Product and intended use
Define the configuration, people, information and tasks being assessed.
What the product is and how it is licensed
Public claim only
The product page, checked 2026-09-27, states that PDF AI Assistant 'is included with any Foxit account', is 'Built into Foxit PDF Editor and the free Foxit PDF Reader, on desktop, cloud, and mobile', and is metered by AI credits: 300 a month free with every account, or 2,000 a month on a subscription, which the AI pricing page prices at '$4.99 per month or $49.99 per year' with paid credits pooling through the Foxit Admin Console while 'Free monthly credits are individual and do not pool'. The same pricing page states 'Perpetual versions of PDF Editor don't include the AI Assistant'. A published capability matrix separates what runs where: chat, summarise, translate, Read Aloud, Smart Podcast and Skills are available in the free Reader, while AI Bookmark, Generate Image, edit by chat and PDF Action Inspector require Foxit PDF Editor, the last two on desktop only.
Ask: For an Australian firm, what is the price in Australian dollars, how many credits does a typical contract summary, translation or chat session consume, and what happens when a user's monthly credits run out mid-task?
The three processing modes
Public claim only
The Foxit AI Assistant Security Overview describes three modes for eligible Editor workflows. In Cloud mode the prompt, the document content or context needed for the task and the generated result are processed through 'the applicable regional cloud workflow'. In On-Device mode 'document processing, product-side storage, and model inference occur on the user's device' using 'compatible small language models downloaded to and run on the user's device', which the product page states runs 'through Microsoft Foundry Local' with no internet connection and no credits. In Custom mode 'Foxit product-side document processing and storage remain local, while model requests are sent to the customer-configured external endpoint'. Foxit Workspace, covered by the same document, is cloud only.
Ask: For each mode, what is the minimum Foxit PDF Editor version and device specification, and can an administrator force all users into On-Device or Custom mode and block Cloud mode?
Documented functions
Public claim only
The Security Overview lists the capability groups as document chat and text assistance (Q&A, summaries, rewriting, translation assistance), agentic and tool-enabled workflows ('Natural-language PDF actions, diagram creation, Skills, and supported Connectors on eligible desktop releases'), and vision-enabled workflows ('Formula recognition, MathML generation, and supported image-understanding actions'). The product page adds AI Bookmark for long unformatted documents, Generate Image, Read Aloud, Smart Podcast, formula conversion to MathML or LaTeX, user-created custom Skills, and PDF Action Inspector, which it states 'scans for those behaviors' where a PDF 'can carry JavaScript that alters content without your knowledge, hide data that survives redaction, or print something different from what's on your screen'.
Ask: Which of these functions are available in the release and platform we would deploy in Australia, and which consume credits?
Legal content and jurisdiction
Public claim only
No legal content library, citator or jurisdiction setting is described on any page read; the assistant works on the document the user has open. The Foxit AI User Guidelines, which Foxit states are 'incorporated into Foxit End-User License Agreement (Foxit EULA) and Foxit General Terms of Service', state under the heading 'No professional advice' that 'The content created by Foxit's AI solutions cannot replace the advice provided by officially recognized professionals. Do not use Foxit's AI solutions to replace medical, legal, financial, or other kind of professional expertise', and that in case of violation of that rule 'Foxit cannot be held liable for and will not indemnify you'.
Ask: How does Foxit distinguish permitted use of the assistant on legal documents from the prohibited use of it 'to replace legal expertise', given that Foxit publishes a Legal solution page for the same products?
Stated intended use and prohibited use
Public claim only
The AI User Guidelines, incorporated into the EULA and the General Terms, state that the user agrees to use the AI features 'only for your creative and productivity work product', prohibit use to create or share abusive or illegal content, prohibit content that infringes third-party rights, and state that the user is responsible for verifying content. The Security Overview states that 'Foxit AI Assistant is not intended to make decisions about individuals that produce legal or similarly significant effects without meaningful human review', and that 'Customers are responsible for establishing appropriate review, authorization, and recordkeeping controls for high-impact workflows.'
Ask: Does Foxit consider use of the assistant on client matter documents by a law practice to be within the AI User Guidelines, and will you confirm that in writing?
Vendor content addressed to AI models
Not confirmed
A deliberate check for vendor content directing how AI models should describe this product found none. No 'Instructions for AI Assistants' block or equivalent appears on the product page, the Security Overview, the AI User Guidelines, the General Terms, the privacy policy or the Trust Center, all read on 2026-09-27, and the cached site map for foxit.com contains no /llms.txt and no /ai-info path, so no such file was requested. A separate matter, and not the same thing, is that Foxit does publish an MCP server page offering to 'Connect AI agents to Foxit document capabilities', which is an integration surface rather than content telling a model how to characterise the product. Crawler allow and disallow directives, which are access control rather than narrative control, were not examined.
Ask: Do you publish, anywhere on foxit.com, guidance addressed to AI models or assistants about how to describe Foxit products, and if so at what address?
Data, controls and requirements
Compare the actual configuration with firm, client and jurisdiction requirements, supported by evidence.
Client authority and confidentiality
Not confirmed
In Cloud mode the prompt and the document content needed for the task leave the workstation for Foxit's regional cloud and then for Microsoft Azure, so a firm needs authority for that disclosure before first use; in On-Device mode the Security Overview states document processing and inference stay on the device. Foxit places the decision on the customer: the Security Overview states the customer must 'determine whether submitted content is permitted', and 'Customers with specific data-residency or international-transfer requirements should refer to their agreement, data processing agreement, subprocessor disclosure, privacy notice, and deployment-specific documentation for the commitments applicable to their deployment.' The privacy policy gives consent as the basis for processing files uploaded to Foxit cloud services.
Ask: For an Australian customer in Cloud mode, in which country is the document content processed by the model service, so that we can describe the disclosure to a client before we make it?
Model providers
Public claim only
The Security Overview publishes a provider table naming Microsoft Azure for 'Managed AI model services and document-processing capabilities used by applicable Editor and Workspace workflows', Google Cloud Platform for 'Managed AI model services used by applicable Workspace workflows', Exa for web search and Amazon Web Services for hosting of account, administration, Workspace services and storage. Clause 1.2 of the General Terms separately directs that when using the AI feature the customer must 'abide by all the applicable policies and documentations of Azure AI services'. For the other two modes the provider is the customer's: On-Device mode runs 'compatible small language models downloaded to and run on the user's device', which the product page attributes to Microsoft Foundry Local, and Custom mode uses 'a user-configured, OpenAI-compatible external model service'. No model name or version is published for Cloud mode.
Ask: Which specific models and versions serve Cloud mode for an Australian customer, in which Azure regions do they run, and what notice is given before a model or provider changes?
Subprocessor disclosure
Public claim only
Foxit publishes no dated subprocessor list. The nearest published disclosure is the provider table in the Security Overview, which names Microsoft Azure, Google Cloud Platform, Exa and Amazon Web Services and states that it 'identifies the principal providers at a high level', that the applicable commitments, 'subprocessor disclosures, processing locations, and data-processing terms are governed by the relevant customer agreement, data processing agreement, privacy notice, subprocessor disclosure, or deployment-specific documentation', and that 'Foxit reviews providers through its security, privacy, legal, and vendor-management processes'. No processing location is given for any provider, no entity details are given, and no change-notification or objection mechanism is published.
Ask: Provide the dated subprocessor disclosure referred to in the Security Overview, with each provider's processing locations for an Australian customer, and our notice and objection rights when it changes.
Training on customer content
Public claim only
The Security Overview states that 'Foxit does not, and will not authorize or instruct its subprocessors to, use customer documents, prompts, AI outputs, or other customer content submitted to or processed through the products on the customer's behalf to train or fine-tune any artificial intelligence or machine learning model, unless the customer has expressly authorized that use.' That statement sits in a document which itself provides that any applicable customer agreement, data processing agreement, privacy notice, subprocessor disclosure or product documentation controls over it. The contractual position is broader and carries no training carve-out: clause 4.1 of the General Terms grants Foxit 'a non-exclusive, worldwide, royalty-free, sub-licensable, and transferrable license to use, reproduce, publicly display, distribute, modify, publicly perform, and translate the content as needed in response to user driven actions ... and for the purpose of operating and improving the Products and/or Services'. Clause 1 of the AI User Guidelines is a restriction on the user, not a Foxit commitment: it states the user will not use the AI features 'to train artificial intelligence or machine learning models'.
Ask: Will Foxit warrant the no-training statement in our contract, and confirm that the clause 4.1 licence to use content for 'operating and improving' the products excludes any model training, fine-tuning or evaluation?
Feedback, telemetry and human review of content
Public claim only
The Security Overview states that 'Product analytics and operational telemetry are limited to information used to administer entitlements and usage, operate and secure the service, diagnose problems, provide support, and improve product operation' and that telemetry 'is governed by applicable settings, privacy notices, and agreements'. It also states that Foxit 'may apply input or output controls appropriate to the feature, model, and risk, including policy enforcement, restricted-use rules, content-safety controls, source references, user confirmations, and monitoring for abuse or security events'. The privacy policy states of files uploaded to Foxit cloud services that 'During that time, we neither look at the files nor mine any data from them' and that their content is 'not monitored without your explicit permission'.
Ask: What does abuse and security monitoring of AI requests capture and retain, who can read it, and can telemetry and any in-product feedback control be disabled for our users?
Retention of prompts, documents and outputs
Public claim only
The Security Overview states that Cloud-mode request data 'follows the applicable temporary processing lifecycle' whose 'exact interval is defined by applicable product, privacy, contractual, or deployment-specific documentation' - no interval in hours or days is published. On-Device data is 'Controlled by the local product, device, and customer-managed environment'; in Custom mode retention of requests and responses at the external endpoint is governed by the customer's own configuration and that provider's terms. Workspace content is retained until the user deletes it or the Project, Workspace or account is deleted. The privacy policy states that files uploaded to Foxit cloud services 'will be deleted once the process is completed, and the output files will typically be deleted in one day', with no backups of transitional uploads or output, while contract and business transaction data is kept for ten years from the last year of the business relationship.
Ask: State the actual retention interval in hours for a Cloud-mode prompt, the document content sent with it and the generated output, and identify the document in which that interval is defined.
Deletion and customer control
Public claim only
Clause 2.2 of the General Terms states 'You are solely responsible for deleting your account along with the related account data. You will not be able to access any data or file after the termination of account.' The privacy policy states that on a request to remove an account Foxit 'will delete personal data related to your account', with the ten-year retention of contract and business transaction data as the exception. The Security Overview states that Workspace-derived artefacts - 'Parsed document representations, retrieval indexes, embeddings, chunks, and similar processing artifacts' - are removed when their Source or Project is deleted, 'subject to applicable backup, legal, security, and compliance retention requirements'. No deletion confirmation or certificate of destruction is described, and no administrator-initiated deletion of an individual AI request is described.
Ask: Can an administrator delete all AI request data for a named user on demand, and will Foxit confirm the deletion in writing, including copies held by Microsoft Azure or Google Cloud?
Foxit and provider access to content
Public claim only
The privacy policy states of files uploaded to Foxit cloud services that 'we neither look at the files nor mine any data from them' and that 'the content of such files are not monitored without your explicit permission'. The Security Overview states that stored product data 'is protected through technical and organizational safeguards intended to limit access to authorized users, services, and personnel', and that Foxit applies 'provider review' among its controls. No break-glass or support-access process is described, no statement is made about whether Microsoft or Google personnel can read AI request content, and no customer-inspectable access log is described.
Ask: Which Foxit personnel can read a Cloud-mode AI request or its output, from which countries, under what approval, and can we obtain a record of any such access?
Administrative control and approvals
Public claim only
The Security Overview states that 'Authentication, access control, sharing, and enterprise administration options depend on the applicable product, account type, deployment, and customer agreement' and that users and organisations are responsible for managing account access and any credentials they authorise for third-party services. It adds that Foxit Workspace 'can require explicit user approval for sensitive or consequential operations, including selected client tool calls and Connector actions', with approval and tool availability depending on client, connected service, scopes, Project, account and configuration. The product page states credits pool 'across your team when managed through the Foxit Admin Console'. Which specific AI capabilities an administrator can switch off for a user is not published.
Ask: Through the Foxit Admin Console, can we disable Cloud mode, individual AI capabilities, Connectors and Workspace per user or group, and enforce single sign-on?
Encryption
Public claim only
The Security Overview states that 'Foxit uses encrypted network transport for authentication, cloud AI requests, Workspace traffic, content transfer, service integrations, and response delivery' and that 'Stored product data is protected through technical and organizational safeguards intended to limit access to authorized users, services, and personnel.' No protocol version, cipher, at-rest encryption algorithm, key-management arrangement or customer-managed key option is stated for AI Assistant in any document read.
Ask: State the transport protocol version and the at-rest encryption algorithm for AI request data and Workspace storage, who holds and rotates the keys, and whether a customer-managed key option exists.
Independent security assurance
Public claim only
The Trust Center certification list read on 2026-09-27 records SOC 2 (Type 2) against 'PDF Editor, eSign Business, eSign Essentials'; TAA Compliance and Section 508 Compliance against 'PDF Editor'; 'GDPR Support'; 'HIPAA Ready', with Foxit 'prepared to enter into a Business Associate Agreement'; 21 CFR Part 11 Support; and FINRA Support. No entry names AI Assistant, Foxit AI or Workspace, no ISO/IEC 27001 and no ISO/IEC 42001 appears, and no report, certificate or bridge letter is published or linked. The Security Overview points instead to a Foxit Development Security Overview whitepaper, which was not read.
Ask: Is AI Assistant, including the Foxit-managed cloud workflow and Workspace, inside the scope of the SOC 2 Type 2 report, and will you provide the report and its scope statement under non-disclosure?
Contracting entity, liability and forum
Public claim only
The General Terms of Service, effective 27 June 2025, are 'a legal agreement between you, or an entity you are acting on behalf of, and Foxit Software Incorporated and its authorized affiliates'; no affiliate is named. Clause 14.1 applies the law of the State of California and gives exclusive venue to 'the federal or state courts located in San Jose, California'. Clause 9 limits total liability to 'the amount actually received by Foxit from you for the Products and/or Services' and clause 8 supplies the products 'AS IS'. The AI User Guidelines add that where the no-professional-advice rule is breached 'Foxit cannot be held liable for and will not indemnify you'. No Australian contracting entity, Australian address, ACN or ABN appears in the terms, the AI User Guidelines, the privacy policy, the Trust Center, the Security Overview or the product page; the privacy policy's responsible body is Foxit Software Inc. of Fremont, California, with Foxit Europe GmbH in Berlin named for European data protection questions. No data processing agreement is published or linked on any page read.
Ask: Which entity would contract with an Australian law practice, what is its ACN or ABN, and will Foxit execute an Australian-law data processing agreement with a negotiated liability cap?
Cross-border transfers and Australian Privacy Principle 8
Public claim only
The Security Overview states that Cloud-mode AI Assistant and Workspace 'operate on the same regional Foxit cloud architecture', that 'Available regional service instances may include the United States, Canada, Australia, the European Union, and Japan', that where regional deployment is supported data 'is routed to the customer's assigned regional instance and is logically isolated', and that account email address and password are 'currently stored in the United States, regardless of user location'. It then qualifies that: 'not every provider or service is available in every Foxit region', 'If required processing occurs outside the assigned region, it is governed by applicable Foxit customer terms and Foxit's arrangements with that provider', and 'A globally available service or endpoint does not, by itself, mean that data remains in the customer's assigned region.' The privacy policy states servers are 'located in the United States, Germany, Canada, Australia, and Japan' and that for customers from Australia information 'will be stored on the local server' except account email and password.
Ask: Confirm in writing that an Australian customer is assigned the Australian instance for AI Assistant, and name every country in which a Cloud-mode prompt, the document content sent with it and the output are processed, cached or logged.
Incident notification
Not confirmed
No incident or breach notification commitment was found. The General Terms contain no notification clause, the privacy policy describes security measures and data subject rights without a notification undertaking, the Security Overview describes 'monitoring for abuse or security events' without any obligation to notify, and the Trust Center links a Security Advisories page and a Responsible Disclosure Policy that were not read within the evidence budget.
Ask: State the contractual notification period from discovery for an incident affecting AI request data, whether it extends to incidents at Microsoft Azure, Google Cloud, Exa or AWS, and what Foxit provides for a Notifiable Data Breaches assessment.
Change notice for terms, features and processing paths
Public claim only
Clause 1.3 of the General Terms provides that Foxit 'may make changes to the Agreement from time to time', that notice may be given 'by revising the date at the top of the Agreement', that 'Unless otherwise specified, the amended terms shall be effective immediately', and that continued use confirms acceptance; the AI User Guidelines are Additional Terms subject to the same clause. The Security Overview carries no version or date and states that 'Feature availability and processing paths vary by release, platform, location, account, configuration, and entitlement', so the provider and region behind a given feature can change with a release.
Ask: What advance notice will we receive before a model provider, a processing region or an AI capability changes, and will you notify us directly rather than by revising an undated page?
Export and exit
Public claim only
Clause 2.2 of the General Terms states that after termination of an account 'You will not be able to access any data or file after the termination of account', and clause 10.2 allows Foxit to terminate or suspend access 'without refunds for any pre-paid fees' in its sole discretion on breach, non-payment, illegality or where Foxit elects to discontinue or modify the products. No export mechanism is published for chat histories, custom Skills, Connector configurations or Workspace Projects, and no treatment of unused prepaid credits on exit is published.
Ask: On exit, in what format can we export chat histories, custom Skills and Workspace Projects, what happens to unused prepaid credits, and how long do we have to retrieve anything?
Connectors and third-party services
Public claim only
The Security Overview states Connectors 'are explicitly configured, tool calls are surfaced in the product experience, and selected operations can require user approval', that access is limited to the OAuth scopes, API keys or other credentials authorised by the user or organisation, and that once data reaches a customer-authorised third-party service 'that service is responsible for its processing, storage, security, and retention under the customer's authorization and the provider's terms'. Clause 1.2 of the General Terms states that third-party products and services 'including plug-ins and extensions' are not Products or Services under the agreement and that their use 'is solely between you and the third party'. Foxit separately publishes an MCP server page offering to connect AI agents to Foxit document capabilities.
Ask: Which Connectors can our users enable in the release we would deploy, can an administrator restrict the list, and what is logged when a Connector or the MCP server is used?
Open-web retrieval
Public claim only
The Security Overview names the web-search provider and when it is used: Exa supplies 'Web-search capabilities used when a user invokes a feature requiring current public-web information', receiving 'Search queries and task context needed to retrieve and return relevant public-web results'. It also lists 'web-search queries' among tool and connector data whose purpose is to 'Execute user-requested searches, tool calls, external model requests, or connected-service actions'. No processing location is given for Exa, and no administrative control over web retrieval is described.
Ask: Which AI Assistant actions send a query to Exa, what does that query contain from our document, in which country is it processed, and can web retrieval be turned off for our users?
Actions the assistant can take on a document
Public claim only
The product page states that in Foxit PDF Editor on desktop the assistant performs edits described in plain language 'including redacting sensitive data, reordering pages, applying watermarks, adding page numbers, converting file formats, and protecting files', and gives as examples 'Redact every Social Security number in a contract' and 'Add a watermark to all 200 pages'. The Security Overview states for agentic and tool-enabled workflows that 'Consequential document actions can require user review or confirmation', and for Workspace that it 'can require explicit user approval for sensitive or consequential operations' and that 'Artifact changes can be presented for review before the revised output is retained'. Which actions are gated, and whether an edit can be undone, is not published.
Ask: Which chat-invoked actions are classified as consequential and therefore confirmed, is each edit reversible, and can redaction by chat be disabled for our users?
Human review before use
Public claim only
The obligation is placed on the user by contract: clause 4.3 of the General Terms states 'You shall evaluate the accuracy of any Output as appropriate for your use case, including by using a human review of the Output', that the user is solely responsible for loss from using the output, and that the user undertakes 'not to claim or assert any rights against Foxit or AI service provider'. The AI User Guidelines state 'Be guided by your judgment when reviewing and validating the content of the generated outputs. Outputs may contain hateful, abusive, profane, inaccurate or ungrounded, misleading or false information.' The Security Overview states the product 'is not intended to make decisions about individuals that produce legal or similarly significant effects without meaningful human review'. No product control records that a review occurred.
Ask: Does the product provide any record that a named person reviewed an AI output before it was applied to a document or sent outside the firm, and can review be made mandatory for named actions?
