Beta Program | Last updated: November 2026
This document is intended to give prospective beta customers a clear, plain-language explanation of how InPolicy AI handles data. For full contractual details, our Data Processing Agreement (DPA) is available at https://inpolicy.ai/legal/dpa. This document is provided for informational purposes only and does not create any contractual obligations beyond those set forth in InPolicy’s Terms of Service, Privacy Policy, Data Processing Agreement, Security Addendum, Acceptable Use Policy, and Cookie Policy (collectively, the “Legal Documents”). In the event of any inconsistency between this document and any Legal Document, the applicable Legal Document shall control. All Legal Documents are available at https://inpolicy.ai/legal.
1. What Data InPolicy Collects and Stores
Account and Organization Data
When a user creates an account, InPolicy stores basic account information (name, email address, role) and creates an organizational profile. This data is used solely to manage access and deliver the service. InPolicy does not use this data for any purpose other than operating the Service, communicating with you about the Service, and complying with legal obligations. This data is encrypted at rest and in transit, and access is limited to authorized personnel on a need-to-know basis.
Policy Documents
Organizations configure InPolicy by uploading policy documents (PDFs, URLs, or other files) that define their compliance requirements. These documents are stored securely on Google Cloud Platform (GCP) infrastructure located in the United States. Policy documents are owned by the customer organization and are treated as confidential.
InPolicy staff will not access your policy documents unless you explicitly request it (for example, as part of a supported onboarding or customer service engagement). Any such access is logged and traceable. Policy documents are encrypted at rest using AES-256 encryption and in transit using TLS 1.2 or higher. Access to policy documents is controlled through role-based access controls, and all access events are logged with timestamps, user identity, and the nature of the access. Logs are retained for audit and compliance purposes. InPolicy does not share your policy documents with any third party, except as necessary to provide the Service through its subprocessors as described in Section 5.
2. What Data InPolicy Does NOT Store
User Content
InPolicy is designed so that the content your employees are working on — emails, documents, messages — is never stored by InPolicy. This is a core architectural principle, not just a policy commitment.
All user content is processed transiently: it is received, analyzed, and a response is returned. Nothing is written to persistent storage. To be specific: user content exists only in volatile memory (RAM) during the processing cycle. It is not written to disk, not cached, not logged, and not retained in any database or file system. Once the analysis is complete and the response is returned, the content is released from memory. InPolicy’s AI inference subprocessors (Google Vertex AI and TypeSafe AI) operate under written terms that prohibit using content for model training and allow no retention beyond what is needed to return the result and meet legal obligations. Google’s terms also prohibit logging content for any purpose other than returning the inference result; TypeSafe’s terms allow it to derive non-content telemetry (technical logs, hashes, statistics and classifications) to operate and improve its service.
Policy state summaries — system-generated records of which policies are active and applicable to a given conversation, containing no message text or personal information — do not constitute Content and may be retained for the duration of an active enforcement session, subject to a maximum of 30 days.
Violation Logs
When a policy violation is detected, InPolicy logs anonymous metadata only: timestamp, the policy rule triggered and its version number, the surface it was found on (for example email or document), severity level, and engagement signals (e.g. whether the user accepted, dismissed, or resolved the suggested change). The same metadata is recorded while a workspace is in its setup stage, when InPolicy checks what people write but shows no suggestions, so that a rule that fires too often can be corrected before anyone sees it. This metadata is not linked to any individual user identity — it cannot be used to identify who was working on a document or what they were writing. It is retained for aggregate analytics purposes only and is not used for individual monitoring or follow-up. To be clear: violation metadata does not contain any text from the document being reviewed, any identifier of the individual user, or any information that could be used to reconstruct the content of the communication. InPolicy applies technical safeguards to ensure that violation metadata cannot be re-linked to individual users or specific content. This metadata is used exclusively to generate aggregate reports for customer organizations (e.g., total violations by policy category, resolution rates) and to improve the accuracy of InPolicy’s detection capabilities.
3. How Data Flows Through InPolicy
Browser Extension
- The extension extracts text from the document, email or chat message the user is actively working on, together with its conversation context: the earlier messages in the open email thread, or up to 200 recent messages (including other participants' messages) in an open chat conversation, each with the author's display name, its time, whether the author is outside the organization, and any attachment names. Conversation context is held only in the browser's memory until the browser is closed or the user signs out.
- That text is transmitted to the InPolicy policy violation endpoint over an encrypted connection (TLS).
- InPolicy's backend selects the relevant policy context for that organization and constructs a prompt.
- The text, its message header (sender, recipients, subject), its conversation context, the applicable policy rules and any relevant Tenant Knowledge Base facts are sent to our AI inference providers (Google, via Vertex AI, and TypeSafe AI) over encrypted connections.
- The LLM response (violation analysis and suggestions) is returned to the user's browser.
No user content is stored at any point in this flow. All data transmitted between the browser extension and InPolicy’s servers is encrypted using TLS 1.2 or higher. All data transmitted between InPolicy’s servers and our AI inference providers is similarly encrypted in transit. InPolicy’s backend infrastructure runs on Google Cloud Platform with network-level isolation, and all inter-service communication within InPolicy’s backend is encrypted.
Policy Bot (Google Docs Agent)
- The user shares a Google Doc with the InPolicy Policy Bot's dedicated account for their organization.
- The Policy Bot opens the document and extracts its text.
- The text is processed through the same policy violation endpoint described above.
- The Policy Bot uses document automation to leave comments and suggestions directly in the Google Doc.
- The Policy Bot removes the document from its Google Drive immediately after processing.
A note on Google Docs sharing: When a document is shared with the Policy Bot account and subsequently removed from the bot's Drive, the sharing permission technically persists in Google's system until the document owner revokes it. InPolicy has no mechanism to enumerate or discover documents accessible to that account — exploitation would require an attacker to independently know a specific document ID. We recommend users revoke sharing access after processing is complete, and we are evaluating additional controls to automate this step. InPolicy recommends that organizations establish an internal process for revoking sharing permissions after each processing cycle. The Policy Bot account is secured with multi-factor authentication, and is monitored for unauthorized access attempts. InPolicy will notify customers promptly if any unauthorized access to the Policy Bot account is detected.
Policy Bot (Microsoft 365 Agent)
- The user shares a Word document or file from their Microsoft 365 tenant with the InPolicy Policy Bot's dedicated Microsoft 365 account for their organization.
- The Policy Bot opens the document and extracts its text.
- The text is processed through the same policy violation endpoint described above.
- The Policy Bot leaves comments and suggestions directly in the document.
- The Policy Bot removes its own access to the document immediately after processing; the document is never copied into InPolicy's tenant.
The Policy Bot account for Microsoft 365 is secured with multi-factor authentication and is monitored for unauthorized access attempts. Sharing notifications and document metadata (sharer name and email, document title) are received by the bot account and retained only as long as needed to process the request.
4. How InPolicy Understands Context
Accurate policy enforcement requires more than checking a message against a rule. Many violations — an unsupported product claim, a reference to a competitor that violates policy, a statement that conflicts with a regulatory requirement — are only detectable if the system understands the facts necessary to apply the relevant policies. InPolicy maintains a Tenant Knowledge Base for each customer organization to provide this grounding.
Tenant Knowledge Base
The Tenant Knowledge Base is a structured set of facts assembled specifically to fill the gaps needed for effective policy enforcement. It contains only information that is directly relevant to applying your organization's policies — for example, who your competitors are, what your products claim to do, and what regulatory standards apply to your business.
This knowledge base is assembled and maintained by InPolicy, primarily from publicly available sources such as product documentation, public filings, and regulatory registers. To find those sources when your organization is set up, InPolicy runs web searches through Google Search grounding on Vertex AI. The searches carry only your organization’s name and website domain, an industry label taken from your public website, and short descriptions of the business facts being looked for (for example, “list of direct competitors”); they contain no user content and no policy text. Google keeps logs of these searches for up to three days for debugging, so they are an exception to the zero-data-retention and United States data-residency commitments described in Sections 5 and 6. The knowledge base is specific to your organization, is never shared with or accessible to other InPolicy customers, and contains no employee communications or personal data.
Your organization can review and correct the Tenant Knowledge Base at any time by contacting security@inpolicy.ai.
Why this matters: A policy prohibiting unsupported product claims cannot be enforced without knowing what your products actually claim. A policy prohibiting competitor disparagement cannot be enforced without knowing who your competitors are. The Tenant Knowledge Base exists solely to close these gaps — it contains only what is necessary to make enforcement decisions, nothing more.
5. Subprocessors and Third-Party Data Handling
| Subprocessor | Purpose | Data Shared | Location |
|---|---|---|---|
| Google Cloud Platform (GCP) | Infrastructure, database, storage and hosting for the backend and AI services | Customer Data, Usage Data | United States |
| Google Cloud (Vertex AI) | LLM inference | Transient Content only (not stored) | United States |
| TypeSafe AI, Inc. | AI inference (decision-model scoring of candidate policy matches during violation detection) | Transient Content (the text being checked, its message header: sender, recipients and subject, and its conversation context: earlier messages in the email thread or chat conversation, with their authors' names) and the Customer Data needed to judge it (the applicable policy rules with their examples and exceptions, and relevant Tenant Knowledge Base facts); zero data retention, not stored for training | United States |
| Vercel | Web application hosting | Transient request data for routing and authentication; no Customer Data stored | United States (application functions); global edge network for request routing and static assets |
| Twilio SendGrid | Transactional email (invitations, notifications) | Recipient email addresses and message content | United States |
| Backblaze, Inc. (B2 Cloud Storage) | Encrypted off-site backups of the production database | Customer Data, encrypted by InPolicy before upload (AES-256); Backblaze does not hold the encryption key | United States |
| Sentry (Functional Software, Inc.) | Application error monitoring | Error diagnostics: stack traces, error messages and request metadata (method, URL, status). Configured not to capture request bodies or Content. | United States |
| Google Workspace | Policy Bot for Google Docs (dedicated bot account) | Documents shared with the Policy Bot account, read transiently; sharing metadata (sharer name and email, document title) | United States |
| Microsoft 365 | Policy Bot for Microsoft 365 (dedicated bot account) | Documents shared with the Policy Bot account, read transiently; the bot removes its own access after processing; sharing metadata (sharer name and email, document title) | United States |
InPolicy does not share customer data with any third parties other than the subprocessors listed above. Our application observability provider (Pydantic Logfire) receives operational telemetry only; customer document content and AI prompt or response content are never sent to it. InPolicy has executed a Cloud Data Processing Agreement with Google Cloud covering both infrastructure (GCP) and LLM inference (Vertex AI). InPolicy will provide at least 30 days' notice before adding or replacing any subprocessor. The Google Cloud Data Processing Agreement includes contractual commitments that Google will not use customer data for advertising, will not train its AI models on customer data, and will process data only in accordance with InPolicy’s instructions. Google’s Vertex AI service operates under a zero-data-retention policy for inference requests, meaning that prompts and responses are not stored or logged by Google after processing is complete. The one exception is the web searches used to build the Tenant Knowledge Base, which Google logs for up to three days (see Section 4). TypeSafe AI processes content under its Master Customer Agreement, which prohibits including customer data in any dataset used to train a model without InPolicy’s consent (consent InPolicy will not give), with zero data retention enabled for InPolicy’s account, meaning TypeSafe keeps no copies of inputs or outputs beyond what is needed to return the response and meet its legal obligations. TypeSafe’s published subprocessor list places its inference infrastructure in the United States, and its trust center shows a SOC 2 Type II attestation. TypeSafe AI was added in October 2026. InPolicy conducts periodic reviews of its subprocessors’ data protection practices and security certifications. Any new subprocessor will be subject to the same level of due diligence and contractual data protection requirements before engagement.
6. Data Residency
All data processed and stored by InPolicy is hosted in the United States, on Google Cloud Platform infrastructure and the subprocessors listed in Section 5. We do not transfer customer data outside the United States.
7. No Training on Customer Data
InPolicy does not use customer data — including policy documents or user content — to train, fine-tune, or improve any AI model. InPolicy’s AI inference subprocessors are contractually prohibited from using customer data to train their models. To be explicit: InPolicy does not use customer data for fine-tuning, reinforcement learning from human feedback (RLHF), embeddings generation, transfer learning, or any other form of machine learning model development. This prohibition applies to user content regardless of whether it has been anonymized or aggregated. InPolicy’s subprocessor agreements contractually prohibit Google from using any customer data to train or improve Google’s own AI models, and prohibit TypeSafe from including any customer data in a dataset used to train TypeSafe’s models without InPolicy’s consent, which InPolicy will not give. InPolicy may use aggregated, anonymized usage data (which does not include any customer content or personally identifiable information) to improve the Service’s general performance, as described in the Terms of Service.
8. Access Controls
- Customer data is accessible only to a small engineering team on a need-to-access basis.
- Access to customer policy documents requires an explicit, logged support request from the customer.
- All internal access to production systems is authenticated and logged.
InPolicy is preparing for a SOC 2 Type II examination. A security posture report is available upon request. In addition to the measures described above, InPolicy maintains the following access controls: role-based access with the principle of least privilege; multi-factor authentication for all employees with access to production systems; automated monitoring and alerting for anomalous access patterns; regular access reviews to ensure that permissions remain appropriate; and documented onboarding and offboarding procedures that include revocation of access within one business day of a person's departure.
9. No Guarantee of Compliance or Complete Detection
InPolicy does not guarantee the detection of all policy violations. The Service may not identify every instance of non-compliant content, and the absence of a violation flag does not constitute a representation that any communication, document, or other content reviewed by the Service is compliant with any applicable law, regulation, internal policy, or other requirement.
Use of the Service does not transfer compliance responsibility to InPolicy. Customers remain solely responsible for their own compliance programs and for ensuring that their communications and activities comply with all applicable laws and regulations. InPolicy is a detection assistance tool; it is not a compliance officer, legal advisor, or guarantor of regulatory compliance. InPolicy strongly recommends that all Output be reviewed by a qualified compliance or legal professional before any action is taken. The Service is designed to assist human decision-makers, not to replace them. InPolicy’s AI technology is inherently probabilistic and may produce false positives (flagging compliant content as a violation) or false negatives (failing to flag non-compliant content). Neither outcome creates liability for InPolicy, as set forth in the Terms of Service.
10. Data Retention and Deletion
| Data Type | Retention |
|---|---|
| Account and organization data | Retained for the duration of the customer relationship; deleted within 30 days of account closure upon request |
| Policy documents | Retained until deleted by the customer or upon account closure |
| User content (emails, documents, chat messages and their conversation context) | Never stored — not applicable |
| Policy state summaries | Retained for the duration of an active enforcement session; maximum 30 days |
| Violation metadata (anonymous, non-user-linked) | Retained for the duration of the customer relationship; deleted within 30 days of account closure or upon request |
| Tenant Knowledge Base | Retained for the duration of the customer relationship; deleted within 30 days of account closure or upon request |
11. Customer Rights
Customers may request at any time:
- A copy of their stored data
- Deletion of their policy documents and account data
- Review or correction of their Tenant Knowledge Base
- A copy of our Data Processing Agreement (DPA) at https://inpolicy.ai/legal/dpa
- Our SOC 2 report once issued, or our current security posture report
Requests can be directed to: security@inpolicy.ai InPolicy will acknowledge receipt of all customer data requests within five (5) business days and will use commercially reasonable efforts to fulfill requests within thirty (30) days. Data exports will be provided in a commonly used, machine-readable format (such as JSON or CSV). Deletion requests will be confirmed in writing upon completion. Customers may also exercise their rights under applicable privacy laws (including the CCPA, GDPR, and other applicable data protection laws) by contacting privacy@inpolicy.ai.
12. Incident Response
InPolicy maintains a documented incident response plan that covers the identification, containment, eradication, and recovery from security incidents. In the event of a confirmed security breach affecting customer data, InPolicy will notify affected customers within seventy-two (72) hours, as required by applicable law and our Data Processing Agreement. Notifications will include: a description of the nature of the breach; the categories and approximate number of data records affected; the likely consequences of the breach; and the measures taken or proposed to address the breach and mitigate its effects. InPolicy conducts periodic tabletop exercises to test its incident response procedures and incorporates lessons learned into its security program.
13. Prohibited Data Types
The Service is not designed to process, and customers should not submit, the following types of data: Protected Health Information (PHI) as defined under HIPAA; payment card data subject to PCI DSS; data classified under ITAR or EAR; government classified or controlled unclassified information (CUI); personal data of children under eighteen (18); or any data whose processing through a cloud-based AI service would violate applicable law or regulation. InPolicy does not offer Business Associate Agreements and is not HIPAA-compliant. Submitting prohibited data types constitutes a material breach of the Terms of Service.
14. Geographic Restrictions
The Service is currently not available to organizations or users based in the United Kingdom. All data is stored and processed in the United States. InPolicy does not transfer customer data outside the United States. For customers subject to the GDPR who transfer personal data from the European Economic Area, InPolicy offers Standard Contractual Clauses (Module 2, Controller to Processor) as part of its Data Processing Agreement.
This document reflects InPolicy AI's data handling practices as of the beta program launch. It will be updated as our infrastructure and certification status evolves. For questions, contact us at security@inpolicy.ai.