# Privacy Policy How PrivateMind handles, protects and governs the data you bring to private AI. **Version 1.7**, effective 2026-10-29. ## 1. Introduction and Purpose This AI Privacy Policy explains how Options Technology Limited ("Options") handles personal data in connection with PrivateMind. It works together with the Client's contract, the AI-specific Data Processing Agreement for PrivateMind (the "DPA"), the [Terms of Service](https://www.privatemind.com/terms/), the [Fair Use Policy](https://docs.privatemind.com/fair-use-policy.html) and the [Cookie Notice](https://www.privatemind.com/cookie-notice/). If this policy conflicts with the DPA on data protection, the DPA prevails. ## 2. Privacy at a Glance - Your prompts, files and outputs (Protected AI Data) stay inside your organisation's isolated PrivateMind tenant, and your organisation, not Options, is the controller of that data. Options processes it only on your instructions. - Protected AI Data is never used to train, fine-tune or improve any AI model, ours or anyone else's, except with your prior written consent. - Options does not read your prompts, except a flagged request examined in a security review (Section 8). Security monitoring uses system metadata and automated safety screening. - Protected AI Data is hosted in your agreed region; see our [hosting locations](https://www.privatemind.com/products/infrastructure/#locations) for the regions available today. - Options is the controller only of the limited Service Data needed to run your account (contacts, billing, authentication records, support tickets and website analytics). ## 3. Scope of this Policy This policy covers the standard PrivateMind service: private AI hosting on Options-controlled infrastructure using open-source (open-weight) models, delivered via webchat and API to external Clients, including EEA organisations. It also covers Options' own internal users of the same standard service, on the same terms. Website cookies and analytics are covered in the separate [Cookie Notice](https://www.privatemind.com/cookie-notice/). **Out of scope, assessed case by case:** any bespoke arrangement or specific Client requirement beyond the standard service (for example access to non-open-source or frontier models, dedicated or unusual configurations, new jurisdictions) is assessed individually and contractually gated, with supplemental privacy terms and (where needed) an updated impact assessment. Anything not covered by this policy is handled in the Client's contract or DPA, not assumed into the standard service. ## 4. Data Protection Roles Under the standard service: - For Service Data, Options is the Controller and determines purposes and means. - For Protected AI Data, the Client is the Controller and Options is the Processor, processing only on the Client's documented instructions under the contract and DPA. - The Client remains responsible for having a lawful basis for any personal data submitted to PrivateMind, and for giving End-Users the privacy information they need. ## 5. What Data Options Processes PrivateMind is an AI system: users are informed at the point of interaction that they are interacting with AI, and AI-generated content is marked as such in a machine-readable way (EU AI Act, Art. 50). Clients act as deployers of AI systems under their own authority. **Service Data (Options as Controller):** account-holder and administrator contacts, organisation details, login/authentication records (MFA, SSO/SAML), API-key metadata (keys are stored only in hashed form, never in clear text), billing and usage records, support tickets, and website/analytics data (see the [Cookie Notice](https://www.privatemind.com/cookie-notice/)). **Protected AI Data (Options as Processor):** - prompts/queries submitted and model responses returned; - files/uploads and connected sources (documents, connectors, retrieval-augmented content); - conversation history, memory and session data retained per the Client's settings and contract; - metadata (timestamps, model used, token counts, user/tenant identifiers); - embeddings/vector-store content where retrieval features are used (treated as personal data); - operational, security and abuse-detection logs (see Logging and Monitoring, Section 8). **Infrastructure data (no Protected AI Data):** service improvement uses only performance and capacity data about the infrastructure and its components (system load, resource allocation, throughput, latency). Prompts, outputs and other Protected AI Data are never used for service improvement. **No training on Protected AI Data:** prompts and outputs are never used to train, fine-tune or otherwise improve any model (Options' or any third party's), except with the Client's prior written consent. ## 6. Purposes and Legal Bases As Controller of Service Data, Options uses it to: - provide and bill for the service, and manage the account and API keys, on the basis of contract; - secure, monitor and operate the platform, and detect abuse and fraud, on the basis of legitimate interest (documented assessment); - improve the service using infrastructure performance and capacity data, on the basis of legitimate interest, subject to the Client opt-out (see Service-Improvement Opt-Out, Section 9); - comply with law (tax, accounting, sanctions and export control), on the basis of legal obligation; - send service and product messages, on the basis of legitimate interest or consent, with easy opt-out. As Processor of Protected AI Data, Options processes it only per the Client's contract and instructions. The legal basis for Protected AI Data is the Client's own. ## 7. Tenant Isolation and Data Location **Demo Service.** The Demo Service is a free shared evaluation environment in which content is stored in a shared database. Do not submit to the Demo Service: (a) production data; (b) personal data relating to any individual, other than your own account registration information; or (c) material non-public information. Paid Plans offer isolated or dedicated environments without these constraints. **Paid Plans.** Each Client's data is held in its own logically isolated tenant and database; no Client has access to another Client's data, sessions or logs. Protected AI Data is hosted in the region agreed in the Client's contract. All available hosting locations, and the data-residency information for each, are listed on our [hosting locations](https://www.privatemind.com/products/infrastructure/#locations) page. PrivateMind is engineered so Protected AI Data stays within the agreed environment. ## 8. Logging and Monitoring Usage of PrivateMind (webchat and API) is logged and monitored for security, abuse detection and service integrity, including usage patterns and rate-limit adherence, policy-violation indicators, system load and resource allocation, and suspicious or anomalous behaviour. Prompt-content monitoring is automated. Where safety screening is enabled for the Client's organisation, each user message is also checked by a safety model that Options runs on its own infrastructure. The check keeps a verdict, its categories and a one-way hash of the message, not the message itself. If an API request is flagged, its content is stored encrypted for 90 days and can be decrypted only through an audited security review. Prompts are not otherwise read; prompt content is processed by the model only to generate the response, within the Client's tenant. | Log type | Purpose | Content | Retention | | --- | --- | --- | --- | | Access/security logs | Security, audit, login records | User/tenant IDs, IP addresses, user agents, timestamps, auth events | Up to 7 years | | Usage/rate-limit logs | Billing, usage reporting, fair-use and capacity enforcement | Token counts, model used, cost, user and key IDs, timestamps | 7 years | | Safety-screening verdicts | Abuse detection and security review | Verdict, categories, a hash of the message, user and tenant IDs, timestamp; not prompt content | 400 days | | Flagged API requests | Security review | The flagged request, encrypted | 90 days | | Protected AI Data logs | Not generally created. If the Client's contract enables content logging, the scope and retention are stated in the DPA. | Per contract | Per the Client's DPA | ## 9. Service-Improvement Opt-Out The Client may opt out of the use of infrastructure performance and capacity data for service improvement via the contract or by emailing the DPO ([DataPrivacy@options-it.com](mailto:DataPrivacy@options-it.com)); an in-product opt-out control is planned and will be added to this policy when available. Opting out does not affect security monitoring under Logging and Monitoring (Section 8), which is required to protect the service. ## 10. Security Options maintains appropriate technical and organisational measures under its ISO/IEC 27001-certified ISMS and currently inflight ISO/IEC 42001 AI management system certification, including: - Encryption in transit (TLS) and at rest (AES-256-GCM); - Tenant isolation; restricted egress; sandboxed workloads; - Identity and access management; MFA required for webchat - where the Client connects via the Client's own identity provider (SSO/SAML), enforcing MFA is the Client's responsibility; - API-key security requirements: rotation and safekeeping are the Client's responsibility, and permission isolation so connected content stays within the requesting End-User's access; - Protections and monitoring against prompt injection and jailbreak (unauthorised security testing by Clients is prohibited); - security monitoring and SIEM integration across the infrastructure; - upstream model provenance, integrity and licence review before any model is deployed. ## 11. Sub-Processors Options uses the following categories of sub-processors to deliver PrivateMind: - Options group entities: infrastructure and operations affiliates delivering the service, including follow-the-sun support teams (Belfast/UK, New York/US, Toronto/Canada); - Microsoft: Microsoft 365 and email communications; - Payment providers: for credit card transactions where relevant. The current list of named sub-processors, the underlying model and infrastructure providers and data-residency information is available on request. Options gives Clients at least 30 days' notice before authorising a new sub-processor, and Clients may object on reasonable data-protection grounds. Sub-processors are bound by flow-down terms consistent with the DPA, and Options remains liable for their performance. No sub-processor will ever have access to Protected AI Data. ## 12. International Data Transfers Where personal data is transferred outside the UK/EEA, Options relies on appropriate safeguards: an adequacy decision, the UK International Data Transfer Agreement (IDTA) or UK Addendum, or EU Standard Contractual Clauses, with supplementary measures and, where required, a Transfer Impact Assessment, as set out in the DPA. Cross-border support access is role-restricted and logged. ## 13. Retention and Deletion Protected AI Data and logs are retained per the contract or DPA and the retention periods in Logging and Monitoring (Section 8). On termination or request, Options deletes or returns Protected AI Data (including from vector stores and embeddings) per the DPA, subject to legal holds. As Controller, Options retains Service Data only for legal and accounting periods and no longer. ## 14. Data portability and switching (EU Data Act) Clients may self-service export Protected AI Data and their audit records in structured, commonly used, machine-readable formats at any time, free of charge. Following termination, data remains retrievable for at least 30 days, after which it is deleted (including from backups). We do not charge egress or switching fees. ## 15. Your Rights, UK and EEA Both Clients and End-Users have the rights of access, rectification, erasure, restriction, objection (including to direct marketing), portability, withdrawal of consent, and protection from solely automated decisions with legal or similarly significant effect. **Protected AI Data:** the Client is responsible for requests relating to Protected AI Data; Options assists under the DPA. End-Users should contact their own organisation first. **Service Data:** requests may be sent directly to the DPO at [DataPrivacy@options-it.com](mailto:DataPrivacy@options-it.com). Options responds to valid requests within one month (extendable by two months for complex or numerous requests). Data subjects may complain to the UK ICO ([ico.org.uk](https://ico.org.uk/)) or the relevant EU supervisory authority. ## 16. Your Rights, US Residents (CCPA/CPRA) and Canada To the extent the California Consumer Privacy Act, as amended by the CPRA, applies to Options' processing of Service Data, and to comparable US state privacy laws: - Options collects only the categories of Service Data described in What Data Options Processes (Section 5), for the purposes described in Purposes and Legal Bases (Section 6). - Options does not sell or share personal data as those terms are defined by the CCPA/CPRA. - California residents have the rights to know, delete, correct and port their personal data, and the right not to receive discriminatory treatment for exercising those rights; requests may be sent to [DataPrivacy@options-it.com](mailto:DataPrivacy@options-it.com). Options responds within the statutory timeframes. - This section is a notice only; the Client remains responsible for its own compliance obligations regarding Protected AI Data. Canada: where Canadian federal or provincial privacy law (including PIPEDA) applies, Options complies with it, and individuals may contact the DPO to exercise available rights. ## 17. Children PrivateMind is an enterprise service and is not directed at children. Options does not knowingly collect personal data from anyone under 18, and Protected AI Data must not contain personal data of children where prohibited by law. ## 18. Security Incident and Breach Notification If Options confirms a security incident affecting the service or Protected AI Data, Options will notify the affected Client without undue delay and in any event within 48 hours of confirmation. Where the incident is a personal-data breach, Options will also notify the ICO within 72 hours (and the relevant EU authority for EEA data) where required, and will support the Client with any notification to the Client's own data subjects. The Client must report security incidents affecting the Client's environment (including leaked credentials) to Options without undue delay via [privatemindsupport@options-it.com](mailto:privatemindsupport@options-it.com) (operational incidents) or [DataPrivacy@options-it.com](mailto:DataPrivacy@options-it.com) (privacy or data incidents); either contact will route the report appropriately. ## 19. AI Transparency, Limitations and Human Oversight - End-Users interact with an AI system. This disclosure is built into the product interface (EU AI Act Article 50(1)). - PrivateMind outputs are assistive and advisory, not authoritative: outputs can be inaccurate, incomplete or fabricated ("hallucinated"), may reflect bias, and are subject to a knowledge cut-off. Outputs must be verified before reliance. - PrivateMind must not be used for fully automated decision-making with legal or similarly significant effect (for example credit, employment, insurance) without qualified human review, and outputs are not a substitute for regulated professional advice. - Where required, Clients must disclose AI-assisted generation. ## 20. EU AI Act, Intended-Purpose Boundary PrivateMind is a general-purpose private AI platform for enterprise productivity, coding, analysis and research support on open-source models. PrivateMind is not designed or intended to be a high-risk AI system (Annex III), for example, for employment or recruitment decisions, credit scoring, biometric identification, or access to essential private or public services. **Contractual gate:** any use case that is, or could become, high-risk or beyond the documented intended purpose requires prior written agreement and an updated AI impact assessment and EU AI Act review by Options before use. Absent that agreement, such use is outside the permitted scope of the service. Options' obligations are re-assessed on any change of market, user population or system purpose. The Client remains responsible for compliance with the policies of any LLM available within PrivateMind. ## 21. Regulated Industries and Records The Client is solely responsible for the Client's own regulatory obligations relating to inputs submitted to PrivateMind or reliance on outputs (for example FCA, SEC, MAS), including assessing whether AI-generated content constitutes a business record (for example SEC Rule 17a-4, MiFID II, FCA SYSC) and retaining it accordingly. Unless agreed in an Enterprise arrangement, special-category personal data, credentials and secrets, and payment-card or other regulated financial data must not be submitted to PrivateMind. Access is not permitted from prohibited or sanctioned jurisdictions, or by sanctioned parties. ## 22. Other Jurisdictions and Governing Law This policy and the privacy it describes are governed by UK and EU data protection law (UK GDPR, DPA 2018, EU GDPR), and, for matters not concerning data protection, by the laws of England and Wales, whose courts have exclusive jurisdiction, unless the Client's contract provides otherwise. Where other data-protection laws apply to Options (for example in respect of personnel or contacts in North America), Options complies with them (see also Your Rights, US Residents (CCPA/CPRA) and Canada, Section 16); the Client's responsibilities for the Client's own jurisdictions are set out in the contract. ## 23. Changes to this Policy This policy is reviewed at least annually and kept in step with the Terms of Service, Fair Use Policy and the DPA. Material changes affecting how Options processes personal data are notified in advance with at least 30 days' notice where feasible. Continued use of the services constitutes acceptance. The latest version is published at [www.privatemind.com/privacy](https://www.privatemind.com/privacy) and [docs.privatemind.com/privacy-policy](https://docs.privatemind.com/privacy-policy). ## 24. Key Terms and Definitions - **"Client"**: the organisation contracting with Options for PrivateMind. - **"Controller" / "Processor"**: as defined in UK GDPR / EU GDPR Article 4. Roles for each data category are set out in Data Protection Roles (Section 4). - **"Demo Service"**: a free, shared evaluation environment. - **"DPO"**: Options' Data Protection Officer. - **"End-Users"**: the Client's personnel and other users the Client authorises to use PrivateMind. - **"Options"**: Options Technology Limited, company no. 02872304, 6th Floor, 11 Strand, London, WC2N 5HR (ICO-registered controller, registration number ZA27022). The contracting Options entity for your account is identified at sign-up or in your contract. It will be Options Technology Limited or an Options group Affiliate if otherwise identified. This policy applies to PrivateMind regardless of which Options entity contracts. - **"Paid Plans"**: paid shared or dedicated services, as described at sign-up, in the applicable plan description or in an order form. - **"Protected AI Data"**: data processed inside the Client's tenant: prompts, outputs, files/uploads, connected sources, conversation history and related metadata. The Client is the Controller of Protected AI Data; Options is the Processor. - **"Service Data"**: data relating to the account and commercial relationship: organisation and admin contacts, billing, authentication/MFA records, API-key metadata, support tickets, and website/marketing data. Options is the Controller of Service Data. ## 25. Contact Options Data Protection Officer (privacy questions, data subject requests, opt-outs): [DataPrivacy@options-it.com](mailto:DataPrivacy@options-it.com) Operational support and security incidents: [privatemindsupport@options-it.com](mailto:privatemindsupport@options-it.com)