The AI gateway your security team needs.

Prompt injection detection, DLP rules, and audit logging -- all at the Fastly edge, before requests ever reach an LLM provider.

Data Flow

How your data moves

Your App → Fastly Edge POP → LLM Provider (direct) | | ├─ Injection scan └─ Response ├─ DLP rules ↓ ├─ Cache (KV) Your App ├─ Audit log | * Prompts not stored beyond cache TTL * Audit logs = metadata only * API keys in Fastly Secret Store

ModelRouter runs on Fastly Compute@Edge. Your prompts pass through the nearest POP, are scanned in-memory, and forwarded directly to the LLM provider. No intermediate servers. No third-party logging.

Before / After

What changes when you add ModelRouter

Without ModelRouter

× Employees paste source code into ChatGPT

× Injection attacks extract system prompts

× No visibility into AI usage

× SSNs leaked through prompts

× One provider goes down, product breaks

With ModelRouter

✓ DLP rules block code before it leaves

✓ 30+ patterns blocked in <1ms

✓ Full audit: who, what, when, which model

✓ Regex DLP catches PII patterns

✓ Fallback across 7 providers

Compliance

SOC 2 Readiness Checklist

ModelRouter maps directly to SOC 2 Trust Service Criteria for AI workloads.

✓ CC6.1 Access Control -- Bearer token authentication per user/service. Keys stored in Fastly Secret Store, never in code.

✓ CC7.2 System Operations -- Full audit logging of every request: timestamp, user, model, provider, tier, injection result.

✓ CC7.3 Monitoring -- Real-time audit log viewer with CSV export. Filter by user, date range, provider, or injection status.

✓ CC8.1 Change Management -- DLP rules and routing configuration updated via API. Version-controlled, auditable changes.

✓ CC9.1 Risk Mitigation -- Prompt injection detection (30+ patterns) + custom DLP rules. Blocks threats before they reach providers.

These mappings cover the most common AI-specific controls requested by SOC 2 auditors. ModelRouter does not replace a full compliance program but provides the technical controls needed for AI governance.

Try It

See it in action

Test prompt injection detection live. Enter an attack and watch ModelRouter catch it in real time.

Try the Security Demo → Create Account →