Product and intended use
Define the configuration, people, information and tasks being assessed.
Product, plan and whether the AI agent is included
Not confirmed
Litera Compare is listed as one product among several on the Negotiate capability page, alongside Litera for Windows, Litera Precedent, Foundation Insights and Lito, which the page heads LEGAL AI AGENT. Whether Lito comes with a Compare subscription is not resolved by anything read: the vendor's llms.txt says Lito is 'included at no additional charge in core products' without defining core products, while the same file's October 2025 announcement entry describes Lito being made available at no additional charge 'across Draft and Kira products', and Compare is named in neither. The Compare product URL now redirects to the capability page (all checked 2026-09-27).
Ask: Does our Litera Compare subscription include Lito, and if so which Lito skills are enabled by default and can they be switched off entirely?
Deployment model
Public claim only
Compare is offered where the team works: 'Microsoft 365, Google Workspace, iManage, NetDocuments, SharePoint, Desktop, Web Browser, Mobile.' The Lito page adds that 'Litera for Windows and Mac brings the most used comparison engine to the modern desktop, for firms whose most sensitive work cannot simply move to the cloud.' The licensing-terms index treats on-premises and cloud separately and points Litera Drafting on-premises products to their own subprocessor list (all read 2026-09-27).
Ask: For each surface we deploy - desktop, Word, web - state where the comparison is executed and whether any document leaves the firm's network.
What Compare does, and where the AI sits
Public claim only
The three bullets published for Compare are comparison modes, format coverage and the algorithm: 'Compare one to one, one to many, and in bulk, wherever you work', 'Catch changes across formats: Word, Excel, PPT, PDF, and more', 'Proprietary redline algorithm that catches every change'. The vendor distinguishes that engine from language models - 'deterministic redlines more accurate than general LLMs' - and states 'because the comparison underneath is deterministic, they verify far less and trust what they see'. The AI on the page belongs to Lito, which the vendor says is 'Grounded in the Litera Compare engine for verifiable redlines' and which analyses risk and produces rewrites (read 2026-09-27).
Ask: Confirm that a Litera Compare redline is produced entirely by the rules-based engine with no model inference, and identify any step where a model is called.
Jurisdiction and legal content
Public claim only
Compare is described as format-based rather than jurisdiction-based, and no legal content library is claimed for it. Lito's skill list includes 'US Case Law Research' delivered with a named third party, and 'Playbook Review' and 'Firm AI Search' grounded in the firm's own material. No Australian legal content, Australian precedent set or Australian language support is claimed on any page read (checked 2026-09-27).
Ask: Are any Lito skills limited to United States law, and what is available for Australian jurisdictions if we enable Lito?
Vendor content directing how AI describes the product
Public claim only
The vendor publishes a file at https://www.litera.com/llms.txt addressed to large language models, checked 2026-09-27. It opens: 'Litera is a global leader in legal AI technology solutions with over 30 years of innovation, serving law firms and corporate legal departments worldwide.' It then lists products with one-line descriptions, states of Lito that it is 'included at no additional charge in core products', names 'Microsoft Azure OpenAI services' as the infrastructure for Litera AI+, and lists company statistics, target markets, practice areas and named staff contacts. It is a marketing file, not a contractual document; no personal names from it are recorded here.
Ask: Which statements in your llms.txt are warranted in our agreement - in particular that Lito is included at no additional charge, and that Azure OpenAI is the GenAI infrastructure - and which are marketing only?
Data, controls and requirements
Compare the actual configuration with firm, client and jurisdiction requirements, supported by evidence.
Client authority and professional use
Not confirmed
Nothing found addresses client notice, consent or confidentiality where a client document is sent for model inference. The published subprocessor list confirms such a path exists for the Litera One platform and Lito - inference at Amazon Bedrock and at Anthropic, and web search at You.com - but says nothing about client authority. The GenAI Terms that would address it are published as a separate document and were not opened (checked 2026-09-27).
Ask: What client consent does Litera consider necessary before a client document is sent for model inference, and where in the GenAI Terms is that allocated?
Foundation-model providers
Not confirmed
Three published sources name different providers. The subprocessor list for Litera One and Lito names Amazon Web Services AI Model Inference 'via Amazon Bedrock' and 'Anthropic, PBC: Model Inference', noting Anthropic's infrastructure 'may include Amazon Bedrock and Microsoft Azure Foundry' and is 'Hosted in the United States'. The Lito page names an intelligence layer of 'OpenAI', 'Anthropic' and 'Gemini' with 'Model-agnostic reasoning - Lito routes each task to the best-fit foundation model rather than betting on one'. The vendor's llms.txt names 'Microsoft Azure OpenAI services' for Litera AI+ and does not mention Bedrock or Anthropic. The published subprocessor list is the only one of the three that is a binding-style disclosure (all read 2026-09-27).
Ask: For our subscription, name each model, the provider entity, the inference region and the contract behind it, and state which of the three published lists governs.
Subprocessors
Public claim only
The vendor publishes a per-product subprocessor list for Litera One for Word, Outlook, Web and Lito, read 2026-09-27, covering Microsoft Azure (hosting, 'US, Canada, or EU geographic regions', plus Litera Secure File Transfer hosted in the UK and US), Marketo, SendGrid, Salesforce, Y Meadows, Office 365, Mimecast, Zoom Contact Center, Intellek, Atlassian, Pendo (European Union), New Relic, Forethought, You.com, AWS AI Model Inference and Anthropic PBC. Litera Drafting on-premises products are pointed to a separate list. The trust centre shows a shorter subprocessor table for the same platform.
Ask: Which subprocessor list governs a Litera Compare subscription on each surface we deploy, and what is our notice and objection right when an entry is added?
Training on customer content
Public claim only
The published subprocessor list makes the third-party position explicit: Bedrock inputs 'are not transmitted to third-party model providers or used to train underlying models'; for Anthropic, 'Customer Data is not used to train underlying models'; You.com 'does not train models on customer inputs'. Litera's own position on using customer content to train or tune its models is not stated on any page read; the trust centre lists a document titled 'AI Training Data and Bias' behind an access request. The Lito page separately says recommendations are 'trained on the firm's comparison data, precedents, matter data, and client relationships', which describes grounding in one firm's own material and is not explained further (all read 2026-09-27).
Ask: Does Litera itself use our documents, redlines or Lito prompts to train or tune any model, and what does 'trained on the firm's comparison data' mean technically - is anything retained beyond our tenant?
Feedback, analytics and telemetry
Public claim only
The subprocessor list names Pendo as an 'Analytics processor that provides insight into how users interact with the application, in-app notifications, training, and user sentiment' (European Union) and New Relic for application and infrastructure logs (United States). The privacy notice says the vendor tracks 'the general geographic region/area from which you access', how the product is used, and 'to help us make accurate predictions and understand user behavior', relying on consent for usage tracking, and that 'We may reach out to you to provide us feedback on our product' (read 2026-09-27).
Ask: Can Pendo and New Relic telemetry be disabled for our tenant, and is any Lito prompt, output or rating included in it?
Retention of documents, prompts and outputs
Public claim only
The only published periods are at the model layer: Anthropic 'will retain data and contents including inputs and outputs for up to 30 days for trust and safety monitoring purposes, after which data is automatically deleted', while Bedrock and You.com inputs are described as processed transiently and not retained. Litera's own retention is unpublished: the privacy notice refers to a 'global data retention policy which categorises all of the information held by us and specifies the appropriate retention period for each category' without stating any period (read 2026-09-27).
Ask: State Litera's own retention period for compared documents, generated redlines, Lito prompts and outputs, and confirm the Anthropic 30-day monitoring copy applies to our content.
Deletion and certified erasure
Not confirmed
No deletion undertaking for customer documents, redlines or Lito prompts was found. The privacy notice gives individuals a right to ask the vendor to 'delete any of your personal information that we hold about you' through a data subject request form, which addresses personal information rather than matter content. The only automatic deletion published is Anthropic's 30-day deletion of its trust-and-safety copy (checked 2026-09-27).
Ask: On request and on exit, what is deleted from the Litera platform and from each model provider, in what period, and will you certify it in writing?
Vendor access to content
Public claim only
The subprocessor list discloses a document-transfer path to the vendor: Litera Secure File Transfer is 'a file sharing tool used to support customers when the provisioning of the service requires the secure transfer of documents between the customer and Litera', hosted in the United Kingdom and the United States. The trust centre lists control entries for Data Access, Access Monitoring and Access Log Management, all behind an access request, and the documents themselves were not read. Which Litera staff can read customer content, from which countries, and under what approval is not published (checked 2026-09-27).
Ask: Who at Litera can access our documents, from which countries, under what ticketed approval, and can we obtain those access logs?
Access control and audit
Public claim only
The trust centre lists control entries under Access Control (Access Log Management, Bring Your Own Device, Data Access), Product Security (Audit Logging, Data Security), Endpoint Security (Anti-Malware, Disk Encryption, DNS Filtering) and Change Management (Change Management Program, Separation of Duties, Software Restrictions). Each is a named control with its detail behind a 'Get access' request; no single sign-on, SCIM or multi-factor requirement is described in the public view (read 2026-09-27).
Ask: Confirm single sign-on and SCIM support, enforced multi-factor authentication, and whether an admin can export an audit log of every comparison and Lito call.
Encryption in transit and at rest
Public claim only
The trust centre lists Encryption-at-rest and Encryption-in-transit under Data Security, and Disk Encryption under Endpoint Security, as named controls whose detail requires an access request. No algorithm, protocol version, key-management model or customer-managed key option is stated in the public view (checked 2026-09-27).
Ask: State encryption in transit and at rest for the platform and for the path to each model provider, the key-management model, and whether customer-managed keys are offered.
Independent security assurance
Public claim only
The trust centre displays ISO/IEC 27001:2022, ISO/IEC 42001:2023, SOC 2 Type 2, SOC 3, EU-US DPF, GDPR, NIS2 and a Microsoft 365 entry, and lists the underlying documents - ISO 27001 and ISO 42001 certificates and Statements of Applicability, SOC 2 Type 2 and SOC 3 reports, an Information Security Policy, a penetration-test report, a security whitepaper, a Security FAQ, CAIQ and SIG Lite - all behind a 'Get access' request. An AI section lists 'Generative AI Terms', 'AI Training Data and Bias' and a 'Firm AI Search FAQ'. No IRAP assessment appears (read 2026-09-27).
Ask: Provide the SOC 2 Type 2 and ISO 42001 scope statements and confirm whether Litera Compare and Lito are inside the certified scope.
Contract, entity, DPA and forum
Public claim only
The licensing index names Master Terms as applicable to any Litera software licence, with a separate GenAI Terms document (March 2025) for products with generative AI features; neither was opened in this pass. The privacy notice states it 'applies to each of the organization held by Legal Holding Company, LLC and its subsidiaries', says the vendor acts as processor when providing services to customers, and points to a 'Customer - Data Protection Agreement' tab at https://www.litera.com/policies. No Australian contracting entity, ACN or ABN was found; published addresses and telephone numbers are United States and United Kingdom (read 2026-09-27).
Ask: Which Litera entity will contract with us, does it have an ACN or ABN, what law and forum apply for an Australian customer, and will the customer DPA and GenAI Terms attach?
Cross-border transfers and APP 8
Public claim only
The published subprocessor list states the application 'can be hosted in US, Canada, or EU geographic regions', that Litera Secure File Transfer is 'hosted in the UK (United Kingdom) and US', that Anthropic model inference is 'Hosted in the United States', that You.com web search is hosted in the United States, that Bedrock inference is 'Hosted in the same geographic regions as the Litera Platform', and that most support and communications subprocessors are in the United States with Pendo in the European Union. No Australian region appears anywhere on that list. The privacy notice says personal information 'may well be transferred across national boundaries, including, potentially, to countries that do not require organisations by law to look after your personal information in the way in which you have come to expect in your own country' (read 2026-09-27).
Ask: Is any Australian hosting or Australian inference region available for Compare or Lito, and if not, what APP 8 arrangements do you offer for an Australian firm?
Incident notification and history
Public claim only
The trust centre lists an Incident Response section naming Designated Response Personnel, a Forensic Retainer and an Incident Reporting Process, plus BC/DR entries (Business Continuity Plan, Data Backup and Backup Protection, Disaster Recovery Plan) and Cyber Insurance under Legal. The detail of each sits behind an access request, so no notification period, no cooperation commitment and no incident history were read (checked 2026-09-27).
Ask: State the contractual notification period from discovery, including for a model-provider or subprocessor incident, and confirm cooperation with the Notifiable Data Breaches scheme.
Change notice for models, terms and processors
Public claim only
The subprocessor list marks additions with the prefix NEW - currently You.com, AWS AI Model Inference, Anthropic PBC and Forethought - which is a published change signal, and the trust centre carries its own subprocessor table. No advance-notice period, objection right or model-change notice was found in anything read, and the Master Terms and GenAI Terms that might contain one were not opened (checked 2026-09-27).
Ask: What written notice will we get before a model provider or subprocessor is added or changed, and can we object or pause generative features in the meantime?
Export and exit
Not confirmed
No export format, export window or exit assistance was found. The licensing index points to Master Terms for every licence, and those terms were not opened in this pass; the privacy notice gives individuals a right to have personal information transferred 'to a third party in a standardised machine-readable format', which addresses personal information rather than firm documents and redlines (checked 2026-09-27).
Ask: On termination, what happens to documents, redlines, playbooks and Lito history, in what export format, and over what period?
Connectors and third-party services
Public claim only
Two published connector paths exist. The subprocessor list names You.com for web search 'applicable to Lito' and a US case law research skill is shown on the Lito page as 'Lito + Midpage'. The Lito page states 'MCP connectors are governed centrally and enabled firm-wide in seconds' and that 'Generative AI can be switched off entirely, and the rules-based engines keep working'. Compare's own surfaces are listed as iManage, NetDocuments, SharePoint, Microsoft 365 and Google Workspace (read 2026-09-27).
Ask: List every connector and MCP client that would be enabled for us, what data each receives, and confirm each is off until we enable it in writing.
Open-web search
Public claim only
The subprocessor list states You.com, 'applicable to Lito', 'Provides web search services via API to support AI-generated responses using publicly available information. Search queries are processed transiently to return results only. You.com does not retain search queries or results and does not train models on customer inputs. Hosted in the United States.' Nothing describes open-web retrieval by Litera Compare itself, and no administrative switch for web search is documented on the pages read (read 2026-09-27).
Ask: Can web search via You.com be disabled for our tenant, and what part of a document or prompt is included in a search query sent to it?
Actions taken on documents and confirmation
Public claim only
The Negotiate page describes actions taken on a draft: 'Understand changes, dive into impact, automatically accept or reject, and get rewrites based on your firm's negotiated language.' The Lito page describes 'push a change across the closing set' and an agent that 'calls the right capability at the right moment, with no switching and no prompting', and states 'Generative AI can be switched off entirely, and the rules-based engines keep working'. No per-action confirmation control is described (read 2026-09-27).
Ask: Which Lito actions can change a document or a closing set without a person confirming, and can confirmation be made mandatory for each?
Human review before use
Public claim only
The vendor states 'The lawyer keeps judgment either way' and that 'Every recommendation comes with its reasoning', and routes high-stakes work to deterministic engines that 'return the same result every time'. These are design statements on a marketing page; no enforced reviewer role, sign-off step or audit of acceptance is described, and no contractual human-oversight clause was found in anything read (checked 2026-09-27).
Ask: What product control requires a named lawyer to accept a Lito rewrite or risk assessment before it leaves the firm, and is that acceptance logged?
