API Discovery

Continuously scans network traffic, API gateways, and load balancer logs to build a real-time inventory of every exposed inference endpoint. This process identifies active, deprecated, and zombie APIs that are no longer maintained but remain accessible.

• Analyzes outbound proxy logs and DNS queries
• Integrates with service mesh telemetry for east-west traffic discovery
• Detects undocumented endpoints created outside formal DevOps pipelines

Analyzes request and response structures to automatically classify discovered endpoints by function and data sensitivity. The system fingerprints the OpenAPI specification, GraphQL schema, or gRPC protobuf to determine if an API serves model inference, training data, or administrative functions.

• Identifies endpoints exposing PII or proprietary model weights
• Flags deviations from approved schema contracts
• Maps data flows to compliance frameworks like SOC 2 and GDPR

Assigns a dynamic risk score to every discovered API based on its exposure level, authentication posture, and data sensitivity. A shadow API—one created outside governance—that serves unauthenticated model predictions receives a critical score and triggers an immediate alert.

• Factors: authentication type, network exposure, data classification
• Correlates with vulnerability databases for known CVE exposure
• Prioritizes remediation based on business impact analysis

Identifies and quarantines zombie APIs—endpoints that are still live but associated with retired model versions or decommissioned services. The system cross-references the discovered inventory with the CI/CD pipeline and model registry to flag endpoints with no active deployment manifest.

• Automatically applies a 'deprecated' header to zombie endpoints
• Tracks invocation frequency to support safe sunsetting
• Prevents attackers from exploiting forgotten, unpatched inference paths

Monitors the API landscape for configuration drift, instantly detecting when a previously secure endpoint becomes exposed due to a firewall rule change or a misconfigured Kubernetes Ingress. This closes the gap between periodic audits.

• Compares current state against a Policy as Code baseline
• Detects newly opened ports and changed authentication policies
• Integrates with SIEM platforms for automated incident response

Visualizes the complex web of dependencies between discovered inference APIs and backend services, model registries, and data stores. This map reveals hidden attack paths where compromising a low-risk API could provide lateral movement to a high-value feature store or vector database.

• Builds a live graph of service-to-service communication
• Highlights transitive dependencies that bypass direct access controls
• Essential for blast radius analysis during incident response

A defensive technique that restricts the number of API requests a client can make within a specific timeframe. For discovered and cataloged inference endpoints, rate limiting prevents model extraction attacks, denial-of-service, and credential stuffing by throttling excessive query patterns.

• Implemented via token bucket or sliding window algorithms
• Critical for protecting newly discovered shadow APIs that lack full security controls
• Should be paired with UEBA to distinguish legitimate bursts from malicious extraction

A granular permission set attached directly to a specific model artifact or version, defining which users, groups, or service accounts are authorized to invoke it for inference. After API discovery catalogs all endpoints, ACLs enforce the least privilege principle on each discovered resource.

• Prevents unauthorized invocation of deprecated model versions still exposed via zombie APIs
• Integrates with RBAC and OAuth 2.0 Client Credentials Flow for machine-to-machine authorization
• Essential for ensuring discovered internal-only APIs are not accidentally exposed externally

A chronological, tamper-proof record of all access and query events against a model serving system, stored in WORM-compliant storage. Once all APIs are discovered, an immutable audit trail provides non-repudiation for every inference call, enabling forensic analysis of unauthorized access.

• Captures who queried which model endpoint, when, and with what payload
• Essential for detecting queries against undocumented zombie APIs that were previously invisible
• Supports compliance with EU AI Act and SOC 2 reporting requirements

A specialized WAF configured with rulesets to inspect and filter HTTP traffic to machine learning APIs. After API discovery maps the full attack surface, a WAF enforces protection against prompt injection, SQL injection, and adversarial perturbations at the network edge.

• Blocks OWASP Top 10 attacks and model-specific threats before they reach the inference runtime
• Can be deployed inline with discovered APIs to provide immediate protection while native security controls are implemented
• Integrates with RASP for defense-in-depth on critical model serving endpoints

A component in a zero-trust architecture that intercepts requests to a protected resource, such as an inference endpoint, and enforces the access decision made by an external policy engine like Open Policy Agent (OPA). For every discovered API, a PEP ensures continuous verification without implicit trust.

• Decouples authorization logic from the model serving application using Policy as Code
• Enforces Just-In-Time (JIT) Access for administrative operations on discovered endpoints
• Prevents lateral movement by validating every request, regardless of network origin