📋 Description
Self-hosting externally built models involves deploying third-party AI models within an organization's own infrastructure rather than relying on external APIs. This approach reduces external dependencies, enhances data privacy, and provides greater control over security configurations, system performance, and integration with internal workflows.
By hosting models locally (on-premise or in private cloud environments), organizations can better enforce access controls, monitor performance, and manage compliance requirements, especially when handling sensitive or regulated data. Self-hosting can apply to pre-trained models downloaded from providers (e.g., Hugging Face, OpenLLM) or to proprietary models developed by external vendors but run internally.
📉 How It Reduces Risks
- Reduces Data Leakage Risks:
- Eliminates the need to send data to external services, minimizing the risk of unauthorized data access or breaches.
- Improves Compliance and Auditability:
- Ensures sensitive data never leaves the organization’s infrastructure, supporting strict regulatory compliance.
- Enhances Security and Monitoring:
- Self-hosted environments can implement customized logging, firewalls, and monitoring to detect and block unusual behavior.
- Prevents Dependency on Third-party Availability:
- Reduces downtime or service disruptions due to API rate limits, outages, or changes in vendor terms.
📎 Suggested Evidence
- Network Architecture Diagrams
- Documentation showing how the model is deployed within an internal firewall or private cloud.
- System Access Logs
- Logs showing restricted access to model endpoints and absence of outbound API calls.
- Model Hosting Policies
- Internal security and deployment policies that prohibit or restrict external API reliance.
- Compliance Audits
- Reports demonstrating how self-hosting supports requirements for data localization and audit trails
- Security Patching Schedules
- Records of periodic security reviews and updates applied to hosted models.
