Security, Privacy & AI Governance by Design. Here Are the Facts.

AI Exchange is responsible for the security and operation of the AI Exchange platform; responsibilities outside that platform boundary remain with the relevant participants unless AI Exchange explicitly assumes them under the applicable service or transaction terms.

Terms, Privacy, Cookies and Legal Notice are governed, versioned documents. English governs; no translation is approved. Each document states what is still pending before production.

Not claimed: The operating legal entity, its registration and its address.

Listed as pending before production in the Legal Notice.

Not claimed: The platform’s controller or processor role under GDPR, a data-protection contact, and contractual data-residency commitments.

Not yet decided. The processing locations above are the configured facts, not a commitment.

Protected information identified in supplied material is replaced by a safe, non-identifying representation before it may become durable platform context, in both the stored content and its identifiers, while the useful meaning and the record of where it came from are preserved.

This governs information the platform has identified as protected. It is not a claim that every possible identifier in every document is detected.

When an unauthorised representation of protected information reaches a durable boundary, the platform refuses the write rather than storing it, and records that the refusal happened.

The application and its data are deployed in Azure Spain Central; the AI models (Azure AI Foundry) in Azure Sweden Central. Both are EU regions.

Document understanding is bound to the EU data zone of the model resource. Live voice discovery is configured to allow global processing; EU-only processing is therefore not guaranteed for that path.

Where EU-only processing is required, live voice should not be used.

Database backups are retained for 7 days in the same region. Geo-redundant backup is off.

Hosting, database, storage and models: Microsoft Azure. Transactional email: Hostinger. No other processor is declared.

This is the set the configuration names, not a reviewed subprocessor list.

An account must verify its email address before it can sign in. Sign-in is refused until the address is verified.

A password-reset request for an unknown address is answered exactly like one for a known address, so the form cannot be used to discover accounts.

Organizations are separate tenants. A private organization’s profile is withheld from other signed-in users, not only from anonymous visitors: a session is not authorization over another organization.

Not claimed: Single sign-on or federated sign-in for customer accounts, and multi-factor authentication.

Customer sign-in is email and password with mandatory verification. No customer SSO or MFA is offered today.

HTTPS is enforced on the public surface.

Workloads authenticate to platform infrastructure by identity rather than stored credentials; secrets are held in a key store and referenced, never copied into configuration.

Infrastructure is defined as code, released through a controlled process, and its deployed posture is checked against AI Exchange’s declared architecture and controls.

This is the platform checking itself. It is not an independent audit.

Artifacts uploaded for a capability are assessed for known vulnerabilities in their components before publication, and the assessment is bound to the exact material assessed.

A capability version whose uploaded artifacts carry unresolved security findings cannot be published. The refusal is made by the server and cannot be overridden from the publishing form.

Uploaded files are analysed in an isolated environment for known vulnerabilities, exposed secrets and insecure configuration before any assistant or reader may use them; material that fails that analysis is never used. Antivirus-style malware scanning is not performed today.

Model access is governed through a provider-neutral gateway to deployments AI Exchange operates on Azure AI Foundry, under the platform’s own identity; application code holds no provider keys, and no other model provider is used today.

AI Exchange’s assistants help an organization describe a business need, help a builder describe a capability, and assess how well a capability fits a stated need. A fourth assistant is reachable only by AI Exchange’s own platform operators and answers questions about the platform from its governed records. They produce assistance and recommendations, never binding decisions.

AI Exchange does not claim a formal model-evaluation programme today.

Discovery does not produce a verdict. A search result, a fit assessment and a buyer approval are distinct records, and fit is assessed against the organization’s own declared context and constraints.

Consequential actions require a human decision: publishing a capability version, selecting a proposal, and contacting an external party on the platform’s behalf. No automated component performs them.

Not claimed: The platform’s role under the EU AI Act, its transparency position for AI-generated output, and model evaluation results.

Not claimed: No assessment of the platform against the NIST AI Risk Management Framework has been performed yet.

The framework is held as a designated reference source; the assessment itself is pending.

A signed-in person can download their personal data (access and portability) and record access, portability, erasure and rectification requests from their account page, each with a one-month deadline. Erasure is carried out by an authorised operator through a governed procedure whose result is verified before the request is closed.

A person who cannot sign in goes through the contact form. Backups keep a copy for 7 days after erasure.

AI Exchange applies data-minimisation and data-protection-by-design controls, so information the platform processes is not automatically retained. Sensitive or customer-specific information, including information supplied unintentionally, is subject to governed minimisation, retention and deletion controls designed to prevent unnecessary persistence -- and to prevent it being reconstructed from what the platform derived from it.

Retention periods are not yet decided, and nothing is expired automatically until they are. A deleted object remains in the storage provider’s protected recovery window before it disappears. Retention by external model providers is stated separately and is not established.

Not claimed: Retention periods for account data, discovery conversations and uploaded material.

Not yet decided. Data is kept while the account exists and nothing is expired automatically until a period is decided.

No certification — SOC 2, ISO 27001 or any other — is claimed, and no independent audit or attestation is claimed.

Not claimed: No assessment of the platform against the CSA Cloud Controls Matrix has been performed yet.

Not claimed: A dedicated security contact, a published vulnerability-disclosure policy, an incident-response process and penetration-test results.

To report a security concern today, use the contact form.

AI Exchange currently operates a single pre-launch environment. No separate production environment exists yet.

Releases follow a controlled process: automation prepares and verifies a release; publishing it to a production environment is a decision reserved to a person.

Not claimed: Availability targets, maintenance windows and status reporting.

A self-assessment of the deployed workload against the five pillars of the Azure Well-Architected Framework was completed on 2026-09-18. It found established controls and material improvement priorities; the scored results are supplied under deeper assurance.

A self-assessment by AI Exchange, not a review by Microsoft or an independent party.

Azure Advisor provides an additional ongoing configuration and operational signal for the deployed subscription; its recommendations are tracked together with the review findings.