Customer data used to train models
No
Customer data and prompts are not used to train foundation models.
AI Data Security
AI data security answers for enterprise review
DataSafeHouse supports procurement, security, and architecture review with direct answers on model training, deployment boundaries, data isolation, and supported AI supplier options.
Deployment options
On premises or AWS VPC
Deployments can be delivered inside a customer-managed environment, including on-premises infrastructure or an AWS VPC.
Tenant and data isolation
Supported
Tenant and application boundaries, scoped keys, and role-based administrative controls support customer-specific segregation.
Supplier model options
Deployment-specific
Supported model and provider options may include AWS Bedrock-based models, OpenAI-compatible endpoints, Gemini, and local or self-hosted models.
Artifacts Reviewed
Materials typically provided during review
Data flow diagram and deployment topology
Customer-specific data flow diagrams and deployment topology documentation are prepared during solution design and provided during architecture or security review.
Integration inventory and authentication methods
Integration scope, connector boundaries, and authentication methods are documented per deployment based on the systems included in the engagement.
Technical documentation
Architecture, deployment, and operational documentation are available during technical review, including platform control points and operational assumptions.
Hosting platform
Supported deployment models include customer-managed on-premises environments and AWS VPC deployments.
Questionnaire Answers
AI data security questionnaire
Governance structure for AI in place for use, safety, and ethics?
YesDataSafeHouse applies governance through scoped administrative access, role-based console operations, policy inheritance, provider and model controls, rate limits, and audit-ready event telemetry.
Security protocols for model training and deployment?
YesCustomer data and prompts are not used to train foundation models. Deployments can run on premises or in an AWS VPC, with deployment-specific network, access, and operations controls defined during implementation.
Policy in place that restricts use of unapproved AI systems?
YesProvider and model access can be controlled through policy resolution at tenant and application scope, including explicit allowlists and request-time enforcement.
Customer AI data segregated and protected from unauthorized access?
YesTenant and application boundaries, scoped keys, role-based administrative permissions, and deployment-specific network boundaries support segregation and access control.
Private AI tenant available for customer-specific deployment?
Deployment-specificDataSafeHouse supports customer-specific tenant and application boundaries. Dedicated deployment models can be scoped according to customer requirements, including on-premises or AWS VPC environments.
Trains foundation models using customer data or prompts?
NoDataSafeHouse does not use customer data or prompts to train foundation models.
AI data retention and disposal plan?
Deployment-specificRetention and disposal requirements are defined per deployment and customer policy. Environment-specific handling is documented during implementation and security review.
Names of AI suppliers
Deployment-specificSupported model and provider options may include AWS Bedrock-based models, OpenAI-compatible endpoints, Gemini, and local or self-hosted models, depending on deployment design and customer policy.
Customer-specific evidence packages, architecture diagrams, integration inventories, and deployment-specific controls are typically provided during solution review or security due diligence.