Luminance vs Summize: how they compare in 2026
Luminance and Summize both bring AI to in house contract work, and both are British, but they meet the business in different places. Luminance, founded in Cambridge in 2015 and used by more than 1,000 enterprises, is built around negotiation and portfolio analysis. It proposes markups in Microsoft Word from approved positions and earlier concessions, and Spotlight compares large contract sets with the company's standards. Its own Luna Crescent models work alongside outside models, with a second model checking important answers. Summize, from Manchester, puts the contract system inside Outlook, Teams, Slack, Salesforce, HubSpot and Jira, so requests and questions start where people already work. The pricing unit differs. Luminance quotes on the number of documents held, counting each version, while Summize licenses per user, with renewal increases capped at 10 percent. Luminance offers a SOC 2 Type II report through a trust center and lets customers choose their AWS region. Summize holds ISO 27001 but names no hosting region.
Which one fits
Choose Luminance if
- Negotiation is the bottleneck. Luminance proposes markups in Word from approved positions, earlier concessions and comparable agreements.
- You need answers across thousands of signed contracts. Spotlight connects agreements with their amendments and compares large sets with your standards.
- Your AI review asks how outputs are checked. Important answers are tested by a second model or a critic grounded in the contract, and failed checks go to human review.
- Security wants reports, not badges. Luminance offers its ISO 27001 certificates and SOC 2 Type II report through a trust center on Vanta.
- You buy through AWS. Luminance Corporate is sold on AWS Marketplace as a private offer on a 36 month term, priced by document count.
Choose Summize if
- Your business lives in Teams, Slack and Salesforce. Summize brings requests, contract answers and approvals into those tools, with no separate platform to learn.
- Headcount, not document volume, should set the bill. Summize licenses per user by license type, and renewal increases are capped at 10 percent.
- You want a rollout with a fixed shape. Implementation runs in three to four week sprints, and a full rollout typically completes within twelve weeks.
- Sales needs contract terms inside the CRM. Summize runs inside Salesforce and HubSpot, so a salesperson can check renewal terms without leaving it.
- You want contractual protection on the software itself. The published agreement carries an intellectual property indemnity and a warranty that the software is free of material errors.
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.
Luminance uses models throughout contract interpretation, negotiation and portfolio analysis. Luna Crescent specializes in extraction, classification and other repetitive legal tasks, while additional models handle work that benefits from broader generative capabilities. A routing layer selects the model for each task, and separate validation models test important answers. The repository connects agreements and amendments into context for later questions, and negotiation uses prior concessions and approved positions to generate markups. Workflows turn those findings into contract actions across legal and business teams. The October 2026 release added External Sources, which brings cited research from legal sources, the open web and LexisNexis Protégé into answers, and a Template Agent, with New Lumi taking actions across connected workflows in beta. The central service depends on model generated understanding of contractual language and relationships, rather than a template library alone.
The product comes in three layers, and only the third is AI. A Knowledge Layer holds playbooks and policies. A Contract Operations Layer runs the lifecycle from first request to signature inside existing tools. The agentic AI layer, SIA, is described as surfacing that knowledge as instant answers. Without SIA, a working contract request, repository and operations system remains, embedded in Outlook, Teams, Slack and Salesforce. Summize presents the layers as distinct, each one powering the next.
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.
Luna Crescent is tested against ContractIQ Bench, with 189,000 manually annotated data points covering provisions such as liability caps, termination rights and confidentiality. Testing includes unseen concepts and blind review by legal specialists. Luminance claims 5 percent higher accuracy in blind legal expert tests, without naming what that is measured against, and says Luna is four times faster than general purpose frontier models on contract intelligence tasks. Its test harness measures precision and recall, with no absolute result given. Separate validation checks important outputs against model agreement, source evidence and task requirements. A failed check can lead to a retry and a visible request for human review. The model is also trained to identify missing information rather than fill the gap. Public sample tasks, a reproducible evaluation set and detailed error rates are not available.
SIA is described as giving instant, reliable answers grounded in the customer's own knowledge and standards. That identifies the grounding source as the Knowledge Layer of playbooks and policies, but no retrieval method is described behind it. The security page says Summize evaluates the effectiveness, quality and security of its AI models on a consistent basis. No result, test set, cadence or scope is published for those evaluations. Summize publishes no accuracy figure. Its materials do not discuss hallucination or name a failure mode, and the published performance figures all measure speed and volume rather than correctness.
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.
Business users can generate and progress routine agreements under legal approved templates and role permissions, with nonstandard terms routed to legal. Proposed negotiation language appears in Word. For important AI outputs, a separate model or panel checks agreement, supporting evidence and task requirements. Every required critic check must pass. A failed check prompts another attempt using the feedback, followed by a review flag if the problem persists, which sends a doubtful answer back to a person. Customer decisions and workflow actions retain an audit trail. The AI terms require automatic event logs kept for the life of the agreement, interfaces that support human oversight, and corrective action by Luminance when the system does not conform. The agreement requires manual interpretation of all results. An NDA Negotiation Agent that moves routine negotiations forward automatically is announced and not yet live.
Summize says the business becomes more self sufficient while legal remains in full control. It says questions that used to reach legal's inbox are answered instantly by the people who needed them. Its materials do not say what SIA answers alone and what a lawyer approves, and they state no threshold. The Word review workflow shows failed or missing clauses and lets reviewers accept, modify or reject AI recommendations. That control applies to contract review, while the materials do not explain how a wrong SIA answer is corrected.
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.
Current customer examples tie specific organizations to contracting work. AMD reports finding precedent language in under two minutes, Emma Sleep reports twice as many contracts reviewed per quarter, and Avianca claims 90 percent less document review time. Repository examples include Yokogawa's 99 percent faster response to business questions and IDEXX finding sanctions ties in 20 minutes. Trench Group ran a tariff analysis in two hours, compared with two weeks and about €11,000. Workflow examples include ProSapient cutting its process from eight steps to three, Fluence reporting 50 percent fewer contract turns and Hitachi moving from NDA generation to signature in under five minutes. The homepage claims more than 1.3 million questions answered and 75 percent less repetitive review work, and adds Staples, Imerys and Cryptomathic to the customer logos. These are vendor presented customer results, without a common study protocol or dated measurement period. The current customer base exceeds 1,000 enterprises.
Three named customers give attributed quotes. Steven McGeagh at Huel speaks to ease of adoption, Julia Trius at Edpuzzle to receiving agreements already in the required format, and Derek Ihnen at Boon Edam to reduced review time. Customer logos include Revolut, SeatGeek, Miami Heat, Matillion, Sigma Computing, CodeRabbit, KSE and IPC Systems. Summize cites three times faster contract creation, a 40 percent reduction in deal length, a 50 percent reduction in processing time and six times more contracts reviewed. It puts an NDA at two minutes against two hours. A 4,062 percent ROI is attributed to Nucleus Research, a named analyst firm. None of these figures is tied to a named customer or dated. Huel, Edpuzzle and Boon Edam each also have a case study page.
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.
Each customer has an isolated environment with separate workloads, data and layered access controls. The agreement treats private learning and outputs as customer property and confidential information, and limits support access to service delivery. It also licenses customer data for product improvements, private learning and usage insights; there is no general prohibition on learning from customer content. Private learning stays with that customer, while nonidentifying usage insights can be retained for improvement. Inputs sent to external models are permitted disclosures directed by the customer. No specific clause protects attorney client privilege or work product, and the model list gives no provider retention limits.
Under clause 6.1 of the SaaS terms, Summize may store, transmit and process Customer Data only as needed to provide the services, and gets no other rights in it. Clause 6.2 requires physical, technical and organizational measures in line with good industry practice. Clause 6.5 imposes confidentiality with named exceptions. Clause 6.6 requires daily encrypted backups available to the customer on request, and clause 9.2 requires prompt deletion or return of all Customer Data at the customer's choice on termination. Summize says prompts, completions, embeddings and training data are exclusive to each customer and not available to OpenAI or used to improve OpenAI models. Clause 6.4 reserves the right to use Customer Data in aggregated, anonymized form to improve the software through machine learning, so the agreement permits training. Separation between customers is described as siloed AI, without detail on how it is enforced. Summize's published materials do not address privilege or work product.
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.
The master agreement prohibits treating the product as legal advice or a substitute for relevant independent or internal legal advice. All results require manual interpretation, and customers must tell third party users about those limits. AI Systems Terms add human oversight duties and describe the intended use as legal document analysis, research, contract analytics and legal workflows. The product also serves procurement, finance, sales and other business functions, with exceptions routed to legal. Luminance sets out no lawyer competence procedures and no boundaries for supported jurisdictions. The license permits professional services firms to use the software when serving their own clients.
Summize markets contract answers to people who are not lawyers, in sales, finance, HR and procurement teams. It says questions that used to reach legal's inbox are taken care of, giving the examples of a salesperson checking renewal terms and a CFO querying payment performance without involving legal. Its materials do not say what the output is and is not, and no disclaimer separates information from legal advice. There is no competence or supervision language, and no jurisdiction limit is stated, although customers are in the UK and the US. The SaaS terms contain warranties, indemnities and liability caps but no provision on legal advice. The website terms of use disclaim only the website. The marketing line that legal remains in full control describes workflow, not advice.
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.
AI Systems Terms, version 3.0 of 1 October 2026, require ongoing risk management, records of risks and tests, and data governance that considers bias, representativeness and missing data. Model selection ranks candidates on accuracy, speed, reliability and cost, in that order. Luminance says it has tested more than 30 models through a test harness measuring precision and recall, blind comparisons in Luminance Arena and five validation stages. The evaluation process also includes complete workflow testing through ALumNI, blind expert review and a human sign off before a reversible production rollout. Luna Crescent releases must clear ContractIQ Bench and expert review before replacing a current version. Luminance names no individually accountable AI governance owner and gives no results on uneven performance across matter types, parties or populations.
The security page sets out four AI commitments, each in one sentence: siloed AI practices, regular assessments of effectiveness, quality and security, ongoing updates to AI capabilities, and a customer feedback loop. No one at Summize is named as accountable for AI outcomes. No testing regime before release is described, and no assessment result is published. Nothing addresses bias or uneven output across contract types or counterparties. The company's General Counsel has written an explainer on the EU AI Act, which comments on the regulation rather than describing Summize's own governance. ISO 27001 is an information security standard and does not cover AI governance.
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.
Customers can delete data during use, with removal from active stores within 30 days and backups within a further 30 days; faster deletion can be requested. Routine backups occur every 24 hours, remain within the hosting region and are kept for 14 to 28 days under the master agreement. Access follows least privilege, with multifactor authentication, optional single sign on and isolated customer workloads. The subprocessor list of 1 October 2026 names hosting, model, support and operations services, including Intercom and ClickUp. New providers require 30 days notice and allow 15 days to object. The data agreement requires notice of a personal data breach within 24 hours, mitigation and information about the affected records and likely consequences.
On termination, clause 9.2 of the SaaS terms requires Summize to promptly delete or return all Customer Data at the customer's option. Retention after that is allowed only where law requires, and confidentiality obligations continue over anything retained. Clause 6.6 requires backups no less frequently than daily, secure and encrypted, in a commonly used machine readable format and available to the customer on request. Access control rests on ISO 27001, with DevSecOps practices and internal password, equipment and data confidentiality policies described. Clause 3.2 gives an annual audit right over user and password compliance. Summize states that it runs regular third party penetration testing. The agreement sets no retention period for the term itself, only for its end. No subprocessor list is published. Clause 8.4 refers to Appendix 3 as defining Sub-Processors, but the appendix is not included in the published document.
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.
Version 3.1 of the master agreement, dated 6 October 2026, provides an IP indemnity for the product, with exclusions for unauthorized combinations, continued infringing use and unsupported versions. A performance warranty allows correction, replacement or a refund of unused prepaid fees. General liability is capped at the previous 12 months' fees, rising to three times that amount for specified data obligations. Confidentiality and IP ownership breaches are outside those caps, and certain legally nonexcludable losses remain protected. Claims require notice within nine months. The agreement separately disclaims losses caused by output errors or omissions and treats customer directed model transmissions as permitted disclosures. No insurance commitment is specified.
The SaaS terms give the customer an intellectual property infringement indemnity in clause 7.2, covering claims, liabilities, losses, damages and reasonable costs. Clause 7.3 sets one named carve out, for unapproved combinations. Under clause 7.4, which sets out how claims are handled, Summize may not settle without unconditionally releasing the customer. Each liability cap is the greater of a fixed sum or a multiple of charges, so the fixed sum sets a minimum. Clause 8.3 caps general liability at the greater of £100,000 or 150 percent of total charges. Clause 8.2 sets a separate, higher cap for breach of the security and data protection obligations, at the greater of £500,000 or 500 percent of total charges. Clause 8.1 preserves liability for willful misconduct, and clause 8.4 makes Summize liable for its subcontractors and subprocessors as for itself. Clause 3.6 warrants that the software meets its specification in all material respects and is free of material errors and defects. Clause 3.5 requires reasonable skill and care, clause 3.8 warrants against malicious code, and clause 7.1 warrants that Summize holds the rights to supply the service. The agreement does not disclaim output accuracy, and Summize publishes no insurance position.
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.
Negotiation generates context based markups directly in Microsoft Word using approved language, earlier concessions and comparable agreements. The workflow product connects contract actions, tasks and approvals across business teams and lets other assistants access contract capabilities through MCP. Lumi MCP, in beta since October 2026, brings Luminance into Slack, ChatGPT and Outlook, and External Sources, live since the same release, brings cited research from LexisNexis Protégé and the open web into answers. The platform also names DocuSign connections. Its technical terms allow uploads from document management systems and virtual data rooms. Connector specific configuration, permission inheritance and the exact fields moving in each direction are not explained. These connections bring AI contract work into existing applications while retaining the central agreement record.
Summize works inside Outlook, Microsoft Teams, Slack, Gmail, Microsoft Word, Salesforce, HubSpot and Jira. It adds Summize Sign for e signature and a Claude integration, and each integration has its own dedicated page. The examples given are a salesperson checking renewal terms inside Salesforce and a CFO querying payment performance across the supplier base. Contract context and assistance are available in the tool, without a new login to manage. Summize says there is no separate platform to adopt. Implementation runs in sprints of three to four weeks under a methodology Summize calls HERO, with in house implementation staff. A full rollout typically completes within twelve weeks.
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 standard hosted product runs in an isolated customer environment on AWS. Customers choose an available AWS data center through the order form, with backups in the same region and recovery objectives of 24 hours for a complete primary data center failure. Additional inference can run on AWS, Azure or Google infrastructure. Azure and Google processing follows an EU or UK hosting choice; for other hosting choices the processing sits in an international location that Luminance does not name. Direct OpenAI API processing takes place in the United States, and a second OpenAI entry, for agentic workflows using Codex, is processed in the EU. Luminance publishes neither the full list of available hosting regions nor the routing of each model task, and a storage choice alone does not fix every processing location.
On its home page Summize says data never leaves the customer's environment, and its FAQ says the platform is built on Azure enterprise infrastructure. That names the infrastructure provider, but no region is given and nothing says where customer contract data is stored. The only published hosting location is in the website terms of use, which say the Website is hosted on servers in the United Kingdom. That covers the marketing site, not the software. Summize does not say whether the platform is single tenant or multitenant. Siloed AI, as Summize describes it, covers how prompts and embeddings are handled rather than a tenancy model.
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.
ISO 27001 covers the information security management system for development, sales, support and operation, including the Cambridge technical headquarters, remote workers and cloud processing. Regular SOC 2 Type II examinations sit alongside independent penetration tests. A trust center on Vanta offers the ISO certificates and the SOC 2 report. Contractual measures, in the technical and organizational measures of 1 October 2026, include encrypted storage and transport, nightly operating system updates, daily backups, least privilege access and a dedicated data protection officer. Senior leaders review the security program quarterly, and a security advisory board supports it. Audit information and certificates are also available through the contractual review process. The assurance statements leave out the auditor names, report dates and examination periods.
ISO 27001 certification is held, with a badge naming ISOQAR as the certification body and UKAS as the accrediting body. No certificate number, scope, statement of applicability, or issue and expiry dates are published, and there is no trust portal or route to a report. Penetration testing is described as regular and third party, with no firm or dates named. A security whitepaper by CTO Richard Somerfield is offered behind a form. Summize does not hold SOC 2. It says it audits infrastructure in line with standards including SOC II and intends to work toward SOC II accreditation.
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.
Luna Crescent is a family of proprietary legal models built from a general purpose base model that Luminance does not name, with specialist contract training. It works alongside other models selected by task through benchmarks and expert review, from more than 30 tested. The subprocessor register of 1 October 2026 identifies AWS, Microsoft Azure and Google Cloud for inference inside cloud environments, direct OpenAI API processing in the United States, and a second OpenAI entry for agentic workflows using Codex, processed in the EU. The AI terms distinguish models hosted within infrastructure from requests sent to an external model provider. New subprocessors require advance notice, but Luminance does not name the individual foundation models, their versions or the task routing. No equivalent notice commitment covers every model substitution within an existing provider.
The security page names OpenAI and states that customer prompts, completions, embeddings and training data are not available to OpenAI or used to improve OpenAI models. The FAQ adds that SIA is built on Azure enterprise infrastructure. No specific model or version is named, and Summize publishes no commitment to notify customers when a model or provider changes. No subprocessor register is published. Appendix 3, cited in the agreement as defining Sub-Processors, does not appear in the published text.
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.
Subscription allowances measure the number of files held in the product, and each version counts as a separate document. Annual uploads are subject to a fair use ceiling of twice the contracted document allowance, regardless of later deletion. Fees, payment timing and the initial term are set in the order form rather than a public rate table. Luminance Corporate is also sold through AWS Marketplace as a private offer on a 36 month term, priced by document count. Renewal ordinarily repeats the initial term unless either party gives 90 days notice. The standard increase ceiling is the annual change in the UK Consumer Prices Index, compounded, plus three percentage points, unless the order form says otherwise. Luminance gives no prices for assisted implementation or extra modules.
The pricing structure sits in the published SaaS terms, and the site has no pricing page. The terms define the Order Form as setting out user numbers and the type of software license, so the charge is per user by license type. The initial term renews automatically for twelve month periods unless 90 days' written notice is given. Clause 2.2(b) limits any renewal increase to 10 percent over the previous twelve month term for the same plan, tier and package, unless a higher cap is agreed. Payment is due 30 days from a valid invoice. Fees exclude VAT and sales tax but include other taxes, and expenses need prior written approval. No figure, band, tier name or feature split is published.
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.
Legal, procurement, finance, sales, executive and HR teams use negotiation, repository analysis and workflows across commercial agreements. The license also permits professional services organizations to use the product for their clients and includes terms for United States government customers. The legal suite defines Corporate, Diligence and Discovery applications, while the AI terms describe legal document analysis, research, analytics and related legal workflows. Examples span NDA creation, supplier review, liability positions, tariffs and renewal obligations. Luminance does not set out firm size segments, supported research jurisdictions, language limits or excluded practice areas. Current product positioning extends across an enterprise's contract work.
Five buying teams (Legal, Sales, Finance, HR and Procurement) and eight sectors each have a dedicated page. The sectors are Software, Sports, Finance, Manufacturing, Business Services, Media and Internet, Retail and Telecommunications. Four use cases are published as request, review, repository and analytics. NDAs are named as a specific contract type with a stated handling time. The buyer is a corporate in house function rather than a law firm, and law firms are neither claimed nor excluded. No organization size is stated, and no contract types beyond NDAs are listed as supported. Government use is not addressed, and nothing says where the product stops.
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?
The master agreement licenses customer data for product improvements, private learning, usage insights and outputs. Private learning means recognizing new clauses and contract types from customer training; it belongs to the customer, remains confidential and is not shared across customers. Nonidentifying usage insights may be retained indefinitely for improvement. No general training prohibition overrides those rights. Inputs used with additional model features may also be transmitted to the named provider at the customer's direction.
Clause 6.4 of the published SaaS terms reserves Summize's right to use Customer Data in aggregated, anonymized form to improve the Services or Software through automated decision processing or machine learning analysis. The clause applies despite the security and data protection clauses before it. It names machine learning and applies to Customer Data, and the anonymization qualifier does not remove the permission. The marketing statements concern outside models.
The home page promises a contractual guarantee that data is never used to train external models. The security page says prompts, completions, embeddings and training data are not available to OpenAI or used to improve OpenAI models. Clause 6.4, by contrast, permits Summize's own machine learning on aggregated, anonymized Customer Data. The marketing does not mention clause 6.4, and the clause does not carry the commitment on outside models.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
Customers can delete data at any time. Active data stores remove it within 30 days, with backup removal within a further 30 days and expedited deletion available on request. The master agreement also sets a retention period after the term ends for retrieving and deleting data. Routine backups have a separate 14 to 28 day retention period. No setting guarantees zero retention across application records, logs and external model processing.
Backups are covered by clause 6.6 of the published agreement. It requires Summize to perform and maintain backups of all Customer Data no less frequently than daily, secure and encrypted, in a commonly used machine readable format and available to the customer on request. Clause 9.2 covers the end of the relationship. It requires prompt deletion or return of all Customer Data at the customer's option on termination. Retention after that is permitted only where and for as long as law requires, and confidentiality continues over anything retained.
The agreement does not state how long prompts, outputs or uploaded contracts are held during the term, and Summize describes no configurable retention window.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
Each customer instance isolates workloads and data through logical, technical and access controls. Administrators configure roles, multifactor authentication and single sign on, while support access is limited to service delivery and may follow additional customer policies. The product applies its own access controls around the contract workspace. No current commitment describes inheriting an external document system's permissions at retrieval time or enforcing a law firm's matter level ethical walls.
The security page describes siloed AI practices under which customer prompts, completions, embeddings and training data are available exclusively to that customer and not to other customers. The home page FAQ repeats this as a siloed AI approach that keeps each customer's prompts, data and outputs exclusive to them. Summize does not describe the isolation mechanism or the tenancy model. Beyond the agreement's user and password provisions, nothing says how access is administered. Nothing addresses separation between matters or contract sets inside a single customer.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
The confidentiality clause requires notice of legally required disclosure where lawful, giving the customer a reasonable opportunity to respond. Luminance must cooperate with reasonable efforts to resist, limit or delay disclosure at the customer's expense. A separate clause also requires prompt notice and cooperation when responding to a subpoena or judicial, administrative or arbitral order. No aggregate transparency report is specified.
Clause 6.5 of the SaaS terms requires both parties to keep the other's confidential information, expressly including Customer Data, confidential. Disclosure to a third party is allowed only if required by law or regulation, permitted in writing, or if the information has become public without default. The clause therefore covers compelled disclosure directly. The SaaS terms do not commit Summize to notify the customer of such a request, and reserve no discretion over notice either.
The agreement says a separate confidentiality agreement, if the parties have one, takes precedence over clause 6.5. Summize publishes no such template.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
Contract answers draw on customer agreements, amendments, negotiation history and the platform's standard legal concepts and clauses. Since October 2026, External Sources adds legal sources, the open web and LexisNexis Protégé for cited research. Luna Crescent separately learns from more than 3.4 million proprietary legal data points, with an evaluation set of 189,000 annotated points. The customer retains its data and private learning.
The sources and rights behind the general training collection, and any primary law collection used for legal research, are not identified individually or accompanied by an update schedule.
Summize grounds its answers in the customer's own Knowledge Layer of playbooks and policies and in that customer's contract repository, rather than retrieving primary law. It identifies no external corpus and names no external database, publisher or content license. No jurisdictional coverage is claimed.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
Negotiation and repository answers are grounded in contracts, amendments and earlier agreed positions. Since October 2026, External Sources brings cited research from legal sources, the open web and LexisNexis Protégé into answers, and the AI terms include legal research among the product's intended functions. No named citator, subsequent history calculation or treatment flag accompanies that research. Source grounded validation checks whether a response follows the supplied evidence, which is separate from checking whether a legal authority remains valid.
Primary law sits outside what Summize retrieves. It works on the customer's own contracts, playbooks and policies to answer contract questions and run the contract lifecycle. Checking authority for subsequent history does not arise for a product of this kind, and Summize publishes nothing on it.
Refusal and Uncertainty Behavior
What does the product do when the answer is not in the corpus?
Luna Crescent is trained to identify information that is absent from the contract rather than invent it. Differential Validation separately checks important responses through independent model agreement or a source grounded critic. A failed check can trigger a retry using the identified problem; continued failure results in a review flag. The current explanation describes those behaviors without a public evaluation of their refusal rate or missed unsupported answers.
SIA is described as giving instant, reliable answers grounded in the customer's own knowledge and standards, a claim about the normal case rather than the failure case. Summize does not say what the product does when it cannot ground an answer. It describes no abstention path, no behavior for when no answer is available, and no confidence or grounding measure shown to the user. Its materials, the dedicated security page included, do not discuss hallucination.
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, which records 2,262 court decisions worldwide and was updated on 2 November 2026, has no entry naming Luminance as the source of fabricated legal citations. Luminance's principal workflows concern contracts, but its AI terms also include legal research, and External Sources now brings cited research into answers. The AI Systems Terms require human oversight, and the master agreement assigns the customer responsibility for manually interpreting and using output, including checking its relevance to the particular legal task.
The AI Hallucination Cases database maintained by Damien Charlotin tracks decisions worldwide where a court addressed hallucinated AI content, and records the tool implicated where known. It records no court order, opinion or disciplinary record naming Summize. Summize is a corporate contract intelligence system rather than a litigation tool.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
Professional services customers can identify mandatory requirements imposed on technology providers by their governing body before the order is signed. The master agreement commits Luminance to meet those requirements when compatible with applicable law, and permits regulatory access and assistance in specified circumstances. All output still requires manual interpretation. The agreement does not map these commitments to a named bar opinion or a particular lawyer ethics rule.
An explainer on the EU AI Act comes from Summize's General Counsel. It comments on the regulation and does not engage with bar or ethics guidance about Summize's own product. Its site and terms name no bar association, law society, regulator or ethics opinion. The buyer is a corporate in house function rather than a regulated practitioner in private practice.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
Luminance licenses its product both for internal business work and for services delivered to clients. Current customer examples include faster precedent retrieval, shorter review and fewer workflow steps, while corporate deployments also report operational savings. The contract permits customers to earn revenue from the resulting output. No guidance connects the time saved to a law firm's client invoice, and the workflow audit trail is not presented as a legal time billing record.
Summize claims time and cost savings. It cites three times faster contract creation, a 40 percent reduction in deal length, a 50 percent reduction in processing time, six times more contracts reviewed, and two minutes against two hours for an NDA. It attributes a 4,062 percent return on investment to Nucleus Research. Its materials do not address billing or disclosure, including what happens to a bill when AI assisted work takes less time.
No record of AI assisted work by matter is described. The buyer is an in house department rather than a firm billing a client.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
The subprocessor list of 1 October 2026 identifies AWS, Azure, Google Cloud and OpenAI for hosting or inference, including a second OpenAI entry for agentic workflows using Codex, plus named support and operations providers such as Intercom and ClickUp. It distinguishes processing inside cloud infrastructure from inputs sent directly to OpenAI and states broad processing locations. The data agreement requires 30 days advance notice of a provider change, with 15 days for objections. No separate client consent or outside counsel AI disclosure template accompanies the legal suite.
The public security page names OpenAI as a model provider and says customer prompts, completions, embeddings and training data are not available to OpenAI or used to improve OpenAI models. The home page FAQ adds that the system runs on Azure. Together they say who handles customer content, without a sales conversation. No subprocessor register or forwardable pack for clients is published. Clause 8.4 of the SaaS terms cites Appendix 3 as defining Sub-Processors, but the appendix does not appear in the published document. Summize publishes no data processing agreement, and its security whitepaper sits behind a form.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
Workflows keep an audit trail of actions and decisions, and the AI terms require proportionate event logs retained for the agreement's duration. Important responses can carry a validation result or a request for review, with contract evidence behind the answer. The legal agreement also provides assistance to regulators. Luminance describes no per document export that combines the exact model used, the retrieved sources and a person's verification.
Summize publishes nothing on court disclosure or verification certification. Its materials describe no audit trail or activity export, do not disclose to the customer the model behind a given answer, and mention no record of human verification. The product is a corporate contract intelligence system rather than a litigation tool.
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.
- Good Law Verification
Product updates
The latest logged changes from each vendor, newest first, each linked to its source.
Luminance
-
Product / capability
Luminance released External Sources, which lets users ask Lumi to search curated legal research sources, web search and LexisNexis Protégé and get cited answers inside the contract workflow.
Source
In summary
Luminance
Luminance is a contract platform from Cambridge, founded in 2015, for legal teams and the business functions that work with them. Its negotiation product proposes markups in Microsoft Word using approved positions, earlier concessions and related agreements, Spotlight analyzes the repository, and workflows let business users progress routine contracts within legal's controls. The October 2026 release made cited research from LexisNexis Protégé and the open web live in answers, with Lumi MCP for Slack, ChatGPT and Outlook in beta. Named customers include AMD, Avianca, Yokogawa, Hitachi and IDEXX. According to the AI Legal Index, Luminance's agreement licenses customer data for product improvement and private learning, and each customer runs in an isolated AWS environment.
Summize
Summize is a contract intelligence system from Manchester for in house legal teams, built to work inside Outlook, Teams, Slack, Gmail, Word, Salesforce, HubSpot and Jira. A Knowledge Layer holds playbooks and policies, a Contract Operations Layer runs requests through review to signature with Summize Sign, and SIA agents answer contract questions across the business. Implementation runs in sprints, and a full rollout typically completes within twelve weeks. Named customers include Huel, Edpuzzle and Boon Edam. According to the AI Legal Index, Summize holds ISO 27001 certification from ISOQAR, and its subscription agreement allows aggregated, anonymized customer data to improve the software, while prompts and embeddings are kept from OpenAI.
Questions buyers ask
How does Luminance compare to Summize?
Both serve in house legal teams and the business around them. Luminance centers on negotiation in Word, portfolio analysis and its own Luna Crescent models. Summize centers on reaching people inside Outlook, Teams, Slack, Salesforce and other tools, with playbooks and SIA agents answering contract questions. From the AI Legal Index, based on each vendor's own published materials as of November 16, 2026. No vendor pays for placement.
What do Luminance and Summize cost?
Neither publishes a price. Luminance quotes on a document allowance, counting each version, with annual uploads capped at twice the allowance, and also sells through AWS Marketplace on a 36 month term. Summize licenses per user by license type, renews for twelve month periods and caps renewal increases at 10 percent. From the AI Legal Index, based on each vendor's own published materials as of November 16, 2026. No vendor pays for placement.
Do Luminance and Summize train AI on customer contracts?
Both agreements allow some use of customer data. Luminance's agreement licenses it for product improvements and private learning, which belongs to the customer and is not shared. Summize's agreement allows aggregated, anonymized customer data to improve the software through machine learning, while prompts and embeddings are kept from OpenAI. From the AI Legal Index, based on each vendor's own published materials as of November 16, 2026. No vendor pays for placement.
Where do Luminance and Summize host data?
Luminance runs each customer in an isolated AWS environment in a region chosen through the order form, with some inference on Azure, Google and OpenAI. Summize says its platform is built on Azure enterprise infrastructure and that data never leaves the customer's environment, without naming a region. From the AI Legal Index, based on each vendor's own published materials as of November 16, 2026. No vendor pays for placement.
Which security certifications do Luminance and Summize hold?
Luminance holds ISO 27001 and completes regular SOC 2 Type II examinations, with the certificates and report offered through a trust center on Vanta. Summize holds ISO 27001 certification from ISOQAR under UKAS, with no scope, dates or report route published. From the AI Legal Index, based on each vendor's own published materials as of November 16, 2026. No vendor pays for placement.
Neither vendor paid for inclusion, placement or a grade, and neither reviewed this page before it published. Everything on it comes from public material on the dates shown. How the index grades.
Neither vendor publishes a price. Luminance does not name the base model behind Luna Crescent or list every hosting region. Summize names no hosting region for customer data and publishes no subprocessor list.