Privacy Policy

Privacy Policy

How we collect, use, and protect your personal information across the Difinity platform.

Version: 2.3 | Effective Date: 13 November 2026 | Last Updated: 13 November 2026

Version 1.x, effective 1 March 2026, was replaced by this version.

Difinity Pty Ltd (ABN 82 686 692 759)


This Privacy Policy explains how Difinity Pty Ltd ("Difinity", "we", "us" or "our") handles personal information through difinity.ai, the Difinity.ai governed chat and agent platform, the chat and agent workspace at chat.difinity.ai, Hub, the Platform API, Flow, the tool gateway, our public tools, and related support and business communications (together, the "Services").

This Policy applies when Difinity handles personal information as a controller or business for its own purposes. When a customer determines why and how Customer Data is processed through the Platform, Difinity generally acts as that customer's processor or service provider. The Data Processing Agreement governs that processing.

Difinity handles personal information in line with the Australian Privacy Principles in the Privacy Act 1988 (Cth), whether or not the Act requires us to. Where they apply, we also follow the EU GDPR, UK GDPR, Canadian privacy law, United States state privacy laws and applicable privacy laws in the Middle East. Local law may give you additional rights.


1. Who is responsible

The entity responsible for this Policy is:

1.1 EU and UK representatives

Difinity has not yet appointed a representative in the European Union or the United Kingdom. Contact privacy@difinity.ai for any request.

1.2 Privacy contact

Difinity has not identified a statutory Data Protection Officer in this Policy. Contact privacy@difinity.ai to exercise a privacy right or ask about our handling of personal information.


2. Our roles

Difinity acts as a controller or business for account administration, authentication administration, billing, security, support, our website, public tools, marketing and our own business records.

Difinity acts as a processor, service provider or sub-processor when it handles Customer Data on a customer's documented instructions to provide the Platform. If the customer is itself a processor, Difinity acts as its sub-processor. The customer remains responsible for its own legal basis, notices, instructions, connected-system authority and configuration. Difinity remains responsible for the duties that apply to it.

A connected model provider or business system may instead be in a direct relationship with the customer. This is the case when a customer supplies its own provider key or appoints a connector destination. The DPA and Sub-processors page explain the distinction.


3. Personal information we handle

What we handle depends on the Services a person or customer uses.

3.1 Account and authentication information

This can include name, business email address, organisation, role, telephone number if provided, account and organisation identifiers, permissions, login timestamps, authentication identifiers, session and refresh-token data, and security events. Supabase Auth provides authentication. This Policy does not say that Difinity stores a password chosen by a person.

3.2 Platform configuration and administration

This can include organisation settings, people and groups, agent identities and instructions, policies, use cases, model settings, connector and MCP server configuration, enabled tools, scopes, permissions, approval requirements, credential references and administration activity.

A credential reference identifies a stored credential without revealing it. Provider keys and connector credentials are write-only through customer-facing interfaces, and no stored credential is returned through those interfaces. The tool gateway may hold the credential needed to carry out a customer-authorised action.

3.3 Conversations, agents and governed runs

Customer Data can include:

  • conversation messages and transcripts;
  • attachments, including file bytes, file names, types, sizes, extracted text, ownership and expiry metadata;
  • prompts, responses and agent instructions;
  • model and tool arguments, results and error information;
  • proposed and completed actions in connected systems;
  • policy decisions, personal-information control results and refusal reasons;
  • approvals, including who was permitted to decide and the decision made;
  • run status, timing, model and provider metadata, usage and cost data; and
  • the separate append-only Run Trail, which records available evidence about a governed run.

The Platform API stores conversations, messages, configuration, approvals, Run Trails, usage and related records. Flow applies configured controls and runs model and agent work. The tool gateway holds connector credentials, decides whether a proposed action may proceed and acts on the connected system. It runs in private subnets behind an internal load balancer, with no public address, and accepts traffic only from that load balancer. Where detection and replacement are enabled, tool arguments and results are recorded in the Run Trail as redacted stand-ins, not the original detected personal values. If detection is turned off, the Run Trail stores the raw text. The tool gateway also keeps a digest of a tool result, and the Run Trail records a reference to the gateway's record. Separately, the tool gateway keeps its own record of submitted calls, including refused and automatically allowed ones, with their arguments, and a memory of standing approvals as the action ran. That record is not redacted and has no automatic expiry.

3.4 Billing and credits

This can include the billing contact, Stripe customer and transaction identifiers, invoice information, Credits ledger entries, Usage Events, provider-cost records, taxes, payment status and related account records. Paid plans are charged by card through Stripe, with a tax invoice, and GST is added for Australian organisations. Credits are a prepaid ledger for eligible model and tool work funded by Difinity, with usage settled shortly after each run.

3.5 Website, public tools and marketing

When a person visits difinity.ai, requests a demo, subscribes, or uses a public tool, we can handle information they enter, such as name, business contact details, organisation, role, answers and free-text responses. We also handle the generated result, report or PDF, delivery status, consent choices and related communications.

Demo sessions. The "See it working" demo on the Site lets a visitor try a sample agent without an account. A demo session ends 24 hours after it starts. When it has ended, we delete what the session created: the conversation and its messages, the copy of the sample agent and its use case, and any files uploaded or created in the session, including the stored copies in AWS S3. Finally we delete the session record itself, including the hashed network address used to limit use. A cleanup job runs every 15 minutes, so deletion follows shortly after the 24 hours are up. If a visitor gives an email address in the demo and agrees to be contacted, that address is also sent to our CRM and to our sales team by email. Those copies are not part of the session and are kept under our usual CRM and email retention, not deleted with the session.

Our public tools use Amazon Web Services, including DynamoDB, S3, Bedrock and SES, to receive answers, generate and store results, and deliver requested reports. Detailed contact and answer summaries may be written to HubSpot as CRM contacts and sales notes. The Cookie Policy explains optional analytics and marketing technologies.

3.6 Communications and support

This can include contact details, the contents of an enquiry or support request, attachments, investigation notes, the organisation concerned and actions taken to respond.

3.7 Website device and event information

Before Analytics consent, PostHog captures no analytics event and stores nothing on the device. On each page load, the Site makes one request through its proxy to PostHog Cloud EU for the project configuration that shapes the page. The request does not repeat while that page stays open. It has no body and no visitor identifier beyond the public project token. As with any web request through the proxy, PostHog Cloud EU receives the IP address, page URL and browser details. No feature-flag evaluation request or analytics event is sent before consent. We rely on our legitimate interests in loading the Site's feature configuration for this request.

After Analytics consent, PostHog can receive event-level information such as the page path, referring information, screen dimensions, user agent, clicks, scroll depth, page exits and engagement timing. Click and copy events send the full page URL, including any query string. A click event can also send the element's ID, classes, selector and page section. A click event can include up to 50 characters of text from the clicked interface element. A copy event records that text was copied, the length of that text and whether it looked like code. It never includes the copied text itself. PostHog can also receive a replay of configured page interactions. PostHog uses local and session storage described in the Cookie Policy, not a PostHog cookie. Rejecting or withdrawing Analytics consent stops capture, clears PostHog local and session storage, and removes difinity_has_visited. PostHog data is processed through PostHog Cloud EU in the European Union (Frankfurt).

Google Analytics, when enabled, loads only after Analytics consent is given. Meta Pixel, when enabled, loads only after Marketing consent is given. The website uses analytics and marketing cookies and similar technologies only with your consent, as described in the Cookie Policy, which identifies their browser technologies, destinations and retention.

3.8 Agent features and what they keep

Agents can use memory, a knowledge library, a code sandbox, triggers and webhooks, alerts, sharing, artifacts, OpenAPI connectors and conversation-access grants. A built-in web search tool sends search queries to Exa. The table says, for each feature, what is stored, how long it is kept and who can see it.

Feature What is stored How long Who can see it
Agent memory Short notes an agent keeps between chats. A personal note belongs to one person for one agent. A shared note belongs to the agent. The note text is stored sealed. A second copy, with common personal details such as emails, phone numbers, card numbers and IP addresses replaced, is turned into a search vector by Amazon Titan on AWS Bedrock in Sydney. Names and addresses typed into a note may not be caught. Until deleted. A person's personal notes go when they leave the organisation. All notes are erased with the organisation. Personal notes: only that person. Shared notes: used by the agent in any run, and managed by people who may edit the agent.
Knowledge Files uploaded to an agent (up to 50 MB each) are stored in AWS S3 in Sydney, converted to text, cut into chunks and embedded with Amazon Titan in Sydney. Connected sources (Confluence, Notion, SharePoint, Google Docs) are searched live with the person's own connection and are not copied; the question is sent to that app as written. Websites are searched through the built-in web search tool, which sends the question to Exa. Until the file or source is deleted or the agent is archived. Erased with the organisation. People who may use the agent get answers drawn from it. People who may edit the agent manage the sources.
Agent sandbox Code an agent writes runs in an AWS Bedrock AgentCore code interpreter. Difinity keeps a session record: task, agent, version, internet mode, timing and call counts. Files made in the sandbox exist only for that run unless the agent shares them as artifacts. A session lasts at most 15 minutes. Session records are held by the tool gateway and are not removed when an organisation is erased. The agent's builder sets the internet mode. Difinity operators can switch the sandbox off or limit it to no internet for the whole service.
Triggers and webhooks A schedule or an inbound webhook starts an agent with nobody signed in. The run acts as the agent's own identity, not as a person. A webhook is checked by its signature. Data delivered by a webhook is cut to 16 KB and held only until the run starts. The run record is kept until the organisation is erased. A trigger pauses if its owner leaves the organisation. The trigger's owner and anyone who can view the agent.
Alerts An alert for a failed run, a waiting approval, a refusal spike, a paused trigger, a changed tool or low credits. It appears in the app and is also emailed through Amazon SES when email is switched on. The email holds the agent name, counts, deadlines, tool or connector names and the credit balance. It never holds message text or tool arguments. It is addressed to the recipient, and it shows the names that administrators typed for agents, tools and connectors. Erased with the organisation. The members named on each alert rule. Each can unsubscribe.
Sharing An agent can be shared with named people, or by a link. Anyone signed in to the same organisation who opens the link becomes a user of that agent. A link has no expiry. It works until the owner revokes it. Only people in the same organisation.
Artifacts A file an agent shares from its sandbox (up to 500 MB) is stored in AWS S3 in Sydney, encrypted with a key tied to the organisation. A preview runs in a locked-down frame that cannot run scripts. 28 days, even if a legal hold applies. Whoever may read that conversation.
OpenAPI connectors An administrator adds a REST API by its OpenAPI document (up to 2 MB). The tool definitions made from it are stored. The organisation's one credential for that API is held as a write-only credential. Until the connector is removed. Erased with the organisation. The raw OpenAPI document is also kept. Administrators manage it. Calls go from the tool gateway directly to the address the administrator registered, which is the customer's own relationship.
Conversation-access grants An organisation administrator can open one person's conversations to a named person for a written reason. The grant, each read, any revocation and its expiry are written to the audit trail. A grant lasts at most 7 days and can be revoked early. The named person can read that person's conversation content for the period. Difinity staff cannot grant access.
AI disclosure For email, chat and support tools, the tool gateway adds the line "This message was sent by an AI agent." to the message before it is approved and sent. It is set per use case. It is not added by custom MCP servers, OpenAPI tools or apps outside the covered list. Part of the message sent. The approver sees it before the message goes.

Search queries sent to Exa may contain personal information. Difinity does not redact personal information from search queries, because doing so would stop the search returning correct results. Exa Labs, Inc. is based in the United States and does not publish where it processes queries, so we say only that processing is outside Australia.

Apps connected through the Composio broker send the arguments and results of each call to Composio in the United States (us-east-1), and Composio holds the linked account and its tokens.

The agent sandbox has three internet modes, chosen per agent version: no internet (the default), listed websites only (not yet available in production), and Full internet. In Full internet mode, code in the sandbox can send data to any website, without passing through the tool gateway.

3.9 How we collect personal information

We collect personal information (a) from you directly, for example when you sign up, request a demo, use a public tool or contact us; (b) from your organisation, when an administrator adds you to it; (c) automatically, through the Services and the Site; and (d) from public sources and other third parties.


4. Why we handle personal information

We handle personal information to:

  • provide, administer and secure the Services;
  • authenticate people and manage organisations, roles and permissions;
  • follow customer instructions for conversations, agents, policies, approvals and connected-system actions;
  • detect and respond to misuse, fraud, security events and service failures;
  • calculate usage, maintain the Credits ledger, invoice and keep financial records;
  • answer enquiries, provide support and administer customer relationships;
  • generate and deliver a public-tool result requested by a person;
  • send marketing where permitted and manage consent, opt-out and suppression records;
  • understand and improve our website and Services where the applicable legal basis permits; and
  • comply with law and establish, exercise or defend legal claims.

Governed run records can contribute operational evidence to wider EU AI Act, ISO/IEC 42001, risk, and audit processes. Difinity does not determine that an organisation or AI system is compliant, and it does not provide ISO/IEC 42001 certification.

Difinity may use data that has been irreversibly anonymised so that a person or customer cannot reasonably be identified. We do not treat replaced, pseudonymised or merely aggregated data as irreversibly anonymised. We do not use Customer content to train a general model or publish a Customer-derived benchmark without the customer's prior written agreement.


5. Legal bases and privacy grounds

Where EU or UK law applies, our legal basis depends on the activity:

  • Contract: account administration, requested Services, billing, transactional messages and support needed to perform a contract or take requested pre-contract steps.
  • Consent: optional Analytics and Marketing technologies, marketing subscriptions and another activity where we ask for consent. Consent can be withdrawn at any time.
  • Legitimate interests: service security, fraud prevention, necessary business administration, responding to business enquiries, loading the Site's feature configuration before an Analytics choice and improving the Services where those interests are not overridden by a person's rights.
  • Legal obligation: tax, accounting, regulatory, sanctions, security-notification and lawful-request duties.
  • Legal claims: establishing, exercising or defending a legal claim where applicable law permits.

Under the Australian Privacy Act, we collect personal information that is reasonably necessary for our functions or activities by lawful and fair means. We collect sensitive information only with consent or where an exception applies. Under PIPEDA and substantially similar Canadian laws, we obtain meaningful consent where required. United States and Middle East privacy grounds apply according to the relevant law and activity.


6. Personal-information controls

Personal-information controls depend on customer configuration and the workflow. A policy may detect and replace supported values, block a request, use a detect-only mode or permit the data. Detection is not guaranteed to identify every personal value.

Where detection and replacement are configured, AWS Bedrock using Qwen currently processes the original content as the personal-information detector. The generation model works on replacement values. The redaction map remains within the Difinity processor and sub-processor boundary while needed for the run, including while an approval is pending.

A tool result is checked for personal information before the generation model reads it. If the check fails, the result is withheld and the run continues without it, rather than sending unchecked content onward. The Run Trail records the kind of personal value and its replacement, never the original value.

For a permitted external action, real values are restored as the last step before the action leaves the tool gateway. If policy requires an Approval, the person sees and approves the restored action that will run. These controls are detection and replacement, not deletion or anonymisation.


7. Who receives personal information

We do not sell personal information for money. We disclose it only for the purposes described here, on a customer's instruction, with a person's direction or consent, or where law permits or requires it.

7.1 Services sub-processors

The current Sub-processors page gives the applicable legal entity, purpose, data, region and transfer details. The Services can use:

  • Amazon Web Services Australia Pty Ltd, including AWS Bedrock, for infrastructure and configured model or detector processing;
  • Composio, in the United States (us-east-1), for brokered connector calls, linked accounts and their tokens;
  • Exa, for the built-in web search and research tool;
  • Supabase Pte. Ltd. for authentication;
  • Stripe Payments Australia Pty Ltd for billing;
  • Google Cloud (Vertex AI), for Gemini models when Difinity funds the call; and
  • OpenAI OpCo, LLC; Anthropic, PBC; or SpaceXAI (formerly xAI), conditionally, only when Difinity funds a call that goes to the vendor directly. Most calls to these models run on AWS Bedrock instead.

When a customer uses its own model-provider key, that provider is the customer's own provider relationship rather than a Difinity sub-processor for that call.

7.2 Customer-authorised destinations

An agent can act through connectors or customer-configured MCP servers. Personal information in an action or tool call may be sent to the business system or provider authorised by the customer. The customer chooses the destination, account, tool and scope. The tool gateway holds the applicable credential and acts; the agent does not hold the credential or connect directly.

7.3 Website and business vendors

Website and business vendors are separate from the Services sub-processors listed at Sub-processors and are not covered by the DPA merely because they support Difinity's controller activity. They include:

  • Amazon Web Services for public tools, report generation, storage and email delivery;
  • HubSpot for CRM and sales operations;
  • PostHog for one project-configuration request per page load before consent and for consented analytics, processed through PostHog Cloud EU in the European Union (Frankfurt);
  • Google LLC for Google Analytics when enabled and consented, with processing in the United States and other countries where Google operates under its Data Processing Terms;
  • Meta Platforms Ireland Limited for Meta Pixel when enabled and consented, including transfers described in the Meta Platforms Privacy Policy and Meta Platforms Business Tools Terms.

7.4 Difinity staff access

Authorised Difinity staff use a separate, role-gated administration surface. They can administer organisations and people, billing, credits and plans, and view aggregate usage counts. They have no access through that surface to conversations, transcripts, run trails, agent definitions, prompts or uploaded files, and they do not sign in as one of an organisation's people.

Staff can be added to an organisation only in a non-admin role, and an added member holds only the permissions of that role. Every staff action on an organisation, including an add, is recorded in a privileged-access log that identifies the staff member and organisation. The application cannot update or delete these entries. The log is deleted together with the organisation when the organisation is erased. It is available to authorised Difinity staff and is not currently visible through customer-facing interfaces. Staff access remains subject to confidentiality and purpose limits. Separately from that surface, a limited number of Difinity engineers hold administrator access to our cloud infrastructure so they can support and maintain the Services and respond to incidents. That access can technically reach stored data. AWS CloudTrail records the management actions taken in the account, but it does not record every read of stored data. The access is used only for those purposes and is limited to those engineers, who are bound by confidentiality duties.

7.5 Operational logs

Difinity's application and gateway logs hold metadata only, such as ids, timing, status and error types, not customer content.

7.6 Law, safety and corporate transactions

We may disclose personal information when legally required, to protect rights and safety, investigate misuse or security incidents, or establish and defend legal claims. Personal information may also transfer as part of a merger, financing, acquisition, reorganisation or sale, subject to applicable notice and protection requirements.


8. Where information is processed

Difinity is headquartered in Sydney, Australia. Difinity hosts the Platform in AWS Australia (Sydney, ap-southeast-2). That is the only live region today. Hosting in European Union (Frankfurt) and United States regions is available by arrangement: a dedicated deployment is set up when a customer signs. Model and connector calls can leave the hosting region. Some models run on AWS or Google Cloud routing that is not limited to Australia: Claude Fable 5.2 and Claude Opus 4.5, OpenAI and xAI Grok models on AWS Bedrock, and Gemini models on Google Cloud Vertex AI. Requests to those models can be processed in regions outside Australia.

Hosting in Sydney does not mean that all personal information stays in Australia. Some processing happens outside Australia.

Difinity is likely to disclose personal information to recipients in the United States (including Composio, Exa, OpenAI, Anthropic, Google, SpaceXAI, Stripe and HubSpot), in Ireland (Meta Platforms Ireland Limited) and in Germany (PostHog Cloud EU, Frankfurt), and in other countries where those providers operate. Exa does not publish where it processes queries, so we say only that processing is outside Australia. Where the customer enables a model provider's models for a use case, personal information in that call can go to that provider's processing locations. Before we disclose personal information to an overseas recipient, we take the steps that are reasonable in the circumstances to make sure it does not breach the Australian Privacy Principles in relation to that information. For a Services sub-processor this means written data-protection terms, listed on the Sub-processors page. We may also rely on another exception in APP 8.2.

Supabase authentication data is processed in the same region as the organisation's hosting region. Services sub-processor destinations are listed at Sub-processors. Website and business data is processed in the vendor destinations identified in Section 7.3 and the Cookie Policy. A connected system or customer-appointed provider processes data in the location chosen for that relationship.

For restricted transfers of Customer Data from the EEA, the DPA uses the EU Standard Contractual Clauses, Module 2 for transfers to Difinity and Module 3 for onward transfers to Sub-processors, where required. For restricted transfers from the UK it uses the applicable UK Addendum or International Data Transfer Agreement. We use transfer assessments and supplementary safeguards where required. We address overseas disclosures under Australian Privacy Principle 8 and use the contractual or statutory mechanism required by applicable Canadian, United States or Middle East law.


9. How long we keep information

We keep personal information only for the period stated below, for the agreed customer configuration, or for as long as an objective legal or operational need applies.

Category Retention period or criterion
Chat attachments and artifacts 28 days, even under a legal hold.
Parked agent-run state awaiting completion Unreadable after 7 days, or one day after the approval deadline if that is later, and never more than 15 days.
Agent memory notes and knowledge files Until deleted. Erased with the organisation.
Trigger run records and alert history Until the organisation is erased.
Agent share links Until the owner revokes them.
Conversation-access grants Up to 7 days, then ended. The record stays in the audit trail.
Operational logs They hold metadata only (ids, timing, status, error types) and no customer content.
Tool gateway call records and standing-approval memory No automatic expiry. Removed on request to privacy@difinity.ai, as described below.
Database backups 7 days.
Run Trails Each organisation sets its own period. The default is seven years and the minimum is six months. When the period ends the records are deleted. After termination, the organisation can export them for 90 days. Deleting the organisation does not delete its Run Trails early; they stay until their own period ends.
Transcripts and archived conversations Kept while the organisation is subscribed, unless it sets a number of days. After termination, the organisation can export them for 90 days, then they are erased. Archiving a conversation does not delete it.
Connector credentials Until the connection is revoked, then the stored connector credential is destroyed.
Customer model-provider keys Where an organisation's own provider keys are enabled for it, until the customer deletes or replaces the key or the account is closed, subject to backup ageing and legal obligations.
Account and authentication records While the account is active, then while needed for security, disputes or legal obligations.
Billing and tax records For the period required by applicable tax, accounting and corporate law.
Demo session content (conversation, sample-agent copy, files, session record) Deleted after the session ends at 24 hours, by a job that runs every 15 minutes.
Public-tool answers and results No automatic deletion. Kept until deleted on request to privacy@difinity.ai.
Public-tool PDF reports 180 days, then deleted.
Public-tool email delivery records No automatic deletion. Kept until deleted on request to privacy@difinity.ai.
Public-tool confirmation links 24 hours for the classifier tool and 7 days for the penalty tool.
Website function logs They contain no form text or email addresses.
PostHog analytics events and session recordings Kept for 12 months. Browser storage is kept until consent is withdrawn or the person clears browser storage; no fixed expiry is set.
Google Analytics data, when enabled Browser identifiers follow the periods in the Cookie Policy. Event data used in Explorations is kept for 2 months by default and can be configured for 14 months.
Meta Pixel data, when enabled Browser identifiers are kept as set by Meta Platforms under its Meta Platforms Cookie Policy. Meta Platforms may retain Event Data for up to 2 years under its Meta Platforms Business Tools Terms.
Marketing contacts, consent and suppression records While subscribed or engaged, then while needed to honour an opt-out, show consent history or meet a legal obligation.
Support and business correspondence While needed to answer the matter, administer the relationship, resolve a dispute or meet a legal obligation.

When we are a processor, an organisation can use the self-service export during its subscription and for 90 days after termination. The export contains Transcripts and the Run Trail only. Transcripts are limited to conversations that the person running the export owns or has been given live access to; an administrator cannot export other people's private conversations. The Run Trail export covers the whole organisation. At the end of that window, our scheduled job deletes the organisation's database records, including conversations, messages, agents, knowledge and memory records, connector records, people, billing records and the privileged-access log.

The job does not delete run trails. A run trail, and the sealed values behind its stand-ins, stays until its own retention period ends (7 years by default).

The job does not remove everything else either. We remove files in storage and stored secrets by hand within 30 days of the end of the window, working from a recorded list. We do not ask Stripe to delete the customer record, which Stripe keeps under its own terms.

Linked accounts held by Composio and the tool gateway's own records of calls, including refused and automatically allowed calls, and its standing-approval memory, which hold tool arguments and have no automatic expiry, are not removed by the job; we will remove them within 30 days of a request to privacy@difinity.ai.

Login accounts held by Supabase are closed by the scheduled job. We do not guarantee that the closed record is deleted; on request to privacy@difinity.ai, we will ask Supabase to delete it. Chat attachments and artifacts expire after 28 days, even under a legal hold.

Database backups containing erased data age out seven days later; older versions of stored files are removed by hand. Connector credentials are destroyed at revocation. Legal holds and mandatory recordkeeping are excepted, apart from the 28-day storage rule above. The DPA governs the process.

The Platform has no self-service conversation-delete feature. A person can still request erasure where applicable; the outcome depends on the controller's instructions, legal exceptions and the separate lifecycle of the Transcript and Run Trail.


10. Your privacy rights

Rights depend on your location, our role and the reason for processing. Contact privacy@difinity.ai. We may need information to verify identity and authority. If Difinity handles the relevant data only for a customer, we will tell you who to ask, and assist as the DPA requires.

10.1 Australia

You may ask us for access to, or correction of, personal information we hold about you (Australian Privacy Principles 12 and 13) by writing to privacy@difinity.ai. We will respond within 30 days. We do not charge for making a request. If we refuse, we will tell you why in writing, unless that would be unreasonable, and how to complain. If we correct information, we will tell any other entity we disclosed it to if you ask and it is practicable. If we hold the information only for a customer, we will tell you who to ask. If we also hold or control it, we will deal with your request ourselves. You may complain to us about our handling of personal information. If you are not satisfied with our response to a complaint, you may contact the Office of the Australian Information Commissioner at oaic.gov.au.

10.2 EEA and United Kingdom

Where the EU GDPR or UK GDPR applies, you may have rights to access, rectification, erasure, restriction, portability, objection, withdrawal of consent and protection from a decision based solely on automated processing that produces legal or similarly significant effects. You may complain to your local supervisory authority.

10.3 United States

Residents of California and other states with comprehensive privacy laws may have rights to know or access, correct, delete, obtain a portable copy, opt out of specified sale, sharing, targeted advertising or profiling, limit specified use of sensitive information, appeal a decision, and receive equal service. The precise rights and exceptions depend on the applicable law. We do not discriminate against a person for exercising a privacy right.

10.4 Canada and the Middle East

Where applicable Canadian law applies, you may request access and correction, challenge compliance and withdraw consent subject to legal or contractual limits. Where UAE, DIFC, ADGM or Saudi law applies, you may have rights to access, correction, erasure, restriction, objection, portability and complaint, subject to the relevant statute.

10.5 Marketing and cookies

Use the unsubscribe link in a marketing email to stop that marketing. Use Cookie settings on the Site to change Analytics or Marketing consent. Withdrawing consent does not affect processing that was lawful before withdrawal.

We respond within the period required by applicable law. Some rights are subject to exceptions, and we will explain a refusal where the law requires it.


11. Security

We use technical and organisational measures intended to protect personal information from unauthorised access, use, change, disclosure and loss. Measures for Customer Data are described in the DPA. They include scoped identity and role controls, write-only credential handling, secrets storage, service-to-service boundaries, configured personal-information checks and separation between the Transcript and Run Trail.

No transmission or storage method is completely secure. If we have reasonable grounds to believe an eligible data breach has happened, we will tell the Office of the Australian Information Commissioner and the people affected, as the law requires. Contact security@difinity.ai or legal@difinity.ai if you believe personal information or an account is at risk.


12. Automated processing and agent actions

The Platform performs several distinct kinds of automated processing. A configured policy can permit, refuse or route work. A model generates output. An agent may propose an action in a connected customer system, and the tool gateway decides whether it may proceed and carries out a permitted action.

Human approval is configurable and is not applied to every action. Where policy requires approval, the person the agent is acting for approves the proposed action, unless the agent's author has named one to ten other people for that tool, in which case any one of them can decide. The customer determines the purpose of its deployment and whether a workflow could produce a legal or similarly significant effect. The customer must select an appropriate legal basis, notices and safeguards. Difinity remains responsible for faithfully implementing documented instructions and its own processor or controller duties.

Difinity's own systems also make one decision using personal information: they refuse spending on Difinity-funded work when a self-serve plan payment is overdue, using account, plan and billing status. A person at Difinity can review that decision. Ask at privacy@difinity.ai.


13. Children

The Services are not directed to children under 18, and we do not knowingly collect their personal information through our own website. Customers must not configure the Platform to process children's information unless the processing is lawful, documented and covered by the agreement with Difinity. Contact privacy@difinity.ai if you believe a child has provided information to us directly.


14. Third-party links

The Site and Services may link to third-party websites or services. Their privacy practices are governed by their own notices unless the DPA states otherwise for a Services sub-processor.


15. Changes to this policy

We may update this Policy when our Services, practices or legal duties change. We will post the new version and Last Updated date and provide any additional notice required by law or contract. Earlier versions should be requested from legal@difinity.ai.


16. Contact and complaints

Contact:

We will acknowledge and investigate a privacy complaint and respond within the period required by applicable law. You may also complain to the regulator with jurisdiction over your complaint.


© 2026 Difinity Pty Ltd. All rights reserved.