Checkbox vs Josef: how they compare in 2026

The product is a workflow and intake system with AI applied to two steps inside it. Checkbox began as a no-code automation platform, which is what the 2022 Gartner Digital Markets recognition was for, and the architecture is still a request pipeline: capture, triage, matter management, automation, reporting. Checkbox AI is described as generative AI to streamline intake and workflow automation, with AI-powered triage routing on matter type, expertise, capacity and business unit, and a separate AI legal chatbot answering routine questions. Those are real shipped capabilities and they accelerate steps that already existed. Remove the models and the platform still captures requests from email, Slack, Teams, Jira and Salesforce, still manages matters, still runs no-code workflows for NDAs and approvals, and still reports on cycle time and workload. The subprocessor list confirms the shape: generative AI is supplied by a third party and enabled only where the customer switches the optional features on, so the AI is a layer a buyer elects rather than the thing being bought.

Josef sells three products: Josef Q for AI question and answer, contract automation, and workflow automation. Only the first is driven by models. The contract and workflow modules are template and rules automation that predate the AI layer, and they are what the L’Oréal case study's 66% turnaround figure rests on. Josef Q is not peripheral: it has its own login at q.joseflegal.com, four dedicated product pages and its own named customers at adidas, Bupa and Cityblock Health. Remove it, though, and a working no code document and workflow platform remains. The models drive a core capability layered on a product that would still work without them.

No accuracy figure exists for a product whose AI answers legal questions and routes legal work. The AI legal chatbot is marketed as giving instant legal help, and AI triage decides which lawyer gets which matter, so wrong answers and wrong routing both have consequences. No accuracy rate, benchmark, test set, evaluation, error mode or confidence measure was found on the home page, security page, legal documents or platform navigation, read. Nothing says what the chatbot answers from, whether only the customer's own policy and FAQ content or the model's general knowledge, and no source linking is described. The only related fact is in the subprocessor annex: generative AI is supplied by OpenAI. Checkbox AI's own product page was not opened and may say more.

Grounding is real and specifically documented. The AI controls page says Josef Q is a closed domain system answering only from uploaded content, and the source list shows the exact page, paragraph and clause an answer drew on, so a user can open and check the support. A document processing engine is described as scanning and optimizing uploaded content before indexing. No measured accuracy figure or test set is published. Josef Q retrieves the customer's own policies rather than primary law, so grounding to primary authority and citation status checking are not capabilities it claims. Against that, the audit trail page carries an unqualified “No hallucinations!” claim, which is marketing running ahead of the mechanism, though it sits alongside a described architectural control and a published refusal path.

What the automation does is described, but not what controls it. Requests are captured across channels, sorted and routed by AI on matter type, lawyer expertise, capacity and business unit, and matters are kept up to date automatically. Self service workflows resolve routine requests without a lawyer. The stated aim is for legal teams to focus on work that matters while the system handles the rest. What is missing is the control structure. No review point lets a lawyer confirm or override an AI routing decision. No confidence threshold or escalation path is published for a request the system cannot classify. Nothing says what happens when a matter is misrouted or the chatbot answers a question it should have escalated. Approvals are a workflow feature the customer configures. That puts a person in the process by design, but is not oversight of the model's output.

The control structure is published in detail. The AI controls page documents human moderation of every answer, adding and removing sources in a Moderation tab, a Tuning Lab for tone, length and spelling, and Smart and Follow-up Suggestions that shape what end users ask. The stop rule is published and specific: outside the uploaded content, the tool answers “Sorry, I don’t know.” The audit trail page adds that all user questions and generated answers are tracked and stored. What is missing is a route back to human judgment. No escalation path from an unanswered question to a lawyer is published, and no thresholds or distinct autonomy modes are published. The review surfaces are the strongest part of the published material.

Checkbox names its customers well, but measures little. Six people are quoted with title, employer, legal team size and company size. Jeannine Moran, Director of Legal Operations at Hitachi Digital, describes a single front door serving 40+ countries, a 60+ legal team and 16,000+ employees. Janene Asgeirsson is Chief Legal Officer at Analog Devices. Richard Conway, Deputy General Counsel and Corporate Secretary at Coca-Cola Europacific Partners, leads a 180+ legal team. Jim Gray, VP and Head of Global Legal Operations at SAP, works with a 400+ legal team across 100,000+ employees. Linh Duong, Senior Legal Counsel at Align Technology, and Sam Bailey, Senior Legal Counsel at Air New Zealand, complete the six. One outcome figure is published: Woolworths Group cut matter volume by up to 80 percent. Another story covers NDA automation at Xero. A logo set includes BMW, PepsiCo, Telefonica, Allianz, Deloitte, PwC, Danone and Telstra. The quotes are qualitative, the team and employee counts size the customer rather than the change, only one outcome figure was found, and no case study shows a date. None of the case studies was opened.

Attribution and figures are both present. The L’Oréal case study names Candy Welsh, Legal Counsel, and reports a 66% drop in contract turnaround, one contract moving from over an hour to 20 minutes, more than 100 such contracts a year, and roughly two weeks of work saved annually. Other named deployments carry named officers. Katherine Roseveare, General Counsel at adidas, is named for a deployment across more than 65,000 people, and Claire Nuske, Head of Legal Operations at Bupa, for more than 40 self service tools. Wendy Chow, General Counsel at Cityblock Health, and Sheila Dusseau, Head of Global Legal Operations at Ferring Pharmaceuticals, are also named. No case study carries a date, and the 66% figure comes with no statement of how it was measured or over what period, so neither the method nor the currency can be assessed.

Checkbox's contract position on confidentiality is complete and readable in advance, but it does not expressly address training or privilege. A buyer can read a published data processing addendum before signing. It includes the EU standard contractual clauses, with Checkbox as processor and the customer as controller. Processing is limited to documented instructions and the stated purpose, and onward disclosure needs documented instructions. Staff get access only as strictly necessary, under confidentiality. At the end, all personal data is deleted with certification or returned, at the customer's choice. Retention is the license term plus thirty days unless earlier deletion is requested in writing. The master services agreement carves data protection and privacy out of the ordinary liability cap. Two gaps. No express statement says customer content is not used to train models, though the clauses' purpose limit constrains it without naming it. And privilege and work product are never mentioned, though the product holds legal matters for in house teams.

No customer agreement is published. The only published documents are website terms and a privacy policy dated 15 February 2024, which applies to Josef as a controller rather than to customer content held as a processor. What is published is real but general. Josef's FAQ says customers keep full ownership of their data and Josef acts as custodian, and the security page states AES-256 at rest and TLS 1.3 in transit with server side keys held by Josef. Nothing addresses training on customer content, tenant separation, or privilege and work product. For an in house buyer the key need is separation between tenants, and no separation of any kind is documented. The ownership and encryption commitments can be read before signing, but the commitments that matter most cannot.

Where the tool stops and legal advice begins goes unaddressed, on a product that markets an AI legal chatbot giving instant legal help to business users who are not lawyers. That framing makes the gap matter. The chatbot and self service workflows are designed so employees resolve legal questions without reaching a lawyer, which is exactly when a clear statement of what the tool is and is not would count. No statement says outputs are not legal advice, that no professional relationship arises, or that answers should be confirmed with counsel. No jurisdiction limit or supervision statement appears. Only part of the master services agreement could be read, and the acceptable use policy and EULA were not opened, so a disclaimer may sit in one of those three documents.

The only published disclaimer covers the wrong thing. The website terms say content on the Josef website is for information purposes only, which addresses the marketing site rather than the output of a tool answering policy questions for employees. The product, meanwhile, is described in advice terms. The Josef Q demo panel offers scalable guidance and advice, and the FAQ describes legal aid organizations and boutique firms using Josef to deliver onboarding, wills and guidance tools to members of the public. No jurisdiction limit, statement of who may rely on an answer, or disclosure requirement for consumers is published, and ABA Formal Opinion 512 is not mentioned. The disclaimer exists but does not reach the product, and the uses the vendor advertises to the public are where the advice line matters most.

Governance of the AI in the product goes unaddressed. There is no responsible AI page, AI policy, ethics statement, governance committee, accountable owner, testing regime or bias evaluation anywhere on the site. The navigation and footer were read in full across platform, solutions, resources, partners and company sections. What exists is information security governance, a different subject. It includes a security management system with named policy areas, thirteen listed security policies in the DPA annex, and a named Chief Information and Security Officer. The gap matters because AI routing decides which lawyer gets which matter, and the chatbot answers questions for people who are not lawyers. Whether either performs unevenly across request types or business units is a live question nothing addresses.

No governance position is published. There is no responsible AI page, no named internal owner, no testing regime before release, and no ISO 42001 or equivalent. Nothing addresses whether answer quality varies across topics, document types or populations. The moderation and review controls are real, but they are an oversight mechanism rather than governance, and security controls are a separate subject. The chief executive's published article on context engineering is a design argument, not a governance disclosure.

Checkbox publishes every element in specific, current terms. Retention has a period: personal data is kept for the license term until thirty days after termination, unless the customer asks in writing for earlier deletion. Deletion is contractual and certified. At the end of processing, Checkbox must either delete all personal data and certify it, or return it and delete copies, at the customer's choice. Subprocessors are named with address, contact and role: Amazon Web Services for hosting and storage, Datadog for logs, SendGrid for workflow email deleted automatically after delivery, and OpenAI for generative AI where the customer turns those features on. Changes need fifteen days' written notice, with a right to object. Incidents must be reported without undue delay, with a contact point, the nature of the breach, the people and records affected, likely consequences and measures taken. Access is limited to staff who strictly need it, under confidentiality. Data is encrypted at rest and in transit. That is backed by thirteen named policies, including a Cryptographic Standard, Data Destruction Standard and Security Incident Response Plan, and by annual third party penetration testing.

Access control is documented and the rest is not. The security page states AES-256 encryption at rest, TLS 1.3 in transit, encryption keys held by Josef on the server side and regular access audits, alongside OWASP and ASD secure coding frameworks, automated vulnerability scanning and annual third party penetration testing. What happens to uploaded documents and prompts afterwards is not addressed. The privacy policy keeps service data for the business relationship and an unstated period after, and describes deletion only as securely deleting or anonymizing once no legitimate business need remains. No incident practice is published. The subprocessor list is not published; the privacy policy says it can be provided on request, and no self service portal or download route is published.

Liability is allocated mutually, with a higher cap for the risks that matter most. Each party's liability is capped at the amount paid in the twelve months before the liability arose. That cap does not apply to intellectual property infringement or misappropriation, data protection and privacy obligations, or indemnities. For those, liability runs to a stated multiple of the ordinary cap. Checkbox indemnifies the customer against third party claims that the software or services infringe intellectual property. It may not settle on terms that impose a nonmonetary duty on the customer, or require an admission of fault, without the customer's written consent. The customer's indemnity is narrowed where a claim results from Checkbox's own negligence or breach. Missing: nothing specific to AI, no warranty on output, and no insurance position found. Only part of the agreement could be read.

Nothing is published on who bears the loss when an answer is wrong. The only published liability clause sits in the website terms and excludes liability arising from use of the Josef website, limited, where it cannot be excluded, to resupplying the relevant services or information. That is a website clause, not a position on product output. No master subscription agreement, customer agreement, service level agreement, warranty or indemnity is published, and the site footer offers only the website terms, the privacy policy and a cookie notice.

Each integration is named with what it does. Slack and Microsoft Teams turn messages into matters, so a request raised in chat becomes a tracked item without leaving the channel. Gmail and Outlook capture email and track matters. Salesforce automates contract generation and approvals, the handoff from sales to legal that often breaks. Ironclad syncs contract requests and tracking, so Checkbox works alongside a CLM rather than against it, and Checkbox markets the pairing. Jira is another intake channel. There is a technology and integration partners page and a referral and services partner program. Missing is depth. No API reference, developer documentation or connector specification was found, nothing describes field mapping or what happens when synced records conflict, and nothing lists unsupported systems. The integration pages were not opened.

Thirteen integrations are named and grouped by function, and the direction of data flow is described. Policies sync in from SharePoint and Confluence, data is pulled from Airtable and Google Sheets, and captured information is pushed out to Slack, Teams, SharePoint and Salesforce. The legal systems are present: HighQ, Xakia, Dazychain and Actionstep are all named. The Xakia entry goes furthest, describing Josef bots launched from Xakia's internal client portal so that requests stay managed within Xakia. Two integrations, Actionstep and Xakia, have linked walkthroughs. No documentation an implementer could use is published: no developer index, no field mapping, and no statement of what a firm must configure. The integrations page answers any unlisted case by inviting a conversation with the team.

Customers choose a data region, but the rest of the deployment model is described only in part. The privacy policy says Checkbox mainly stores data in the data center region the customer requests. The same passage says data may sometimes go to third parties in other places for listed purposes, where data protection laws may differ, with transfers made under applicable law. The subprocessor annex names Amazon Web Services for hosting and storage. The EU standard contractual clauses govern transfers, with the Irish Data Protection Commission as supervisory authority. Missing: no tenancy model, so nothing says whether customers share infrastructure or get a separate environment. There is no list of the regions actually available, and nothing on what changes between deployment tiers.

The security page and the FAQ both state data residency options in the US, Europe, the UK and Australia, a commitment to four regions stated consistently in two places. The tenancy model is never stated: nothing says whether customers sit in shared or separate instances. Processing location is not addressed separately from storage, and two documents pull in different directions. The security page offers regional residency, but the privacy policy says personal information may be transferred and processed outside the user's country, including in the United States, relying on standard data protection clauses for EEA and UK transfers. The two are not necessarily inconsistent, since the privacy policy governs Josef as controller rather than the platform tenancy, but no published document reconciles them.

Four standards are claimed by Checkbox with audit language, but none of the confirming detail is public. It says it has been externally audited and completed a SOC 2 Type II examination, holds ISO/IEC 27001:2022 certification, and complies with ISO/IEC 27017:2015 and ISO/IEC 27018:2019, the cloud personal data standard. The DPA repeats all four, putting the claim in a contract rather than only marketing. Its security management system covers network security, information handling, access control, incident response, backups, change management, risk, vulnerability and patch management, and vendor risk. Listed controls include network intrusion detection, microsegmented firewalls, hardened system images and monitoring of all administrative access. Missing: no auditor, certificate date, examination period or scope statement appears anywhere. A trust portal at trust.checkbox.ai was not opened. The security page offers a contact form to request policies, penetration tests and SOC 2 reports.

SOC 2 Type II is stated on the security page, the FAQ and the home page, and the privacy policy states SOC 2 Type I and II. Annual third party penetration testing is stated, as are GDPR compliance and ISO certification of the underlying server infrastructure. Nothing that would let a reader check any of it is published. No auditor, scope or coverage period is named, no report or summary is offered by any route, and no trust center exists. No Vanta, SafeBase or equivalent portal is linked from the footer or navigation. The standard is named in prose rather than only as a badge image, but there is no route to the evidence at all, not even a gated one.

The AI model provider is named in Checkbox's data processing addendum. Annex III lists OpenAI, LLC, with address and privacy contact, and says it provides generative AI for features in Checkbox where the customer turns those optional features on. That tells a buyer whose model is involved, that the AI is optional, and that the provider sits in the processor chain with the same fifteen day change notice and objection right as any other subprocessor. Amazon Web Services is listed separately for infrastructure, so a reader can tell where workloads run apart from whose model reads the content. Missing: no model or model family is named, and there is no version or change notice commitment specific to the model. Nothing says what data goes to the provider, how long it is kept there, or whether zero retention applies.

Josef acknowledges a large language model underneath and never says whose. The chief executive's published position is that an LLM alone cannot be trusted and needs context engineering, and the audit trail page refers to GenAI engineering. No model, model provider or inference location is named. Hosting region and ISO certified infrastructure are published, but a hosting location shows where the software runs rather than whose model sees client content. There is no commitment to notify customers if the underlying model changes. The only route to the supply chain is the subprocessor list the privacy policy says is available on request.

Pricing is absent at every level, down to the unit of charge. The navigation and footer were read in full and include no pricing page. Every commercial route is a demo booking, a guided tour or a contact sales form. No rate, band, tier, package or per seat, per request or per matter unit was found. Nothing says what implementation or onboarding adds, which matters for a platform whose value depends on configuring workflows. The published agreements set billing mechanics without price. Fees are defined in a Checkbox License Order, and the data processing addendum ties its retention period and its own signing to that order. Two pages address growing and enterprise legal teams, which divides the market without pricing it.

No pricing information is published at any level. There is no pricing page, and the navigation and footer of the home page, the Josef Q and integrations pages, the security page, the AI controls page and the FAQ carry no pricing entry. Every commercial path is a demo booking or a contact form. No tier names, packaging structure, unit of charge or figure is published, so nothing shows what is charged for, let alone how much.

The buyer is identified with precision and the boundaries of the product are left open. Checkbox is unambiguous about who it is for, addressing in-house legal teams throughout and publishing separate pages for growing and for enterprise legal teams plus an in-house legal roles page, and the customer evidence bears the positioning out with legal teams ranging from 60 to 400 lawyers at organizations from 10,000 to 100,000 employees. Practice coverage is expressed as request types rather than practice areas, spanning NDA automation, contract lifecycle steps, legal policy and FAQ handling, and general intake, which suits a front-door product. What is thin is everything about limits. No industry or sector segmentation is published despite a customer base spanning banking, pharmaceuticals, retail, aviation, technology and professional services. Nothing states a minimum viable team size, a request volume the product is designed for, or matter types it does not handle, and no jurisdictional scope is given for a platform sold to multinationals. The claim that it suits both growing and enterprise teams is made without describing what differs between them.

Coverage is set out by segment, with named customers in each. The FAQ and customer navigation name five segments: in house legal, global law firms, boutique and NewLaw firms, legal aid organizations and community legal centers, and law schools. Each has its own page and named users. In house users include L’Oréal and Bumble, and firms include Orrick and Herbert Smith Freehills. NewLaw users include Hive Legal and Polaris Lawyers, legal aid users include Housing Court Answers and Everyday Justice, and education users include Cornell, NYU and the University of Pennsylvania. Document types are named too, including NDAs, MSAs, employment agreements, powers of attorney and client onboarding forms, with question and answer uses grouped under commercial, privacy and data, legal operations and HR. The limits are not stated: Josef's FAQ says anything repeatable can be automated, government use is not addressed, and no page says where the product stops.