Skip to content
Wingback Security

Platform capabilities

A factual reference for the Wingback Security platform: supported model providers, detection and red-team coverage, integrations, framework mappings, deployment models, and endpoint and MCP coverage.

Updated

A plain, factual reference for what the Wingback Security platform does. Wingback delivers Agent Detection & Response across four areas: discovery, runtime defense, red teaming, and governance.

Discovery (AI Surface Intelligence)

  • Agentless discovery of agents, models, MCP servers, and inference paths across cloud, SaaS, code, and endpoints.
  • 30+ AI surface categories; 40+ cloud misconfiguration checks; each asset scored 0 to 100 on configuration, exposure, change history, and ownership.
  • Cloud and SaaS coverage: AWS Bedrock and AgentCore (models, agents, knowledge bases, guardrails, flows), Azure OpenAI and AI Foundry, Microsoft Copilot Studio, Salesforce Agentforce, RAG indexes and vector stores. GCP/Vertex is on the roadmap.
  • CycloneDX 1.6 AI bill of materials (AIBOM) per model and agent, versioned and exportable.

Runtime defense (Agent Detection & Response)

  • Inline enforcement: block, sanitise, throttle, terminate, or alert on agent actions, tool calls, and data flows.
  • AI gateway across 12+ model providers (OpenAI, Anthropic, Google, Mistral, AWS Bedrock, Vertex, Gemini, Groq, Cohere, Ollama, OpenRouter, Perplexity) under one policy. Deploy as a drop-in proxy, a Kubernetes sidecar, or an SDK guard, with no agent code changes.
  • Four-tier prompt defense: signatures, embeddings, a Prompt Guard classifier, and an LLM judge; resistant to base64 and homoglyph obfuscation.
  • Data protection in both directions: 25+ PII entity types plus secrets, with detect, mask, or block.
  • MCP gateway: stdio, SSE, and HTTP transports; per-tenant credential isolation; tool input/output scanning; a curated server catalog.
  • Agent guardrails: per-agent tool and model allowlists, budget caps, recursion limits, handoff and intent tracing, and a kill switch. Every session is replayable.

Red teaming

  • An adaptive, LLM-driven attacker that probes models, agents, and RAG pipelines.
  • 800+ attack techniques (27 techniques across 36 vulnerability classes), refreshed continuously.
  • Every finding is mapped to a framework and each validated finding compiles into a runtime guard, so the same attack cannot run twice.

Governance & compliance

  • Continuous, audit-ready evidence mapped to 15+ frameworks.
  • Plain-English policy compiled to runtime rules; 1-click audit binder export; retention configurable from 7 days to 7 years with tamper-evident audit trails.

Endpoint coverage

  • Browser sensor for Chrome and Edge with purpose-built adapters for 15+ AI web apps; inline warn, mask, or block before a prompt submits.
  • Native macOS and Windows agents that sweep on-disk AI tools, MCP servers, and skills every five minutes.
  • 20+ endpoint posture checks mapped to the OWASP MCP Top 10; a live feed of detection rules and malicious-skill indicators.
  • MDM rollout (Intune, Jamf, GPO); 60-second heartbeat; browser and native sensors merge into one device identity.

Frameworks & standards

OWASP LLM Top 10, OWASP Agentic Top 10, OWASP MCP Top 10, MITRE ATLAS, ISO/IEC 42001, NIST AI RMF, and the EU AI Act. Wingback is SOC 2 Type II certified and ISO 42001-mapped, with ISO 27001 in progress.

Integrations & deployment

  • Agent frameworks: LangChain, LangGraph, CrewAI, AutoGen, LlamaIndex, OpenAI and Anthropic SDKs.
  • SIEM and SOC: Splunk, Datadog, Snowflake, Chronicle, and standard SIEM webhooks; OpenTelemetry export.
  • Deployment: managed SaaS, dedicated single-tenant, or self-hosted in your own VPC via Helm.
Request a demo