Router AI is the infrastructure layer between your apps and AI providers. Every request passes through one consistent security boundary — AWS KMS envelope encryption, key isolation, audit logging, and access control built in, not bolted on.
Router AI is a middleman. Here's exactly what data we see at each layer, what gets persisted, and what doesn't.
A single request passes through multiple network boundaries. Here's what data is visible and persisted at each hop.
Prompt and completion bodies are never written to our database. They exist only in request/response memory during routing.
Security isn't just encryption—it's isolation, redundancy, and blast radius containment built into the infrastructure layer.
No system is perfectly secure. Here's our assessment of realistic attack scenarios, their impact, and architectural mitigations.
These are target response times, not contractual guarantees. We commit to transparent communication during incidents, with acknowledgment and resolution timelines based on severity.
Targets, not guarantees. Actual response times may vary based on incident complexity and available resources.
Severity Examples Acknowledgment Resolution Target Critical API key leak, data breach, production authentication bypass24 hours 7 days High Security vulnerability (RCE, SQL injection), TLS certificate expiry48 hours 30 days Medium Non-critical security issue (XSS in docs, rate limit bypass)72 hours 90 days Low Security suggestion, documentation improvement, feature request1 week Best effort
Common questions about data handling, encryption, compliance, and operational security.
Found a vulnerability? Have a security concern? We take all reports seriously and respond according to the SLA commitments above.
No, by default. Prompt and response bodies pass through our gateway for routing but are not written to our database. We only persist request metadata: model ID, token counts, cost, latency, and HTTP status code. This metadata enables billing, analytics, and incident investigation. If you enable request logging (opt-in feature, not yet available), we may store truncated request/response samples for debugging. This will be clearly documented when the feature ships.
Router AI API keys (sk-*) are stored as irreversible bcrypt hashes — they cannot be recovered even with full database access. BYOK provider keys are protected with AWS KMS envelope encryption: each credential is encrypted with a unique data encryption key (DEK) generated by AWS Key Management Service, wrapped by a hardware-backed master key (FIPS 140-2 Level 2 HSM). The actual encryption uses AES-256-GCM. Decryption requires both KMS API access and a tenant-specific encryption context (tenantId + provider + purpose) — a database dump alone reveals nothing. No key ever appears in application logs, error traces, or metrics. Plaintext DEKs are zeroed from memory immediately after each cryptographic operation.
Yes. Global traffic now routes through AWS us-east-2. The legacy China endpoint has been retired for new integrations. Historical regional infrastructure remains isolated where applicable: Its own database (no cross-region queries) Its own Kubernetes cluster Its own provider credentials Its own domain and TLS certificate Regional data stores and credentials are not merged across environments. New integrations should use the global endpoint.
No. Router AI does not train models or use your request data for any AI development. We are a routing and billing infrastructure layer, not a model provider. Your prompts and responses go directly to upstream providers (OpenAI, Anthropic, etc.). Their data usage policies apply—refer to each provider's terms of service for details on how they handle your data.
If our gateway becomes unavailable, all requests return 503 until service is restored. Active requests fail immediately (no retry on our side). Your application should implement retry logic with exponential backoff. No data is lost—our database persists through outages. Request metadata already recorded remains intact. Once the gateway recovers, new requests resume normally. Mitigation: We use multi-AZ RDS, auto-scaling Gateway pods, and real-time monitoring. For critical workloads, configure provider fallback routes (if available in your plan).
SOC 2 Type II is in our roadmap. In the meantime, this page documents our actual security controls with the same level of specificity a SOC 2 audit would require — you can review the architecture, threat model, and incident response process directly. We do not claim compliance we haven't achieved. If SOC 2 is a hard requirement for your organization, please contact us to discuss timeline — your interest helps us prioritize the audit.
Not at this time. Router AI is not currently HIPAA-compliant and does not sign BAAs. If you handle protected health information (PHI), you should not route it through Router AI until we complete HIPAA certification. HIPAA compliance is planned for a future milestone, but we do not have a firm timeline yet. Contact us if this is a blocker for your use case.
