25, 2026. No vendor pays for placement.
What do Augmetec and Nuix Neo both leave unpublished?
Accuracy and the AI's own record. Neither measures how often its evidence identification or classification is wrong, and neither says what the AI does when it cannot reach a reliable answer. Neither names the models it supplies or engages with bar guidance for the lawyers running investigations. Neither offers a record showing which findings a model produced, and neither describes walls between teams inside one customer, which matters when an investigation concerns colleagues who also use the system. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 25, 2026. No vendor pays for placement.
Three readings to weigh. Augmetec's agreement caps its liability at the lesser of one year's fees or $20,000 and expressly excludes loss of privilege and loss of confidentiality from recovery; those are published terms a buyer of an investigations platform should read closely. Its security page shows ISO 27001, HIPAA, GDPR and SOC marks as images without stating that any certification is held, while its addendum describes benchmarking against ISO 27001 on certain systems. The same addendum lists research into new algorithms and models among the activities for customer data. Augmetec was verified on 5 September 2026 and Nuix Neo on 6 September 2026. Neither vendor reviewed this page.
Neither vendor paid for inclusion, placement or a grade, and neither reviewed this page before it published. Everything above comes from public material on the dates shown. How the index grades.