Tenant Isolation
yAI deploys per tenant inside your Azure or AWS environment. Data, embeddings, model traffic, and audit logs stay inside your perimeter.
Deployed inside your cloud, verified at the claim, and never trained on your data.
Designed to meet the security, privacy, and recordkeeping standards regulated firms run on, with documentation available on request.
yAI deploys per tenant inside your Azure or AWS environment. Data, embeddings, model traffic, and audit logs stay inside your perimeter.
yAI never trains on customer data. Each tenant runs against an isolated model endpoint. No shared corpus. No cross-tenant data flow.
Data is encrypted at rest with AES-256 and in transit with TLS 1.3. Keys live in your Azure Key Vault or AWS KMS. yAI never holds key material.
Private endpoints for storage, secrets, and database services. Network rules deny by default; only HTTPS and authenticated database traffic are allowed. On AWS, Bedrock calls route through VPC endpoints, so model traffic never hits the public internet.
Optional NAT Gateway gives each deployment a static outbound IP. Partners can allowlist your traffic for SFTP transfers, webhooks, and vendor APIs.
Workloads authenticate via Azure Managed Identity and AWS IAM federation. No static API keys in code. Secrets load from your Key Vault at deploy time.
Four roles enforced application-wide: User, Approver, Compliance Officer, and Admin. Approver scope is bound per tenant. Compliance Officer gets read-only visibility into audit and deletion workflows.
Deletions require two approvers. Requests move through pending, approved, rejected, and executed. Self-approval is blocked. Every transition lands in the audit trail with actor, timestamp, and notes.
A post-processing layer checks every claim against the evidence in its cited passages. Each claim returns supported or unsupported, with reasoning and evidence.
Every claim ties back to the source passage that supports it. Citations are structured and exportable, so compliance can trace any output to the underlying document.
AWS Bedrock Guardrails run on every inference call. Content filters block hate, violence, misconduct, and prompt injection. Sensitive-data detection covers emails, phone numbers, SSNs, bank accounts, and credit cards.
Every prompt, retrieval, model response, and generated artifact lands in an immutable audit log. Timestamped, filterable, exportable for compliance review.
Built for SEC 17a-4 books-and-records. Audit events archive daily to write-once Blob storage, retained seven years by default. SHA-256 checksums provide tamper evidence.
Optional reserved Claude capacity per tenant. Sonnet, Haiku, and Opus throughput available on a one-month or six-month commit, keeping production latency off shared capacity.
Third parties that may process customer data on yAI's behalf, bound by data-processing agreements.
| Vendor | Purpose | Region |
|---|---|---|
| Anthropic | Model inference, served via Amazon Bedrock | Customer tenant region |
| OpenAI | Embeddings (text-embedding-3-large) | Customer tenant region |
| Microsoft Azure | Infrastructure host on Azure deployments | Customer-selected |
| Amazon Web Services | Infrastructure host on AWS deployments | Customer-selected |
All subprocessors are bound by data-processing agreements.
yAI was built inside the firms it now serves. Every architectural decision begins with what compliance, IT, and legal will approve. Isolation, audit logging, and guardrails are present by construction.
Full security documentation is available under NDA. We embed on-site during onboarding to walk your security team through the deployment.
| Document | Description |
|---|---|
| Architecture diagrams | Tenant topology and data flow |
| Data flow and retention | What's stored where, retention windows, deletion paths |
| Control mappings | Internal controls aligned to your framework |
| DPA (Data Processing Agreement) | Standard contractual terms |