# AI Policy Statement A summary of the PrivateMind AI Policy: how Options governs the AI systems it provides through PrivateMind. **Version 1.0**, document ID PUB-AI-001. Summary of PrivateMind AI Policy, version 1.0. Approved by: SVP, AI Transformation. ## 1. About this statement Options Technology Ltd. ("Options") provides PrivateMind, a general-purpose private AI platform for enterprise productivity, coding, analysis and research support. It runs open-weight models on Options-controlled infrastructure. This statement summarises the PrivateMind AI Policy, which sets out how Options governs the AI systems it provides through PrivateMind. The policy takes effect on 26 November 2026. ## 2. Scope The AI Management System (AIMS) covers PrivateMind as provided to internal users and clients: the infrastructure, the MLOps control plane, the private inference platform, the PrivateMind chat app, secure workspaces, governance features and AI engineering services. The AIMS does not cover clients' own use of outputs, AI applications clients build on the API, the free demo service, or Options' internal use of AI tools. Options acts as an AI provider. It does not develop or train foundation models. ## 3. Commitments Top management of Options commits to: - meet applicable requirements: law, regulation, contracts, model licences and ISO/IEC 42001; - continually improve the suitability, adequacy and effectiveness of the AIMS; - set and review AI objectives; - protect client data: client prompts, documents and outputs stay under client control and are never used to train, fine-tune or improve any model provided or hosted by Options; - assess risk and impact before any AI system, model or major change goes live, and while it runs; - be transparent with clients about which models they are using, what those models can and cannot do, and how they are hosted and changed; - apply human oversight; - resource the AIMS with people, budget, tools and compute; - support everyone in a role under the policy to contribute to the AIMS and to raise concerns without fear of reprisal. Client data is processed in data centres in London, New York and Iceland. Under the policy, data leaves the client environment only through hybrid routing the client has contracted for and turned on. Hybrid routing is the optional routing of client requests to third-party frontier models. It is disabled by default. The policy requires client notice at least 7 days before a model is changed or retired, or an alias is moved to a different model. New models are announced when they go live. ## 4. Principles The policy sets eight principles and requires the following under each. | Principle | What the policy requires | |---|---| | Privacy and data sovereignty | Client data is isolated per tenant and is never used for training. | | Security | Models run in containers on Options' own servers, from files checked against the publisher's hashes. AI-specific threats such as prompt injection and model tampering are managed. | | Reliability and robustness | Inference is available, performant and consistent across model versions. | | Transparency and explainability | Clients know which model and version answered, its limits and its provenance. | | Fairness and harm avoidance | Known bias and harmful-output risks of each model are assessed and shared with clients. | | Accountability | Every AI system has a named owner, and decisions are recorded. | | Human oversight | Output is assistive and advisory. A named human remains accountable. | | Environmental responsibility | Compute is used efficiently, on low-carbon energy where possible. Energy intensity is measured and reported. | ## 5. Human oversight AI output is assistive and advisory. Options does not support fully automated decision-making with legal or similarly significant effect on an individual. PrivateMind is not designed or intended to be a high-risk AI system. It is not intended for employment or recruitment decisions, credit scoring, biometric identification, access to essential services, or any purpose requiring a legally significant decision without qualified human review. Under the Fair Use Policy and Terms of Service, clients must not make decisions with legal or similarly significant effect on individuals without qualified human review. ## 6. Governance The policy assigns accountability by role. | Role | Accountability under the policy | |---|---| | CEO | Top management for the AIMS. Holds ultimate authority. Approves the AI Policy. | | SVP, AI Transformation | AIMS executive sponsor. Holds the deployment gate for new models and major changes. | | VP, AI & Data | AI System Owner. Owns the AIMS day to day. | | CISO | Accountable for AI risk assessment and AI impact assessment. Holds the DPO function. | | CAO | Owns the legal and regulatory position for PrivateMind. | The policy establishes the PrivateMind Governance Committee. It is the cross-functional body that sets objectives, reviews risk and impact, and decides model onboarding and retirement. The SVP, AI Transformation chairs it. Its members are the VP, AI & Data, the CISO, the CAO, the MD, Product Management, PrivateMind and the VP, Security & Data Analytics. ## 7. Raising a concern Anyone, including clients, end users and suppliers, may report a concern about a PrivateMind AI system or about the adequacy of AI controls. The policy names these channels: - the AI concerns mailbox at [ai-concerns@privatemind.com](mailto:ai-concerns@privatemind.com); - the DPO at [DataPrivacy@options-it.com](mailto:DataPrivacy@options-it.com); - for Options users, the means provided in line with the Whistleblower Policy. The policy requires concerns to be logged, assigned to the control owner, escalated to the AI System Owner and reported to the Committee, within two business days. The person who raised the concern is protected from reprisal and is told the outcome where appropriate. ## 8. Full policy and reissue The full text of the policy is available to clients, prospective clients and auditors on request, subject to redaction of internal control detail. This statement is reissued within 30 days of any approval of the policy. It must not contradict the policy, the AI Privacy Policy or the Fair Use Policy.