Privacy policy
Last updated: 19 November 2026
Translation provided for information only. Only the French version of this document is legally binding.
This policy explains what personal data Aurva Software processes for AIForeGuard, why, on what legal basis, for how long, with which providers, and how to exercise your rights. It covers the website www.aiforeguard.com (presentation and demo requests) and the application app.aiforeguard.com, which is reserved for business customers.
- Controller and contact
- Our two roles: controller and processor
- Processing, legal bases and retention
- Artificial intelligence
- Google user data (Google Workspace directory)
- Recipients and subprocessors
- Transfers outside the European Union
- Security
- Your rights
- Cookies and local storage
- Changes
1. Controller and contact
The controller is AURVA SOFTWARE, a simplified joint-stock company with a single shareholder (SASU) with share capital of €100, whose registered office is at 138 avenue de Versailles, 75016 Paris, France, registered with the Paris Trade and Companies Register under number 130 051 519, publisher of AIForeGuard (hereinafter the “Publisher”, “we”).
In this policy, the “Service” means the website www.aiforeguard.com and the application app.aiforeguard.com; the “Customer” means the business that has subscribed to the Service; a “User” is a person authorised by the Customer to access the application; “Customer Data” means the data the Customer has processed in the application.
- Contact for any question about your data: privacy@aiforeguard.com.
- Data protection officer: no DPO has been appointed to date; any question relating to data protection may be sent to privacy@aiforeguard.com.
2. Our two roles: controller and processor
Controller. The Publisher is the controller of the processing relating to the website (demo requests, contacts), the management of customer accounts, the commercial relationship and invoicing, and the security of the Service (technical logs, error monitoring). This processing is described in §3.1 to 3.5.
Processor. The Publisher acts as a processor, within the meaning of Article 28 GDPR, for all Customer Data: the inventory of its AI agents, metadata and logs read from its services (Microsoft, Amazon Web Services, Google, Mistral, etc.), data about its employees collected by the browser extension, directories, alerts and the audit log (see §3.6). For that data, the controller is the Customer; the Publisher processes it solely on the Customer’s documented instructions, under the terms of the data processing agreement (DPA).
If you are an employee or staff member of a Customer (for example, if your work browser has the AIForeGuard extension installed or if you create AI agents), please contact your employer first: it decides on this processing and is responsible for its compliance. We will forward to it any request we receive directly.
3. Processing, legal bases and retention periods
3.1 Demo requests (website)
- Data: work email address, company name and, if you wish, a message. The form asks for no name, no telephone number and no payment data.
- How it works: your request is sent to us by email, through our provider Resend, to demo@aiforeguard.com; the website itself stores nothing. A hidden field, invisible to people, is used to filter out automated submissions: a request in which it is filled in is rejected. This field collects no additional data.
- Purpose: to answer your request and arrange a demonstration.
- Legal basis: pre-contractual steps taken at your request (Article 6(1)(b) GDPR) or legitimate interest in responding to professional enquiries (Article 6(1)(f)).
- Retention: 3 years after the last contact, unless a contract is concluded (see §3.3).
3.2 Contacts and pre-contractual exchanges
- Data: name, email address, organisation, job title and the content of the exchanges you send us (in particular to contact@aiforeguard.com), including proposals and quotations prepared at your request.
- Purposes: to answer your questions and to prepare a demonstration or an offer.
- Legal basis: pre-contractual steps taken at your request (Article 6(1)(b)) or legitimate interest in responding to professional enquiries (Article 6(1)(f)).
- Retention: 3 years after the last contact, unless a contract is concluded.
3.3 Customer accounts, commercial relationship and invoicing
- Data: the Customer’s company name; name, job title and business contact details of its contacts (signatory, administrators, sales and billing contacts); purchase order or signed quotation, offer, scope and commercial terms; contract status; billing address and VAT number; invoices and payments.
- Payment: no online payment. The Service is invoiced and paid by bank transfer; we collect no card details.
- Legal basis: performance of the contract concluded with the Customer (Article 6(1)(b)), legitimate interest in managing the relationship with its contacts (Article 6(1)(f)) and compliance with our accounting and tax obligations (Article 6(1)(c)).
- Retention: for the duration of the contractual relationship; accounting records (invoices, purchase orders) are then kept for 10 years (Article L123-22 of the French Commercial Code).
3.4 User accounts and access to the application
- Data: identifier provided by the identity provider (issuer and the person’s identifier), email address, display name, role (administrator, security analyst, AI agent manager, auditor or reader), permissions and date of last sign-in; sign-in sessions, of which only a fingerprint (SHA-256) of the token is kept; link with Slack, if the User links their account (Slack identifier and name); mapping between Microsoft Entra groups and roles, if the Customer configures it; invitations, sent as a link.
- Sign-in: Users sign in through their organisation’s identity provider (Microsoft Entra ID, Okta or another OpenID Connect provider); we do not receive their password. For email-and-password accounts created by the Publisher, only a fingerprint of the password (scrypt) is kept, never the password itself.
- Sessions: 8 hours at most; a session ends after 2 hours of inactivity.
- Role and legal basis: the Customer decides who accesses the application and with which rights; for those decisions and their history, the Publisher acts on the Customer’s behalf (see §2). The Publisher also processes this data, as controller, to authenticate Users and secure the Service, on the basis of its legitimate interest in providing the Customer with secure access to the Service (Article 6(1)(f)).
- Retention: for the term of the Customer’s contract; when the Customer removes a User’s access, the history of that User’s actions remains in the audit log. Sessions are deleted 7 days after they expire; invitations, after 7 days.
3.5 Security and technical logs
- Data: technical logs generated by our hosting providers and our servers for each request (IP address, date, request, request identifier, errors); IP addresses of the browsers that download the extension’s rule files; technical error messages sent to our error-monitoring tool, filtered by the application before sending: no IP address, no location, no User identity, no cookies, no headers and no content.
- Purposes: to ensure the security, availability and proper functioning of the Service; to detect and fix errors; to prevent abuse, in particular through rate limits.
- Legal basis: legitimate interest in ensuring the security and proper functioning of the website and the application (Article 6(1)(f)).
- Retention: a limited period applied by each provider (hosting, delivery, error monitoring).
3.6 Data processed on behalf of our Customers
In the application, the Publisher is a processor (see §2). The Customer, as controller, determines the purposes and the legal basis of this processing. In particular, it is responsible for informing its employees in advance and, where the law requires it, consulting the employee representatives (in France, the social and economic committee, or CSE: Articles L1222-4 and L2312-38 of the French Labour Code) before deploying the browser extension or any other monitoring tool, for ensuring that this processing is proportionate, and for carrying out a data protection impact assessment (DPIA) where one is required.
Depending on the connections and options enabled by the Customer, the Service processes:
- Inventory of AI agents (Microsoft Copilot Studio, Azure AI Foundry, AWS Bedrock, AgentCore and Quick, Google Gemini Enterprise, Mistral, etc.): configuration and components of the agents, limited to the fields selected by the Service; agent instructions, kept only as a fingerprint and a length; display names and work email addresses of the people named by an agent (owners, authors); number of conversations per day; Microsoft Entra sign-in logs of agent creators (device, operating system, compliance, application, dates), without IP address or location.
- Approval before publication: request opened for an agent; work email address of its creator, used to send them the link to their request and then the answer, through the integration the Customer chooses (its own Microsoft 365 or Gmail mail system, Slack); the creator’s answers (purpose of the agent, audience, process step, explanations of gaps); the business owner’s opinion and the security decision. The link sent is kept only as a hash.
- Hold of unapproved agents (optional): if the Customer chooses it, an unapproved agent is put in quarantine on its platform, and the quarantine is lifted on approval (only these two actions, logged). Depending on the method chosen: the address of the flow installed in the Customer’s tenant, or the sign-in name of the administrator who linked their account and their delegated token, kept in the secret vault.
- Microsoft 365 Copilot (audit logs): metadata only (application, agent, extensions, sensitivity label identifiers, jailbreak detections, model provider); the user’s identifier, kept as a fingerprint computed with a key specific to the organisation; administrative actions on agents (who, what, when); number of licences.
- Microsoft Defender: device name, Microsoft Entra device identifier, local account name, host or command of the MCP server detected, to identify AI assistants installed on workstations.
- Live control (Copilot Studio): the user’s message, the conversation history and the values passed to tools are read in memory and are not kept; each decision records the agent, the tool, the parameter names, the outcome, the Microsoft Entra user reference and the conversation reference.
- Browser extension (Shadow AI): origin host and destination host among AI services, type of observation, signal codes and daily counters per host (visits, messages, requests, blocks); email address of the browser profile, kept only as a fingerprint (HMAC); push-notification subscription address of the browser, used to wake the extension with an empty message so that it fetches its rules; document labels, read locally on the device; access requests from an employee (name, work email and reason entered by them); employee records (name and optional reference). If the Customer turns on per-person detail (“All usage per person”), the name, size and SHA-256 fingerprint of each file whose upload to an AI service was blocked, with the person and time, as evidence; never the file content, and these items are erased at the end of the chosen period (30, 90 or 180 days) or as soon as the Customer turns the option off. In Netskope mode, the content of a file passes through the AIForeGuard server to the Customer’s Netskope tenant without being kept. As an option, a relay installed on the Customer’s own infrastructure keeps the per-browser detail there (8 days) and sends only totals per unit to the Service.
- Directories (Microsoft Entra ID; SCIM directories such as Okta, OneLogin, JumpCloud, Ping Identity or Microsoft Entra; Google Workspace, see §5): each person’s department and groups, to link equipped browsers to a department; identifiers are kept as fingerprints, with no name or email address in clear text.
- Alerts sent to the Customer’s tools (Microsoft Teams, Slack, Jira, ServiceNow, Microsoft Sentinel, webhook): title, priority, reason, agent name, platform and connection concerned, dates and link to AIForeGuard; on Slack, the name of the person who takes charge of the alert and of the person who resolves it. Never any document content.
- Connection secrets: integration keys, tokens and passwords entrusted by the Customer, including those of alert destinations, kept only in the secrets vault (Azure Key Vault); Jira and Slack tokens are renewed and kept there.
- Audit log: who did what and when in the application; it is append-only and cannot be modified.
Microsoft permissions. To read this information, the Service asks an administrator of the Customer for read permissions: AgentIdentity.Read.All, AuditLog.Read.All, GroupMember.Read.All, User.Read.All, ThreatHunting.Read.All, LicenseAssignment.Read.All and ActivityFeed.Read, as well as delegated access to Dynamics and Azure Resource Manager and the “AI Reader” role. The Customer can withdraw these permissions at any time in Microsoft Entra ID.
What the Service never keeps:
- the text of conversations with AI agents, or the questions and answers of Microsoft 365 Copilot;
- files and document content;
- the values of agents’ environment variables;
- for the browser extension: questions, answers, files, cookies, page titles, or the paths and parameters of the addresses visited;
- the IP address and location of agent creators;
- in clear text: the name and email address of the people listed in directories, the email address of the browser profile and the identifier of Microsoft 365 Copilot users.
The only content examined is read in memory, without being kept: the message, history and tool values read by live control, and the files sent to the Customer’s Netskope tenant in Netskope mode. Customer Data is used solely to provide the Service to the Customer; it is neither sold nor used for advertising.
Retention periods for Customer Data:
| Data | Retention |
|---|---|
| Inventory and current state of agents, settings, and what is still in force: open cases, last decided review of each agent, decisions and exceptions not yet expired, refused publication requests, notices sent, mission contract in force, with the evidence they cite; reports produced | Term of the contract, unless a shorter period is agreed with the Customer |
| Names of the people named by agents | 90 days after the last reading |
| Live control decisions | 90 days |
| Audit log, agent events, notifications, browser extension events, activity counters, AI usage, administration events, collection history, observed agent states, evidence items and detected changes that no dossier cites any more, closed cases (reviews, requests, resolved alerts, plans), superseded versions and declarations | 1, 2 or 3 years, as chosen by the Customer in the Service (3 years by default) |
| Exports (files) and invitations | 7 days |
| Sessions | Deleted 7 days after they expire |
| Per-browser detail in the relay installed on the Customer’s infrastructure (option) | 8 days, on the Customer’s side |
| Daily database backups | 7 days (rotation) |
End of contract. Before the contract ends, the Customer can export its audit log, review dossiers and reports (PDF). No later than thirty (30) days after the end of the contract, a period during which it helps the Customer, on request, to obtain any exports it has not made, the Publisher completely erases the Customer’s organisation; if the erasure is interrupted, it resumes where it stopped, and a receipt, containing no person’s name, records it. Versions of secrets kept in the vault are permanently purged 90 days after their deletion (protection against accidental deletion); backups disappear through rotation within 7 days.
4. Artificial intelligence
AIForeGuard helps its Customers govern their use of AI, but does not itself use any artificial-intelligence model or any large language model (LLM) provider to process data.
- Nothing sent to an AI service. No data, personal or otherwise, is sent to an AI model to be analysed, summarised or drafted.
- Reading the Customer’s AI platforms. The connections to Microsoft Copilot Studio, Microsoft 365 Copilot, Azure AI Foundry, AWS Bedrock, AgentCore and Quick, Google Gemini Enterprise or Mistral are used only to read their configurations and metadata. Copilot Studio live control, when the Customer enables it, examines agents’ tool calls in memory and allows or blocks them according to the Customer’s settings, without keeping their content (see §3.6).
- No training. Customer Data is not used to train any AI model.
- No decisions about individuals. Alerts and blocks apply explicit rules chosen by the Customer; any follow-up, in particular regarding an employee, is for the Customer to decide. The Service takes no decision producing legal effects concerning a person based solely on automated processing (Article 22 GDPR).
5. Google user data (Google Workspace directory, Gmail)
This section applies when a Google Workspace administrator of a Customer authorises AIForeGuard to read their organisation’s directory, an optional feature linked to the browser extension, or when a Customer connects a Gmail mailbox dedicated to sending its emails (§5.4). It describes precisely the data obtained through Google APIs (“Google user data”) and supplements the other sections of this policy.
Limited Use commitment. AIForeGuard’s use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. The use of raw or derived user data received from Google Workspace APIs will adhere to the Google User Data Policy, including the Limited Use requirements.
5.1 Data accessed
- Permissions, read-only:
admin.directory.user.readonly,admin.directory.orgunit.readonly,admin.directory.group.readonlyandopenid email. - Directory: for each person, their department, organisational unit and groups, read again every hour.
- Google Workspace account: the identifier of the Google Workspace account and its domain name, to link that account to a single organisation in AIForeGuard.
- What is kept: only fingerprints of the accounts and their link to a department or group; never a person’s name or email address in clear text.
- Read-only: for the directory, AIForeGuard never creates, modifies or deletes anything in your Google Workspace. Apart from sending emails as described in §5.4, it does not access any other Google service (Gmail, Calendar, Drive, etc.).
5.2 How the data is used
This data is used exclusively for one feature that is visible in the application and chosen by the Customer: linking each browser equipped with the extension to the department or group of the person using it, in order to show AI usage by department. No other use is made of it: no advertising, targeted or otherwise, no resale, no solvency assessment and no lending decisions.
5.3 Sharing and transfer
Your Google user data is never sold or rented, and is never transferred to third parties, in particular to advertisers, data brokers or information resellers. Within AIForeGuard, it is visible only to the Customer’s Users, according to their permissions. Only our hosting providers process it on our behalf, strictly for this feature: Vercel (server functions, Paris), Supabase (database, Paris) and Microsoft Azure Key Vault (refresh token, France Central). It may also be disclosed to comply with applicable law or to investigate a security incident or abuse and, as part of a merger, acquisition or sale of the Publisher’s assets, only with the Customer’s explicit prior consent.
Artificial intelligence. AIForeGuard does not use your Google user data — raw, derived, aggregated or anonymised — to develop, improve or train artificial intelligence or machine learning models, and does not transfer it to any AI service.
Human access. No member of the Publisher’s staff reads your Google user data, except: at the Customer’s express request, for specific data (for example during a support request); when necessary for security purposes, such as investigating a bug or abuse; or to comply with the law.
5.4 Protection, retention and deletion
- Protection: all exchanges are encrypted (TLS); the refresh token issued by Google is kept only in the secrets vault (Azure Key Vault), the database keeping only its reference; each organisation’s data is isolated from that of other organisations.
- Retention: for as long as the Google Workspace directory is linked to AIForeGuard.
- Revocation: when the Customer unlinks Google Workspace, the token is disabled in the vault and revoked with Google. A Google administrator can also remove access at any time from their Google account (myaccount.google.com/permissions) or from the Google Workspace Admin console: the directory is then no longer read.
- End of contract: erasing the organisation erases its tokens and its Google user data; versions of the token in the vault are permanently purged 90 days after their deletion, and database backups disappear within 7 days.
For any request concerning your Google user data: privacy@aiforeguard.com.
5.4 Sending emails through Gmail (optional)
- Permission:
gmail.send(send emails from the connected mailbox) andopenid email(that mailbox’s address). The Customer connects a mailbox dedicated to sending, for example ai-governance@their-domain. - Use: sending from this mailbox the alerts chosen by the Customer and the approval-before-publication messages (request link to an agent’s creator, then the answer). Nothing else is sent.
- No reading: AIForeGuard never reads, files or deletes any email and keeps no copy of the messages sent.
- Token: the refresh token is kept in the secret vault; disconnecting Gmail in AIForeGuard disables it, and the Customer can remove the access at any time in their Google account.
6. Recipients and subprocessors
Data is accessible only to authorised Publisher staff who need it, and to our subprocessors within the limits of their assignment:
| Provider | Assignment | Location |
|---|---|---|
| Vercel Inc. (United States) | Hosting of the website and the application, server functions (including scheduled tasks), transmission of demo requests | Functions: Paris, France; static files: global delivery network (cache) |
| Supabase, Inc. (United States) | Dedicated database and backups (7 days) | European Union — Paris, France |
| Microsoft (Azure Key Vault) | Secrets vault (integration keys, tokens and passwords entrusted by Customers) | France Central |
| Cloudflare, Inc. (United States) | Delivery of the extension’s rule files, signed and encrypted (unreadable by Cloudflare); DNS for the domain | Global network |
| Functional Software, Inc. (Sentry) (United States) | Technical error monitoring, without personal data | European Union — Frankfurt, Germany |
| Google (Google Workspace) | Business email for the aiforeguard.com domain | Google’s global infrastructure |
| Resend | Sending of the website’s demo requests to demo@aiforeguard.com (nothing is stored by the website) | Company established in the United States |
The detailed list, with the data concerned and each provider’s safeguards, is published on the Subprocessors page.
Services connected at the Customer’s initiative. The services the Customer chooses to connect are enabled at its initiative and are governed by the contract between it and their providers: Microsoft (Entra ID, Microsoft 365 and Purview, Copilot Studio and Dataverse, Power Platform, Defender, Sentinel, Teams, Azure AI Foundry), Amazon Web Services (Bedrock, AgentCore, Quick), Google (Gemini Enterprise, Google Workspace directory), Mistral, SCIM directories (Okta, OneLogin, JumpCloud, Ping Identity, Microsoft Entra), the Customer’s identity provider (Okta, OpenID Connect), Slack, Atlassian Jira Cloud and Jira Data Center, ServiceNow, Netskope and the Customer’s webhook address (signed deliveries). What the Service reads from them or sends to them is described in §3.6. The extension is also woken through the notification service built into the Customer’s browser (for example Google for Chrome, Microsoft for Edge), which only receives an empty message.
We may disclose data to the authorities where the law requires it. We neither sell nor rent any personal data.
7. Transfers outside the European Union
The data stored by the application is hosted in the European Union: database in France (Supabase, Paris), server functions in France (Vercel, Paris region), secrets in France (Azure Key Vault, France Central) and error monitoring in Germany (Sentry, Frankfurt).
Some data is nevertheless processed outside the European Union, or in a global infrastructure with no guaranteed region: requests served by the global delivery networks of Vercel (static files) and Cloudflare (the extension’s rule files, signed and encrypted), which see browsers’ IP addresses; business email (Google Workspace); sending of demo requests (Resend); browsers’ subscription addresses, passed to the notification service built into each browser. Providers established in the United States that host data in the European Union may also access it from abroad for support and operational purposes.
These transfers and this access are governed by each provider’s data processing agreement, by the standard contractual clauses adopted by the European Commission for any access from a third country and, where the provider participates in it, by the EU–US Data Privacy Framework. They are supplemented by technical measures: encryption of communications, signed and encrypted rule files, and error monitoring without personal data. You can obtain information about these safeguards by writing to privacy@aiforeguard.com.
8. Security
We implement measures appropriate to the risks, including:
- isolation of organisations through row-level security (RLS) enforced in the database, separate database roles with no way to bypass it, and verification of this isolation after every schema change;
- secrets kept only in the vault (Azure Key Vault), which the application accesses through identity federation, with no stored secret;
- encryption of communications (TLS);
- append-only audit log;
- dual approval (“four eyes”) for mission contracts, lifecycle changes and regulatory profiles;
- roles and permissions per member; short sessions (8 hours at most, 2 hours of inactivity), of which only a fingerprint of the token is kept;
- protection against cross-site request forgery (CSRF), rate limits and a strict content security policy;
- browser rules that are signed (Ed25519) and encrypted (AES-GCM);
- allow-lists of hosts for outgoing calls;
- error monitoring without personal data;
- daily database backups, kept for 7 days.
No measure offers absolute security, however. We will notify any data breach under the conditions provided for by the GDPR: to the CNIL and, where applicable, to the data subjects for processing for which we are the controller; to the Customer, without undue delay, for data we process on its behalf.
9. Your rights
You have the rights of access, rectification, erasure, restriction, objection and portability, as well as the right to give directions concerning the fate of your data after your death (Article 85 of French law no. 78-17 of 6 January 1978 on information technology, data files and civil liberties). You may object at any time, on grounds relating to your particular situation, to processing based on our legitimate interest.
- How to exercise them: write to privacy@aiforeguard.com. We reply within one month, extendable by two months in the cases provided for by the GDPR, and may ask you to prove your identity in the event of reasonable doubt.
- Data processed on behalf of a Customer (for example, data collected by the browser extension on your work computer): please contact your employer, as controller, first; we will forward to it any request we receive and help it to respond (see §2).
You may lodge a complaint with the CNIL, 3 place de Fontenoy, TSA 80715, 75334 Paris Cedex 07, France (www.cnil.fr).
10. Cookies and local storage
Website www.aiforeguard.com: no cookie is set and nothing is stored in your browser’s local storage. The website loads no third-party resources (its fonts are served by the website itself) and uses no analytics, advertising or tracking tool.
Application app.aiforeguard.com: only strictly necessary cookies and preference cookies are used.
| Cookie | Purpose | Duration |
|---|---|---|
__Secure-next-auth.session-token | Sign-in session, encrypted (session identifier only) — strictly necessary | 8 hours at most; ends after 2 hours of inactivity |
__Host-next-auth.csrf-token | Protection against cross-site request forgery (CSRF) — strictly necessary | Until the browser is closed |
__Secure-next-auth.callback-url | Page to reopen after sign-in — strictly necessary | Until the browser is closed |
__Secure-next-auth.state, __Secure-next-auth.pkce.code_verifier, __Secure-next-auth.nonce | Sign-in checks with the identity provider (OAuth, OpenID Connect, PKCE) — strictly necessary | 15 minutes (nonce: until the browser is closed) |
aib_ms_proof, aib_ms_grant, aib_ms_discovery | State of connections to Microsoft (tenant verification, authorisation, discovery of environments) — strictly necessary | 10 minutes |
aifg_slack_oauth, aifg_jira_oauth, aifg_google_directory | State of connections to Slack, Jira Cloud and Google Workspace — strictly necessary | 10 minutes |
aib_locale | Interface language — preference | 1 year |
aib_tz | Browser time zone, to show times in local time — preference | 1 year |
aib_theme | Display theme — preference | 1 year |
aib_bu | Unit chosen as a filter — preference | 1 year |
aiforeguard_last_org | Last organisation opened — preference | 1 year |
| Exploration view cookies (one per organisation) | View and filter chosen on the exploration page — preference | 1 year |
The application also uses the browser’s local storage for display preferences (for example text size or the filters shown) and session storage to link a Slack account, until the tab is closed.
No analytics or advertising cookie is used. These cookies and this storage serve only to operate the application requested by the User or to remember their interface preferences: they are exempt from consent (Article 82 of the French Data Protection Act). This is why no consent banner is shown.
11. Changes
We may update this policy, in particular in the event of new processing or a change of provider. The version in force is published on this page with its update date; in the event of a significant change, Customers are informed by email. Changes of subprocessors are announced to Customers under the conditions set out in the data processing agreement.