Product and intended use
Define the configuration, people, information and tasks being assessed.
Product, entity and plan boundary
Public claim only
Checked 2026-09-27. The vendor publishes the product as 'Leah Agentic CLM' with a wider family named Leah Agentic OS, Leah Legal, Leah Procurement, Leah Finance and Leah Contracting. The contracting company named on the legal page, privacy statement and trust portal is ContractPod Technologies Limited (also written ContractPod Technologies Ltd.), 2 Kingdom Street, 2nd Floor, London W2 6BD. The legal page names five document sets: Privacy Statement, Data Processing Addendum, Security Data Sheet, Master Terms and Annexes, and Marketplace Terms.
Ask: Which entity will sign with us, which of Leah Legal, Procurement, Finance and Contracting are on the Order, and do the Marketplace Terms or the Master Terms and Annexes apply to our purchase?
Tenancy model
Not confirmed
The legal page states 'Because of our multi-tenant infrastructure, all customers benefit from the same security architecture and improvements.' The product page FAQ states that Leah 'supports single-tenant deployment for customers with strict data isolation requirements.' Both pages were read 2026-09-27. Neither says which is the default, what a single-tenant deployment costs, or whether it is on every plan. The trust portal states the platform is hosted on Microsoft Azure and that a web application firewall is deployed 'within all environments across all available regions'.
Ask: Is our tenancy single-tenant or multi-tenant, on which plan, and will the Order Form record it together with the Azure region and the isolation boundary for documents, indexes and model calls?
Functions described by the vendor
Public claim only
Product page read 2026-09-27 sets out six areas: Contract Creation (intake forms with conditional logic, clause libraries, auto-population, an Outlook add-in, and 'One-Drop' upload that 'classifies the contract type, extracts key clauses, obligations, and parties, and kicks off the right workflow'); Review and Negotiation (playbook redlining delivering 'Word-ready redlines with tracked changes', an in-document 'Ask Leah' question-and-answer function, a Word add-in, detection of positions previously negotiated with the same counterparty); Approval and Signature (conditional routing on value, risk flags, jurisdiction or contract type, delegation, escalation timers, DocuSign and Adobe Sign send-for-signature with executed documents synced back); Smart Repository (fuzzy, wildcard, proximity and phrase search, obligation tracking with renewal and notice-period reminders, role-based permissions, 'DeepSights' dashboards); Contract Helpdesk (question-and-answer over the contract set where 'Every response links to the specific clause or document it came from', plus 'Leah Discovery' cross-corpus analysis); and Legal Intake (structured request forms, auto-triage by urgency, contract type and team, workload analytics, automated status updates). Nothing on that page was observed in operation.
Ask: Demonstrate One-Drop classification, playbook redlining and Contract Helpdesk on our own contracts, and state which of the six areas are licensed on our plan rather than available at extra cost.
Governing law and Australian coverage
Not confirmed
No governing law, forum or Australian-law option was found. The Master Terms and Annexes were not read within the credit cap; the legal page states the Master Terms adopt the published Bonterms Cloud Terms v1 and Bonterms DPA v1 as a starting point. The privacy statement addresses GDPR and California rights and does not mention the Australian Privacy Act or the Australian Privacy Principles. No Australian entity appears on any page read.
Ask: Will you accept Australian law and a non-exclusive Australian forum, and an Australian Privacy Principles addendum covering notifiable data breaches, or are the Bonterms defaults and your chosen law mandatory?
Check for vendor content directing how AI describes the product
Not confirmed
Checked deliberately on 2026-09-27 across the four pages read (product page, legal index, privacy statement, trust portal) and against a site map of 59 vendor URLs captured the same day. No page or block directing AI models, assistants or LLMs how to describe the product was found; no /ai-info, /for-ai or llms.txt path appeared in the site map, and no 'Instructions for AI Assistants' heading or equivalent wording appeared in the pages read. The vendor does publish an AI Governance page, which was not read within the credit cap and which is a different thing from content telling AI how to describe the product.
Ask: Do you publish, anywhere on your site or in your files, guidance or instructions to AI models, assistants or crawlers about how your products should be described?
Data, controls and requirements
Compare the actual configuration with firm, client and jurisdiction requirements, supported by evidence.
Authority to put client and counterparty material into the product
Not confirmed
Nothing on the pages read addresses the customer's authority to upload third-party or client material, or any obligation to obtain consent. The described workflow brings counterparty documents in through the Outlook add-in, One-Drop upload and the Word add-in, and sends documents out to DocuSign, Adobe Sign, Signicat and Viafirma. The Master Terms, Annexes and DPA, which would carry the customer warranty and any licence over uploaded data, were not read within the credit cap.
Ask: Which clause of the Master Terms and DPA sets out our warranty over uploaded contracts and the licence we grant you over them, and what wording do you accept in an Australian engagement letter to disclose this processing to clients and counterparties?
Foundation-model providers and versions
Not confirmed
No model provider, model family, model version or inference location is named on any of the four pages read, checked 2026-09-27. The trust portal names Microsoft Azure as the hosting cloud and links to Azure infrastructure and physical security documentation, but names no model provider and does not say where inference runs. The product names its own functions (Ask Leah, One-Drop, Leah Discovery, DeepSights) without naming what sits underneath them.
Ask: Name every model family, version and hosting route behind playbook redlining, One-Drop classification, Ask Leah, Contract Helpdesk and Leah Discovery, say in which country each runs, and confirm versions are pinned and changes notified.
Subprocessors for the service
Not confirmed
No subprocessor list for the service was found. The trust portal states that Leah vets and monitors subprocessors and directs the reader to the DPA 'for more information'; the DPA itself was not read within the credit cap. The privacy statement, read 2026-09-27, lists processors for the website only, naming Qualified, Calendly, Cloudflare and Cloudflare Stream, and Videoask, plus analytics and advertising vendors including StatCounter, Microsoft Clarity, Google Analytics, LinkedIn, Yandex, PostHog, Datadog, Marketo, ZoomInfo, 6sense, Pardot, Clearbit and others. Those are website processors and are not evidence about who processes contract data in the service. Microsoft Azure is named as the hosting cloud on the trust portal.
Ask: Provide the dated subprocessor list that will apply to our tenancy, naming each entity, its country, the data it touches and whether it is in the model path, plus the notice period before you add one.
Training on customer content
Public claim only
The trust portal states, read 2026-09-27: 'Leah does not use user data to develop, improve, or train generalized AI and/or ML models.' The product page FAQ states 'Your contract data is never used to train models'. Neither statement is in a contractual document on the pages read; the DPA and Master Terms were not read. The trust portal wording is limited to 'generalized' models and does not address customer-specific tuning, embeddings, retrieval indexes or retention by any model provider.
Ask: Warrant in the Order or DPA that our contracts, prompts, outputs and embeddings are not used to train, tune or evaluate any model of yours or of any model provider, including customer-specific models, and state the retention position at each provider.
Use of feedback, ratings and corrections
Not confirmed
Nothing on the pages read addresses in-product ratings, thumbs-up or thumbs-down signals, or the use of user corrections to redlines and extractions. The privacy statement covers website analytics and advertising only. The product page describes users accepting tracked changes in Word and reviewing extracted values, which produces correction data, but says nothing about what happens to it.
Ask: When a user edits a Leah redline or corrects an extracted value, where does that correction go, who can see it, and can the feedback signal be confined to our tenant or turned off?
Retention of documents, prompts and logs
Not confirmed
The trust portal, read 2026-09-27, states that 'strict data retention and deletion policies' exist to manage the lifecycle of sensitive information and that logs which track access, use, modification or deletion of collected data are ingested into a SIEM and 'retained for forensic purposes'. No period is given for contract documents, prompts, model outputs, indexes, backups or logs. The privacy statement gives no period either, stating only that data is kept for as long as necessary for the purpose. The DPA was not read within the credit cap.
Ask: State in the Order the retention period for contract documents, Ask Leah prompts and outputs, extraction results, search indexes, application logs and backups, and whether any of those are configurable by us.
Deletion on instruction and on exit
Not confirmed
The trust portal names a 'Data Deletion / Data Retention' control and states that policies exist, without stating any deletion timeframe, scope or evidence. No deletion-on-termination clause, deletion certificate or backup-purge commitment was found on the pages read; the Master Terms, Annexes and DPA were not read within the credit cap.
Ask: Give us a numbered deletion commitment covering documents, prompts, outputs, embeddings, indexes, logs and backups, a timeframe measured in days, and a written certificate of deletion on exit.
Vendor and contractor access to customer content
Public claim only
The trust portal, read 2026-09-27, states that access to data is restricted by role and necessity, that internal access follows least privilege and separation of duties with regular review by administrators and documented rationales, and that access is revoked under termination policies. It also states that all employees undergo a background check and sign a non-disclosure agreement, and that a third-party managed security service provider performs 24/7 SOC monitoring with logs in a central SIEM. No customer-visible record of vendor access, and no per-request customer approval gate, is described.
Ask: List every path by which your staff, your managed security provider or any model provider can see our contract text, and can we require our approval per support ticket and receive an exportable log of every access?
SSO, MFA, roles and audit logging
Public claim only
The trust portal, read 2026-09-27, states that single sign-on is supported through multiple identity providers, that multi-factor authentication is supported via the customer's chosen identity provider and that two-factor authentication is configured through the identity provider when single sign-on is enabled, that role-based access control exists to manage permissions, and that centralised audit logging records all actions and is regularly reviewed. It does not say whether the audit log is exportable to the customer, whether SCIM provisioning or IP allow-listing exist, or whether single-sign-on-only can be enforced.
Ask: On our plan, can we enforce single-sign-on only, use SCIM provisioning and IP allow-listing, and export an audit log covering prompts, document access, exports, permission changes and vendor access?
Encryption in transit and at rest
Not confirmed
Three statements read 2026-09-27 do not line up. The trust portal's Data Security summary says measures 'include robust encryption for data at rest and in transit'. The trust portal section actually headed 'Encryption-at-rest' says only 'All customer data is encrypted in-transit using TLS 1.2' and makes no statement about data at rest. The product page FAQ says 'all data is encrypted in transit and at rest'. No algorithm for data at rest, no key custody or rotation detail, and no customer-managed key option is published on the pages read.
Ask: Confirm in writing whether customer contract data, search indexes, embeddings and backups are encrypted at rest, with which algorithm, who holds and rotates the keys, and whether customer-managed keys are available.
Independent security assurance
Not confirmed
The three pages describe the same assurance differently, all read 2026-09-27. The legal page says Leah 'conducts annual audits including SSAE-18 SOC 1 Type 2 and SSAE-18 SOC 2 Type 2, with certifications available to customers on request'. The trust portal lists compliance as 'HIPAA/HITECH' and 'GDPR' and links documents including a SOC 3 report described in its file name as a 2025 Type 2 SOC 3, an ISO/IEC 27001:2022 certificate for ContractPod Technologies, a TISAX report, a GDPR audit report dated July 2023, a network diagram and two penetration test reports; the HIPAA/HITECH report is on request. The product page FAQ says 'Leah is SOC 2 Type II certified'. What is published for download is a SOC 3 report, not a SOC 2 Type II report. No report was opened. The trust portal states third-party penetration testing every six months and on release of new products, while a separate section states annual independent penetration testing of the external network surface, the CLM web application and all CLM APIs.
Ask: Provide under non-disclosure the current SOC 2 Type II report with its period, scope and exceptions, the ISO/IEC 27001:2022 certificate scope statement, and the latest penetration test that covered the AI features; and reconcile the six-monthly and annual testing statements.
Contract set, DPA and standard form
Public claim only
The legal page, read 2026-09-27, names the contract set: Master Terms and Annexes for Leah SaaS and cloud products, a Data Processing Addendum covering the European Union standard contractual clauses and California personal information, a Security Data Sheet, and separate Marketplace Terms in a PDF dated 12 May 2026. It states that the Master Terms adopt the published Bonterms Cloud Terms v1 and the Bonterms DPA v1 and links to both. None of these documents was read within the credit cap, so liability caps, warranties, indemnities, intellectual property in outputs and any Australian addendum remain unverified.
Ask: Send the Master Terms, Annexes and DPA that would apply to us, marked against the Bonterms baseline, and identify every Leah-specific variation to liability, warranties, output ownership and data protection.
Hosting location, cross-border transfer and Australian residency
Public claim only
The trust portal, read 2026-09-27, states 'Leah infrastructure is hosted by Azure in multiple regions' and does not name any region. The product page FAQ states customers have 'full control over data residency'. The privacy statement names the storage locations European Union, United States of America and United Kingdom, but that statement is expressly about personal information provided via the website, not about the service. No Australian region and no Australian data residency option is named on any page read, and nothing states where model inference runs.
Ask: Is an Australian Azure region available for our tenancy and will you record it on the Order Form, and does model inference for redlining, Ask Leah and Contract Helpdesk also run in Australia or does the prompt leave the country?
Incident response and breach notification
Public claim only
The trust portal, read 2026-09-27, states that a documented Incident Response Plan exists and is reviewed, tested and approved annually, that a process exists for detecting, managing and notifying affected parties of data breaches 'in a timely manner', that a third-party managed security service provider runs 24/7 SOC monitoring with a central SIEM, and that cyber insurance is held. No notification deadline in hours or days is published, no Australian notifiable data breach cooperation commitment was found, and no incident register or history is published. The legal page publishes a whistleblower report form for suspected misconduct or breach of law, which is a separate channel from security incident notification.
Ask: Commit to a notification deadline in hours for any incident affecting our contract data, name the security contact and their hours in Australian time, confirm cooperation with a notifiable data breach assessment, and disclose material incidents in the last 24 months.
Change of terms, models and features
Not confirmed
The privacy statement, read 2026-09-27, states that if the policy changes the vendor 'may notify you on our website, by a blog post, by email, or by any method we determine' and that continued use is subject to the then-current policy. No notice commitment for changes to the service, to model providers or to the security posture was found on the pages read, and the trust portal states that the vendor 'continually reviews the rigor of current encryption standards' without a change-notice mechanism. The Master Terms, which would carry any change clause, were not read.
Ask: What written notice will we get before you change the model provider, the hosting region, the tenancy model or the security commitments, and do we get a testing window and a right to terminate if we do not accept the change?
Export, transition and shutdown
Not confirmed
Nothing on the pages read describes bulk export, an export format, a retrieval window after termination, transition assistance or escrow. The trust portal describes backups for the vendor's own continuity rather than customer-initiated export. The Master Terms and Annexes, which would carry any exit and return obligation, were not read within the credit cap.
Ask: Demonstrate a full tenancy export of contracts, extracted metadata, playbooks, templates, clause libraries and audit logs in open formats, and state the retrieval window after termination and the rate for transition assistance.
Connectors, add-ins and what they move
Public claim only
The product page, read 2026-09-27, names connectors to Salesforce, Microsoft Dynamics, HubSpot, SAP, SAP Ariba, Oracle, Workday, Coupa, ServiceNow, NetSuite, iValua, DocuSign, Adobe Sign, Signicat, Viafirma, SharePoint, OneDrive, Box, Google Drive, Dropbox, NetDocuments, Microsoft Teams, Slack and Outlook, plus open CLM APIs, and describes Microsoft Word and Outlook add-ins. Its FAQ states 'pre-built connectors for CRM, ERP, e-signature, and file storage mean there's no lengthy implementation project'. No connector documentation, permission scope, authentication model or statement of whether documents are copied out of the tenancy was published on the pages read.
Ask: For each connector we would enable, document the identity used, the permission scopes requested, whether document content is copied outside the Leah tenancy, how we revoke it, and how copies are purged on revocation.
Open-web retrieval
Not confirmed
No open-web search, browsing or external retrieval is described on any of the four pages read, checked 2026-09-27. 'Leah Discovery' is described as analysis across the customer's own contract corpus and the Contract Helpdesk as question-and-answer over the contract set, with every response linked to the clause or document it came from. Absence of a description is not confirmation that the feature is absent.
Ask: Can any Leah agent or skill fetch content from the open internet or from an external source outside our connectors, and if so what payload leaves our tenancy and can we disable it?
Actions the platform takes and confirmation gates
Not confirmed
The product page, read 2026-09-27, describes actions taken by the platform: kicking off a workflow from an uploaded document, routing approvals conditionally on deal value, risk flags, jurisdiction or contract type, delegating and auto-escalating on timers, sending documents for signature to DocuSign or Adobe Sign and syncing executed documents back, auto-triaging and prioritising inbound legal requests, and sending automated status updates to requesters. The platform family is published as 'Leah Agentic OS' and includes pages for orchestration, an agent library, workflows and an app builder, which were not read. No list of which actions require an explicit human confirmation, and no statement about isolating untrusted counterparty document text from instructions, was found.
Ask: List every action a Leah agent can take without a human clicking confirm, in particular sending a document to a counterparty or to e-signature, and show how text inside an uploaded counterparty document is prevented from acting as an instruction.
Human review before output leaves the firm
Not confirmed
The product page, read 2026-09-27, describes outputs a person then handles: redlines delivered as 'Word-ready redlines with tracked changes' for a user to accept or reject, and answers where 'Every response links to the specific clause or document it came from ... so your team can verify'. The same page states 'Leah doesn't hallucinate'. No mandatory supervisor approval gate before a redline or answer leaves the organisation is described, and no per-output record of the model used and the reviewer is described.
Ask: Can we require named supervisor sign-off before any Leah redline or drafted clause is sent to a counterparty, and can we export a per-output record showing the model used, the source clauses cited and who approved it?
