Product and intended use
Define the configuration, people, information and tasks being assessed.
How Knowledge Base is licensed and enabled
Public claim only
The security overview dated September 2026 states that 'An organization's Adobe Administrator enables Knowledge Base in Acrobat using a dedicated toggle inside the Acrobat Studio for Enterprise product profile in the Adobe Admin Console', that 'For new customers, this toggle is automatically enabled', and that the organisation 'must have Acrobat Studio for Enterprise and Acrobat Shared Credits to control access and usage'. Users licensed for Acrobat Studio can create, manage and query collections, while 'Any user with an Adobe ID but not licensed to use Acrobat Studio can query Knowledge Base collections that have been shared with them'. The Australian product page publishes no price and directs readers to 'Find out how to access Knowledge Base through Adobe Acrobat Studio enterprise plans.'
Ask: Does our Acrobat Studio enterprise agreement include Knowledge Base and the Shared Credits it consumes, and is the toggle on by default in our tenancy today?
Where the product and its data storage run
Public claim only
The security overview states that 'The Knowledge Base web application, Knowledge Base Generative AI Service, and corresponding data and file storage are hosted in AWS data centers in Virginia, USA and Frankfurt, Germany', and that 'Although Knowledge Base in Acrobat is built on the Acrobat Sign infrastructure, data remains separate and is not transferred or shared between the Knowledge Base in Acrobat and Acrobat Sign solutions.' Adobe's Document Cloud hosting page, last updated 16 September 2026, states that 'Analyzer in Acrobat and Knowledge Base in Acrobat are hosted and associated user-generated content is stored in the Acrobat Sign infrastructure in the United States and Germany.' One component runs locally: a separate Knowledge Base Web Indexer desktop application is used to ingest sites that require authentication.
Ask: Which of the two data centres would hold our collections and vector indexes, can that be fixed in the order documents, and is an Australian region planned with a date?
Documented components, inputs and surfaces
Public claim only
The security overview lists a web application for building curated collections from uploaded documents, files, images and web links; a Web Indexer desktop application for sites requiring authentication such as internal wikis; SharePoint and Google Drive connectors that 'continuously sync'; an AI Assistant conversational interface; chatbots for Slack and Microsoft Teams; a Generative AI Service that parses documents into chunks and indexes them 'using an advanced RAG platform optimized for large document collections'; data storage holding both a vector database and the original documents for citation retrieval; and APIs. It states the service supports 'multi-turn Q&A, search, summarization, and entity extraction'. The Australian product page states collections can run to 'hundreds to thousands of documents' across 'PDF, web content, email, text files and common document formats'.
Ask: What are the published limits on collection size, document size, number of collections and supported formats, and which of our file types would be rejected or silently skipped?
Legal content and jurisdiction
Public claim only
No legal content library, citator or jurisdiction setting is described in either document; answers are grounded only in the collection the customer curates. Adobe's Australian legal use-case page states that Knowledge Base in Acrobat finds answers across connected repositories 'with every response traceable to the specific document and source sentence', and that on the same page Adobe states its AI 'does not constitute legal advice or risk assessment' and that 'Outputs should always be reviewed by qualified legal counsel.'
Ask: Does Knowledge Base apply any Australian legal knowledge or currency checking, or is an answer only as current as the documents a firm has placed in the collection?
Who Adobe names as the intended users
Public claim only
The Australian product page describes the intended content as 'policies and procedures, HR handbooks, legal guidance, support playbooks, contracts and other repositories of institutional knowledge', and the solution brief listed on the page names 'IT, HR, Legal and Customer Support teams'. The security overview describes the audience only as organisations exposing knowledge from curated collections, and the help page states that regulated use cases require the PDF Spaces integration to be disabled before regulated data is ingested.
Ask: Is Knowledge Base supported for client-confidential matter material as distinct from internal firm knowledge, and what does Adobe recommend for collections that mix the two?
Vendor content addressed to AI models, and marketing claims that need testing
Public claim only
A deliberate check for vendor content directing how AI should describe this product found none: no 'instructions for AI' block and no /ai-info page appears on the Australian product page, the help page or the Trust Centre pages read on 2026-09-27. Adobe does publish a file at https://www.adobe.com/llms.txt addressed to large language models, opening '# Adobe Tools > Give your team a creative edge. Create something beautiful, boost productivity, and deliver engaging experiences with Adobe software.'; it is a product directory with one-line descriptions, contains no instruction of the form 'describe this product as X', and does not mention Knowledge Base or Acrobat Studio at all. Crawler allow and disallow directives, which are access control rather than narrative control, were not examined. Separately, the product page describes answers as 'trusted' and 'grounded', which is framing rather than a measured property; the help page is the document that states the certification position plainly.
Ask: Which statements on the Knowledge Base product page are warranted in our agreement, and is the llms.txt file maintained - it does not mention Knowledge Base or Acrobat Studio?
Data, controls and requirements
Compare the actual configuration with firm, client and jurisdiction requirements, supported by evidence.
Client authority and confidentiality
Not confirmed
Building a collection copies documents, including copies of every synced SharePoint or Google Drive document, into AWS data centres in Virginia and Frankfurt, and passes content chunks and queries to Azure OpenAI, so a firm needs authority for an offshore disclosure before ingestion. Section 2.4 of the Australian General Terms states the customer agrees not to collect, process or store Sensitive Personal Information using the services except as directly authorised by Adobe, as intended by the services, or as governed by the Product Specific Terms, and not to make it available to Adobe or its third-party providers; the help page separately requires the PDF Spaces integration to be disabled before regulated data is ingested and warns that 'Disabling PDF Spaces integration doesn't by itself establish compliance with a regulatory framework.'
Ask: Does Adobe authorise the ingestion of client-confidential and sensitive personal information into a Knowledge Base collection, and in which document is that authorisation given?
Foundation-model provider for this product
Public claim only
The security overview dated September 2026 names the provider for this product: the Generative AI Service 'currently leverages Azure OpenAI', which is called both to 'generate semantic embeddings' of ingested content and to generate answers from retrieved content, and the data passed to it is listed as customer queries, 'The relevant user document data needed to answer the user's query' and 'The instructions for how to answer the user's question'. It states that 'Adobe has modified the Azure OpenAI abuse monitoring feature so that Microsoft cannot store or conduct human review of the input and output data processed by Azure OpenAI', that 'Adobe does not use any customer data to train or fine-tune Azure OpenAI', and that Microsoft Azure AI Content Safety is configured to 'Medium' blocking on input and output. No model name or version is published and no customer model choice is described.
Ask: Which Azure OpenAI models serve embedding and answer generation for an Australian customer, what notice would we get before either changes, and can a version be pinned?
Subprocessors reached by ingestion and by a query
Public claim only
For Knowledge Base specifically the security overview names Microsoft twice - Azure OpenAI for embeddings and answers, and Azure AI Content Safety for filtering - and AWS as the host of the application, service and storage; where a query comes from a chatbot, Slack or Microsoft Teams sits in the request path, and where it comes from PDF Spaces the Acrobat Generative AI Service and the user's Adobe Cloud Storage are also in the path. Adobe's sub-processor list, last updated 28 July 2026, is maintained under its data-protection terms with a subscription for updates, but is written at Document Cloud level and does not name Knowledge Base; it explains that the locations it gives are sub-processor headquarters and key hubs rather than the location of Adobe's instance.
Ask: Provide the sub-processor list scoped to Knowledge Base, including the Slack and Teams chatbot path and the PDF Spaces path, with our notice and objection rights when it changes.
Training on customer content
Public claim only
Section 4.3(A) of the Australian General Terms states 'We will not use these rights to train generative AI models on your Content and will not use the sublicence rights to have anyone else train generative AI models on your Content, except at your specific request', and section 2.2(F) excludes Local and Cloud Content from generative AI training except content submitted to Adobe Stock. The Knowledge Base security overview adds that 'Adobe does not use any customer data to train or fine-tune Azure OpenAI.' Content Analytics is a separate mechanism and not model training: the Content Analysis FAQ dated 10 June 2026 states business and school accounts and profiles 'have been automatically opted out', with exceptions where a user is not signed in or joins feedback, beta, user-study or improvement programmes.
Ask: Confirm in our agreement that no collection document, embedding, query or answer trains or fine-tunes any model at Adobe or Microsoft, and that Content Analytics is off for every profile we hold.
Feedback and human review of submitted content
Public claim only
The security overview confirms a feedback mechanism exists inside the product but does not say what Adobe does with it: it lists 'feedback' among the user personal data deleted when an administrator disables or removes a user account, alongside chat history, configured connectors and chatbots and usage records. Section 2.2(G) of the Australian General Terms states human review of Cloud Content may occur in limited circumstances - on the customer's request, where content is made public, where it is flagged or reported as illegal or abusive, or where the customer opts in to a prerelease, beta or improvement programme - and that Local Content is 'never' reviewed. The Document Cloud hosting page states that 'Reported Feedback data submitted by users is handled separately'.
Ask: When a user submits feedback on a Knowledge Base answer, what document content leaves our tenancy, who reads it, and can the control be disabled for all users?
Retention of documents, chats and indexes
Public claim only
The security overview states that all chat interactions are 'retained in Knowledge Base data storage until the relevant document or collection of documents is deleted', and that documents 'along with the content indexes and citations generated from these files' are stored 'until the user deletes the file(s)'. For connected repositories, 'Knowledge Base mirrors the retention policy of the SharePoint/Google Drive folder(s)', except that if the connector's authentication becomes invalid the documents 'will remain in Knowledge Base data storage rather than following the source repository's version, audit, or deletion state'. No maximum retention period and no backup-deletion interval is published.
Ask: Is there a tenancy-level retention or auto-expiry policy for collections, indexes and chat history, and how long do deleted items persist in backups?
Deletion, user offboarding and contract end
Public claim only
The security overview states that when a user deletes a document in the web application 'the document and its corresponding metadata are immediately deleted, no longer accessible to any user, and not recoverable through the Knowledge Base application'. On offboarding, disabling or removing a user deletes that user's personal data but 'content uploaded by that user, including documents in collections and content associated with connectors or chatbots, is retained unless the organization's Adobe Administrator explicitly deletes it', with ownership transferring to the Adobe Admin at collection level, and a footnote states 'Adobe does not currently log these events'. On contract end it states that 'Adobe does not automatically delete customer data upon contract termination or expiration' and that the customer should export or delete content beforehand or request deletion at termination; the Australian General Terms separately allow thirty days to transition content out, after which Adobe reserves the right to delete it.
Ask: At termination, will Adobe delete every collection, index, chat history and connector configuration on written request, within what period, and will it certify the deletion?
Adobe and provider access to collection content
Public claim only
The security overview states that Adobe has modified Azure OpenAI abuse monitoring 'so that Microsoft cannot store or conduct human review of the input and output data processed by Azure OpenAI', and that uploaded documents 'are only visible to the user unless they choose to share the collections in which the documents are ingested with other authorized users within the same organization'. Section 2.2 of the Australian General Terms states that 'A person may review your content on our servers in limited circumstances, such as upon your request, when you choose to let us use your content to improve our products or when your content is flagged or reported as illegal.' No break-glass or support-access process for Knowledge Base data storage is published, and the security overview states that customer-accessible audit logs are not currently available.
Ask: Which Adobe teams can read documents, embeddings or chat history in Knowledge Base data storage, from which countries, under what approval, and will you provide a record of such access on request?
Administrative control, sharing and audit logging
Public claim only
The security overview states that Knowledge Base is enabled by a product administrator toggle in the Acrobat Studio for Enterprise product profile and is 'automatically enabled' for new customers; that users are authenticated by Adobe Identity Management Services; that any Adobe ID holder can query collections shared with them; that sharing is at collection level to authorised users 'within the same organization'; and that the PDF Spaces integration is 'enabled by default' and can only be changed by a product administrator. It also states plainly that 'Document changes are recorded in internal application logs, but customer-accessible audit logs are not currently available.'
Ask: Without customer-accessible audit logs, how can we evidence who queried or exported a collection, and when will such logs be available?
Encryption
Public claim only
The security overview states that 'All data is encrypted in transit over HTTPS using TLS 1.2 or greater', that 'All data stored in Knowledge Base data storage is encrypted at rest using AES 256-bit encryption', that documents are encrypted 'on a per-document basis', and that responses returned to PDF Spaces are stored 'in the user's provisioned encrypted Adobe Cloud Storage'. No key-management detail and no customer-managed key option is described.
Ask: Who holds and rotates the keys for Knowledge Base data storage and the vector database, and is a customer-managed key option available on an Australian enterprise agreement?
Independent security assurance
Public claim only
Adobe's help page dated 23 September 2026 states the position plainly: 'Formal certifications for Knowledge Base in Acrobat are in progress', the service 'runs on infrastructure governed by Adobe's Common Controls Framework (CCF)', and 'Because the Compliance List on the Adobe Trust Center only lists services with a completed, named attestation or report, Knowledge Base in Acrobat is not yet listed there', with certifications expected 'in the November/December timeframe'. It lists SOC 2, SOC 3, ISO 9001, ISO 27001, ISO 27017, ISO 27018, ISO 22301, CSA STAR Level 2, BSI C5 and FedRAMP Tailored as 'Targeted by end of 2026', FDA 21 CFR Part 11 as a future release and FedRAMP Moderate as 'TBD', and states that 'A target date indicates work in progress. It doesn't mean that Knowledge Base currently holds the listed certification or assessment.' HIPAA readiness is listed as available now, with an external review report 'currently in process'.
Ask: On what date will a SOC 2 or ISO 27001 report naming Knowledge Base in Acrobat be available to us, and until then will you warrant that the service is operated under the same controls as the audited Acrobat Sign environment?
Contracting entity, liability and forum
Public claim only
The Australian General Terms state that for customers residing outside North America and Japan the relationship is with Adobe Systems Software Ireland Limited under the laws of Ireland, and that 'For customers in Australia, Adobe Systems Software Ireland Limited is acting as an authorised agent of Adobe Systems Pty Ltd. and is entering into this contract in its capacity as agent for Adobe Systems Pty Ltd.' No ACN or ABN is published. Total liability is limited to the greater of US$100 or the amount paid in the three months preceding the event, the services are supplied 'AS-IS' to the maximum extent permitted by law, and rights under the Competition and Consumer Act 2010 (Cth) are preserved with liability for a statutory guarantee limited at Adobe's election to re-supply or its cost. Section 2.3(B) provides that Data Protection Terms apply to personal data collected from individuals outside the EEA and the UK; those terms, and any Acrobat Studio for Enterprise agreement, were not read.
Ask: Confirm the Australian contracting entity and its ACN for an Acrobat Studio for Enterprise agreement covering Knowledge Base, provide the Data Protection Terms that will apply and the Product Specific Terms, and state any negotiated liability cap.
Cross-border transfers and Australian Privacy Principle 8
Public claim only
The security overview places the application, service and all storage in AWS data centres in Virginia, USA and Frankfurt, Germany, and states that for the model 'Adobe restricts processing of Knowledge Base data to the customer's provisioned data sovereignty zone (either North America or EU)' - there is no Australian zone, so an Australian firm's collections and queries are processed offshore. Section 2.5 of the Australian General Terms states 'We process and store information in the U.S. and other countries. By using our Services and Software, you authorise Adobe to transfer your personal information across national borders and to other countries where Adobe and its partners operate.' Adobe's sub-processor list states it applies transfer protections including European Commission approved Standard Contractual Clauses and the Data Privacy Framework.
Ask: Name every country in which our collections, embeddings, queries and answers are processed, cached or logged, including the Slack and Teams chatbot path, and identify the Australian Privacy Principle 8 protection relied on for each.
Incident notification
Not confirmed
No breach-notification period was found for Knowledge Base. The Australian General Terms contain no notification clause, neither the security overview nor the help page has an incident section, and the worldwide Data Protection Terms at adobe.com/go/dpt-ww, which would ordinarily carry the obligation, were not read within the evidence budget.
Ask: State the contractual notification period from discovery for an incident affecting Knowledge Base collections, whether it extends to incidents at Microsoft, AWS, Slack or Google, and what Adobe provides for a Notifiable Data Breaches assessment.
Change notice for models, terms, defaults and processors
Public claim only
The security overview's model wording is expressly current rather than fixed - 'Knowledge Base currently leverages Azure OpenAI' - and two documented defaults can change the exposure of a collection without any firm action: the Knowledge Base toggle is 'automatically enabled' for new customers, and the PDF Spaces integration 'is enabled by default'. The help page states that certification target dates 'can change while certification and assessment work is underway'. The sub-processor list states Adobe will update the page as sub-processor use changes and offers a subscription to updates, and the Australian General Terms provide for updates to the terms and to Product Specific Terms.
Ask: What advance notice do we get before the model, the sovereignty zone, the hosting region or a default such as the PDF Spaces integration changes, and can we require that new defaults arrive switched off?
Export and exit
Public claim only
The security overview states that on termination 'the customer should export or delete any content they wish to retain or remove prior to account closure', that the customer 'may also notify Adobe at the time of contract termination and request deletion of their data', and that 'Adobe does not automatically delete customer data upon contract termination or expiration'. No export format or mechanism is described for a collection, its indexes, its chat history, its chatbot configurations or its connector settings, and the product does not support document version history. The Australian General Terms require any transition of content out to be completed within thirty days of termination or expiry.
Ask: In what format can we export a collection, its chat history and its chatbot and connector configuration before termination, and what cannot be exported at all?
Connectors and third-party services
Public claim only
The published connection points are a Microsoft SharePoint connector, a Google Drive connector, both continuously syncing and storing copies in Knowledge Base data storage, web links to public websites, a separate Web Indexer desktop application for authenticated internal sites, Adobe-developed chatbots in Slack and Microsoft Teams, Knowledge Base APIs, and the PDF Spaces integration which is enabled by default. Section 3.12 of the Australian General Terms states that third-party software and services made available through the services are not Services and Software as defined in the terms and that their use is governed by the third party's terms. The security overview states that connector synchronisation depends on the user's SharePoint or Google Drive credentials remaining active.
Ask: Whose credentials does each connector hold, can a connector be scoped to named sites or shared drives, and can the Slack and Teams chatbots be blocked centrally?
Open-web retrieval and ingested web content
Public claim only
Two different things must be kept apart. Knowledge Base ingests web content at the customer's direction: the security overview states users 'can also add web links to dynamically retrieve pre-indexed content in their collections', and a Web Indexer desktop application ingests 'web sites/pages that require authentication, such as Wikis and other internal sites'. Separately, none of the three documented query data flows contains an open-web search step; they draw on Knowledge Base data storage and Azure OpenAI only. Adobe's sub-processor list states at Document Cloud level that 'Adobe Document Cloud utilizes the Brave Search web search API', and whether any Knowledge Base action reaches it is not stated.
Ask: Can a Knowledge Base answer draw on anything outside the collection, including Brave Search or model knowledge, and can that be prevented for our tenancy?
Actions the product can take
Public claim only
In the documented flows Knowledge Base reads and indexes content and writes embeddings, responses and chat history into its own storage; where a query comes from PDF Spaces, the Acrobat Generative AI Service 'stores the generated responses and the chat history in the user's provisioned encrypted Adobe Cloud Storage'. No flow describes editing, deleting or writing back to a source document in SharePoint or Google Drive. The one administrative action described is ownership transfer of a collection to the Adobe Admin when a user is removed, which the security overview states Adobe 'does not currently log'.
Ask: Can Knowledge Base or its connectors write to, rename or delete anything in a connected SharePoint or Google Drive repository, and what write scopes does the connector request?
Human review before use
Public claim only
Adobe presents verification as the reader's task: the product page is headed 'Get the answer and the source to prove it' and describes answers 'each linked to the verifiable source', the security overview describes responses 'grounded in the source documents with citations back to the source', and Adobe's Australian legal use-case page states outputs 'should always be reviewed by qualified legal counsel'. No product control that requires, records or evidences a reviewer's sign-off is described, and the security overview states customer-accessible audit logs are not currently available.
Ask: Is there any setting or record showing that a named person checked a Knowledge Base answer against its cited source before it was relied on or sent outside the firm?
