Juro

AI Centrality

How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.

The models are the engine of a core capability on a platform that would function without them. Juro is a full contract lifecycle system first: create from templates, route for approval, negotiate, sign with a native electronic signature built to meet advanced electronic signature requirements, store, and track. Strip out AI Review, AI Extract, AI Assistant and Operator and a working CLM with eSignature and workflow remains, which is a product with its own market. The vendor positions the AI as the differentiator against that baseline, arguing that unlike legacy CLMs its platform does the work rather than accelerating it, and Operator is genuinely model-driven. But the founding product dates from 2016 and the AI features were added on top, which is the B band rather than the A.

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.

Unusually candid about fallibility and entirely unmeasured. The candour is real and it is contractual: clause 3.1 of the published AI terms states that AI Features make mistakes and may produce output that does not accurately reflect real people, places, facts or laws including legal authorities, and the responsible AI page states plainly that generative AI is predictive by nature and can never be 100 per cent accurate. Very few vendors write that into an agreement. Grounding is asserted at the product level, with Operator described as citing its sources when answering questions about the repository. What does not exist anywhere on the pages read on 31 Aug 2026 is measurement: no accuracy figure, no error or hallucination rate, no test set, no benchmark and no published evaluation. The retrieval method behind Operator's citations is not described, and the accuracy mechanisms the vendor does document are input-side aids, the AI playbook for setting context and constraints and prompt shortcuts, rather than measured output quality.

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 oversight commitment is contractual rather than aspirational, which is rare on this axis. Clause 3.1 of the AI terms requires the customer to use human review to identify and correct errors in AI output before using or relying on it for any purpose, and clause 2(e) prohibits misrepresenting AI output as human-generated. The responsible AI page frames the tool as a highly capable trainee lawyer whose output should be checked, and advises using AI features only for tasks the user could complete and verify unaided. Real review surfaces exist in the product rather than being asserted: approval routing is a core module, and the AI playbook lets a customer set context and constraints for AI tasks at template or document level. What is missing is the rest of the control structure. Nothing published states what the agents run unattended in the AI intake and review flow, at what threshold the system stops and escalates, or what happens after an output is wrong.

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.

The deepest customer evidence read in this pull, and still short of the A band on two limbs. Named individuals with named roles, employers and locations each carry a figure: Jessica Zwaan, COO at Talentful, an 86 per cent reduction in time to sign; Victoria Sorving, Chief Legal Officer at Funnel, an 88 per cent reduction in manual contract reviews and 8,000 manual touchpoints removed; Paul Harker, Global Head of Legal at Luno, a 91 per cent time saving per contract; Clio Anderson Garwood, Senior Legal Counsel at Paddle, three and a half hours saved per contract; Pareen Kohlhaas, COO at COOP Careers, five times faster contracting; Chris McFalls at Goldin, 20,000 contracts a year; and Chantelle Zemba, General Counsel at Deliveroo, eight years as a customer. The pricing page adds further figures against named customers including 8,000 dollars saved a year at Tibber and 48 hours a month at Eucalyptus. Missing for an A: no deployment dates and no method behind any figure. Two link errors were noted on 31 Aug 2026, with the Funnel quote linking to a case study for RVU and the ANC quote linking to one for Iptor, and the individual case study pages 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 commitments in the agreement rather than on a marketing page, short of segregation and privilege. Clause 1.2 of the published AI terms prohibits Juro from using customer data as training data for any AI model except a customer-specific model used only to serve that customer, and requires any such model to be erased within 60 days of termination. Clause 1.1 assigns the customer all right title and interest in AI output and treats it as customer data. The model provider is named openly as OpenAI's GPT model provided by Microsoft on Azure servers, and the responsible AI page states that strict confidentiality obligations are in place with both the customer and the technology providers. A data processing agreement and a subprocessor list are published without a gate. Two gaps hold this at B. Nothing published documents how one customer's repository is separated from another's, or how access is enforced between teams inside a customer where sales, HR and procurement all self-serve. And privilege and work product are not addressed directly; the nearest material is a general acknowledgement on the responsible AI page that regulated professionals have confidentiality obligations.

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.

Real material on output reliability, nothing on the advice line, and an audience that makes the gap matter. The published AI terms require human review before reliance and prohibit passing AI output off as human-generated, and the responsible AI page acknowledges that regulated professionals need confidence that AI use will not compromise their confidentiality obligations. But searched the home page, the trust centre index, the responsible AI page, the pricing page and the terms index on 31 Aug 2026 and located no statement that Juro is not a law firm and does not provide legal advice, no ethics or professional responsibility page, no named bar or ethics guidance including ABA Formal Opinion 512, and no jurisdiction limits. The exposure is higher here than for a lawyer-only tool because the product is explicitly sold on the basis that colleagues in sales, HR, procurement and finance self-serve on contracts without legal involvement, and the pricing FAQ says so directly.

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.

A published governance framework with real substance, short of a named owner and any testing results. The responsible AI page, dated 15 January 2026, defines responsible AI as designing systems that align with human values and prevent harm, then names the three harm categories the vendor considers most pertinent to contracting, privacy, confidentiality and accuracy, and sets out what it does about each. The mechanisms are auditable rather than rhetorical: a data protection impact assessment template offered pre-populated with information about the AI Assistant, two published data protection guides covering AI Assistant and AI Extract, an EU AI Act guide in the trust centre written for customers, and the AI playbook as the documented control on context and constraints. What is absent is the bias half of this axis, and the vendor says so itself, listing bias and intellectual property ownership as other challenges it considers less pertinent and does not address. No individual or role is named as accountable for model behaviour and no pre-release testing regime or evaluation result is published.

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.

Substantive published policy across most of this ground. A subprocessor list is published openly at the trust centre and names entities rather than categories: Algolia, AWS and MongoDB for infrastructure, Google Analytics, Heap, Mixpanel, Metabase and Hotjar for analytics, HubSpot, Intercom, Sendgrid and Sumo for communications, Stripe for payments, and Salesforce, Slack and Google as customer-initiated integrations. A data processing agreement is published as a dated downloadable document. The trust centre carries dedicated articles on application security, network and infrastructure security, and application resilience, alongside a public status page and a stated 99.9 per cent platform uptime. Access, portability and erasure rights are set out in plain language. Two gaps, checked 31 Aug 2026. No retention period for contracts, prompts or outputs was located; the only period published anywhere is the 60 days within which a customer-specific model must be erased after termination. And no incident or breach notification practice was located on the pages read, though the application security and infrastructure articles were not opened and may carry it.

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 published position, and the AI-specific part of it is unusually explicit. Juro publishes its Master Services Agreements for US and non-US customers as dated downloadable documents, alongside separate AI terms carrying a last-updated date of 11 March 2025 and a link to the superseded February 2025 version, so a buyer can read the allocation of loss and its history before entering a sales process. Clause 3.2 of the AI terms then does something few vendors do in either direction: it states expressly that any indemnity Juro provides under the MSA relating to non-infringement of intellectual property or other third party rights does not apply to AI output. So there is an indemnity, and the buyer is told plainly that it stops at the AI. Paired with clause 3.1, which puts the duty to identify and correct errors on the customer, the position on wrong output is clear and it is adverse to the buyer. What could not be verified on 31 Aug 2026 is the rest: the Master Services Agreement was not opened, so the liability cap, any warranty and any insurance position were not read, and this grade is rebuttable in either direction on that document.

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.

Real integrations, individually documented, into the systems an in-house team actually works in. Each has its own page: Salesforce, HubSpot, Slack, Microsoft Word, Docusign, Google Drive, Zapier and a REST API, plus MCP connections to Claude and ChatGPT, with Pipedrive, Greenhouse, Workday, SharePoint and Companies House named elsewhere. An integrations index exists. The pricing FAQ adds something implementer-useful and commercially candid, distinguishing out-of-the-box integrations available on all plans, naming Slack, Google Drive and Companies House, from deeper integrations that cost extra, naming Salesforce, HubSpot and Workday. What is missing for an A is depth: the individual integration pages were not opened on 31 Aug 2026, so what each connection actually moves, in which direction, and what a customer must configure was not verified. No document management integration such as iManage or NetDocuments appears, which is consistent with an in-house rather than law firm product.

Deployment Model and Data Residency

Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.

Cloud delivery is implied and neither the tenancy model nor the region is stated on any surface read. AWS appears on the published subprocessor list as infrastructure, and the AI processing location is stated on the responsible AI page as Microsoft Azure servers, which is a real statement about where inference happens. Beyond that, searched the home page, the trust centre index, the responsible AI page, the pricing page and the terms index on 31 Aug 2026 and located no statement of whether the platform is multi-tenant or single-tenant, no region options, no data residency commitment, and no private or on-premises deployment option. Two trust centre articles not opened, on network and infrastructure security and on the GDPR, are the likely home for a residency statement, so this grade is rebuttable on either. The gap is more visible than usual because the vendor states it serves customers in more than 85 countries and publishes an EU Data Act addendum.

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.

The trust centre is genuinely open, which is rarer than it sounds, and the attestation behind it is not evidenced. Juro publishes a trust centre with ten substantive articles readable without a form, an email address or an NDA, covering the UK GDPR, the GDPR, the CCPA, the EU AI Act, the EU Data Act, advanced electronic signatures, application security, network and infrastructure security, application resilience and responsible AI use, filterable by compliance, security and privacy. That is a materially better access route than the gated portals elsewhere in this cohort. SOC 2 Type 2 certification is stated repeatedly, with a badge in the site footer and a dedicated trust centre article described as covering the SOC 2 Type II attestation, and a public status page and a 99.9 per cent uptime figure sit alongside it. What was not located on 31 Aug 2026 is the substance an attestation is judged on: no auditor is named, no coverage period or report date is given, no scope is described, no route to obtain the report is stated, and no penetration test summary or testing partner appears. The infrastructure article was not opened and may name the auditor.

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 supply chain is named to a level most of this market does not reach, without a change commitment. The responsible AI page states that Juro's AI tools are built using OpenAI's GPT model, provided by Microsoft on Azure servers, which identifies the model family, the provider, the delivery route and where inference runs in a single sentence, and the vendor ties the arrangement to how it protects confidentiality. Three gaps, checked 31 Aug 2026. No specific model version is named, and the illustrative discussion on that page refers to GPT-4, which may or may not be what the product runs today. Nothing commits Juro to notifying customers when the provider or model changes. And there is an internal inconsistency worth a buyer's attention: neither OpenAI nor Microsoft appears on the published subprocessor list at the trust centre, which names Algolia, AWS and MongoDB as infrastructure and does not mention the model provider that processes contract content.

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 structure are published without the figure, which is the B band exactly and more than most of this cohort offers. The pricing page states the charging model directly: all plans include unlimited users, unlimited workflows and unlimited templates, and price is driven by monthly contract volume and by the complexity of integrations. The volume bands a buyer is placed into are published, running from fewer than 20 contracts a month through to more than 1,000. Which integrations carry extra cost is named, with Slack, Google Drive and Companies House included and Salesforce, HubSpot and Workday charged. Billing currencies are stated as USD, GBP or EUR, non-profit discounts are offered, and a 20 per cent first-year discount for signing in the month of the demo is published. No figure is reachable: the on-page calculator collects volume, contract types, AI features and integrations and then requires name and email, and the approximate price field rendered as 0 USD when read on 31 Aug 2026. Nothing states what implementation adds.

Firm and Practice Coverage

Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.

Coverage is described with substance on the axis the vendor sells, and the boundaries are left open. Dedicated pages exist for five buying teams, legal, HR, procurement, sales and finance, with legal positioned as the function that controls templates and workflows while the others self-serve within them. Contract types are named concretely on the pricing calculator: NDAs, employment contracts, MSAs, SOWs, order forms, M&A documents and vendor agreements. The vendor states it serves customers in more than 85 countries and gives a usage floor rather than leaving fit vague, saying businesses signing around ten or more contracts a month will see a return. What is absent is the far edge. No law firm segment is addressed, which follows from this being an in-house product but is never stated. Nothing addresses government or court use. And the industry navigation is published but non-functional, with B2B SaaS, food delivery, auto marketplace and travel marketplace all resolving to empty anchors when checked on 31 Aug 2026.