Security at Neural Router

Security is built into the platform, not bolted on. Every request is authenticated, every workspace is isolated, prompt content is minimized and optionally never stored, and policies are enforced server-side so they can't be bypassed per request. The sections below detail our controls; to report an issue, see Vulnerability disclosure.

Our formal compliance program is in progress. We can share our current security overview and data-handling details under NDA for procurement and vendor review.

Customers can sign a Data Processing Addendum (DPA), and zero-retention and residency controls support regulated workloads.

• In transit, TLS 1.2+ for all API and dashboard traffic, with HSTS enforced.
• At rest, AES-256 for stored data and backups.
• Secrets, provider credentials and bring-your-own-key material are encrypted with AES-256-GCM and isolated per workspace. Encryption keys are held in a dedicated secrets manager, separate from the data they protect.

Access to production is least-privilege and role-based, and every administrative action is recorded in the append-only audit trail. Within the product, organizations get role-based access control (owner, admin, member) and granular API key scoping: a key can be limited to specific models, given its own rate limit, and capped with its own spend limit.

API keys are generated with high entropy and shown only once at creation. We store a one-way reference plus the last four characters for display, never the full secret. Revoking a key takes effect immediately. Bring-your-own-provider keys are encrypted at rest and only the last four characters are ever surfaced in the UI.

We minimize what we hold. Routing operates on metadata; prompt and completion content is transient by default and, with zero-retention enabled, is never written to disk. Workspaces can pin inference to a jurisdiction and attach a per-request region audit. We do not use customer prompts or completions to train models. See the Privacy Policy for full detail.

The platform runs on hardened cloud infrastructure with network isolation, automated patching, and infrastructure-as-code change control. Workspaces are logically isolated, and provider failover keeps traffic flowing if an upstream degrades. Inference is routed across many upstream providers and regions; contact us if your deployment requires a specific regional footprint.

We continuously monitor availability and security signals, run health probes against provider endpoints, and auto-suspend providers that breach SLA. We maintain a documented incident response plan and notify affected customers without undue delay, consistent with our contractual and legal obligations.

We welcome reports from security researchers. If you believe you've found a vulnerability, email security@neuralrouter.ai with steps to reproduce. Please give us reasonable time to remediate before public disclosure and avoid accessing or modifying data that isn't yours. We do not pursue legal action against good-faith research that follows this policy.