🧩 Provider Agnostic

Supported Integrations

EthicalZen works with every major AI provider. Add governance guardrails to any LLM with zero code changes through our transparent proxy gateway.

5+ Providers
6 Bedrock Modes
<5ms Overhead

Choose Your Provider

EthicalZen integrates transparently with any LLM provider. Route traffic through the ACVPS Gateway and gain instant safety governance.

G

OpenAI

SDK Wrapper + Proxy
  • GPT-4o, GPT-4, GPT-3.5 Turbo support
  • Streaming response validation
  • Token cost limiting
View Docs
A

Anthropic / Claude

SDK Wrapper + Proxy
  • Claude 3.5, Claude 3, Claude 2 support
  • Response grounding checks
  • Constitutional AI complementary
View Docs
G

Groq

Proxy
  • Ultra-low latency inference
  • LLaMA 3.1, Mixtral support
  • Default demo provider
View Docs

Custom / BYOL

Proxy + Middleware
  • Any REST/HTTP API endpoint
  • Self-hosted models (Ollama, vLLM)
  • Azure OpenAI, custom deployments
View Docs

AWS Bedrock Integration Modes

Six flexible patterns to combine EthicalZen with AWS Bedrock Guardrails. Mix and match by business unit, compliance tier, or risk class.

Mode A: Complement

Keep Bedrock Guardrails for general content moderation; add EthicalZen Smart Guardrails for domain-specific and enterprise controls Bedrock does not provide.

Customer App
ACVPS Gateway
+ Smart Guardrails: domain ML, FMA gaps, embeddings, contracts
Bedrock InvokeModel
+ Bedrock Guardrails: content filters, PII, denied topics

Role Split

ComponentPrimary Responsibility
AWS Bedrock GuardrailsGeneral content filters, basic PII, denied topics, word filters
EthicalZen Smart GuardrailsDomain ML guardrails, FMA gap analysis, deterministic contracts, prompt leakage detection, custom embedding classifiers
ACVPS GatewayTraffic proxy, chaining, logging, and policy version pinning
Example Use Case

Healthcare assistant uses Bedrock Guardrails for general moderation and EthicalZen Smart Guardrails for medical advice safety, mental-health crisis detection, and HIPAA-specific PII patterns that exceed Bedrock defaults.

Implementation Notes

  • Deploy ACVPS as an L7 proxy (sidecar, ingress, or API gateway plugin) and route Bedrock traffic through it
  • Run EthicalZen pre-checks before InvokeModel; let Bedrock Guardrails run at inference; optionally run EthicalZen post-checks for contract validation
  • Pin contract versions and guardrail versions per environment (dev/stage/prod)

Mode B: Orchestration

Design guardrails in EthicalZen; provision and lifecycle-manage Bedrock Guardrails via API at scale across many apps and tenants.

SentryWorks Policy
EthicalZen Portal
Alex Agent + FMA + DC Contracts
Bedrock CreateGuardrail API
guardrailIdentifier + guardrailVersion

Role Split

ComponentPrimary Responsibility
EthicalZenGuardrail design (Alex Agent + FMA), lifecycle management, multi-tenant governance, cross-provider orchestration
AWS BedrockRuntime enforcement for Bedrock-hosted models using guardrailIdentifier + guardrailVersion
SentryWorksOptional: upstream policy definition feeding EthicalZen guardrail designs
Example Use Case

Enterprise with 50 Bedrock-powered apps designs guardrails conversationally in EthicalZen, runs FMA to discover gaps, then auto-pushes configurations to Bedrock using CreateGuardrail/UpdateGuardrail APIs. One contract governs all apps.

Implementation Notes

  • Implement a Bedrock Guardrails provisioning service in EthicalZen with tenant-aware IAM roles and least-privilege permissions
  • Store guardrail metadata (identifier, versions, last-updated, policy hash) for drift detection and rollback
  • Expose promotion workflows (draft → review → active) with approvals and automated tests

Mode C: Migration

Start with Bedrock-only and incrementally transition to provider-agnostic enforcement without changing application code.

Phase 1
App
Bedrock + Guardrails
Phase 2
App
ACVPS Gateway
Bedrock + Guardrails + EZ Monitoring
Phase 3
App
ACVPS + EZ Guardrails
Bedrock (no guardrails)
Phase 4
App
ACVPS + EZ Guardrails
ANY LLM

Role Split

ComponentPrimary Responsibility
ACVPS GatewayTransparent insertion and provider routing via headers (e.g., X-Target-Endpoint)
EthicalZenGuardrail parity mapping, deterministic contracts to preserve continuity, provider-agnostic enforcement
AWS BedrockGradually reduced role from primary enforcement to optional monitoring to removed
Example Use Case

Fintech using Bedrock Guardrails adds direct provider routes (e.g., Anthropic Claude API and Groq for cost). They insert ACVPS, replicate Bedrock rules as EthicalZen Smart Guardrails and contracts, then route to any provider while keeping a consistent safety posture.

Implementation Notes

  • Start with shadow mode: EthicalZen evaluates but does not block; compare results against Bedrock block/allow decisions
  • Automate parity tests: generate a policy test suite and validate equivalence before switching enforcement
  • Introduce provider routing rules by workload (latency tier, cost tier, geography, compliance tier)

Mode D: Dual-Layer

Defense in depth with three validation layers: EthicalZen pre-check, Bedrock at-inference, and EthicalZen post-check. For regulated and high-risk workloads.

Customer App
Layer 1: EthicalZen PRE-check
Smart Guardrails, prompt injection detection, contract enforcement
Layer 2: Bedrock AT-inference
Content filters, contextual grounding, PII redaction
Layer 3: EthicalZen POST-check
Response validation, contract envelope, certificate compliance
Safe Response

Role Split

ComponentPrimary Responsibility
EthicalZenPre-flight guardrails and post-flight validation, audit trail, certificate enforcement
AWS BedrockAt-inference moderation, grounding checks, and native PII redaction
ACVPS GatewayChaining, latency management, and observability across layers
Example Use Case

A regulated bank needs SOC 2 and OCC-aligned controls. EthicalZen contracts enforce financial advice rules (pre + post), while Bedrock handles native moderation and grounding. Compliance evidence includes EthicalZen certificates and Bedrock guardrail version pinning.

Implementation Notes

  • Keep latency budgets explicit (per-layer SLOs) and allow tiered enforcement (strict for prod, relaxed for dev)
  • Use structured refusal responses and standardized reason codes to reduce support burden
  • Instrument block/allow decisions at every layer for drift and incident response

Mode E: Governance

Define policies once in SentryWorks; auto-configure both EthicalZen provider-agnostic guardrails and Bedrock-native guardrails. Policy-as-code governance.

SentryWorks.ai Policy Hub
↓ governance policies
EthicalZen Portal
Alex Agent → Smart Guardrails → DC Contracts → ACVPS Gateway
AWS Bedrock (via API)
CreateGuardrail → guardrailIdentifier → Applied to InvokeModel

Role Split

ComponentPrimary Responsibility
SentryWorksPolicy definition, compliance rules, audit requirements, cross-platform governance
EthicalZenPolicy-to-guardrail translation (FMA + Alex Agent), contract generation, enforcement outside Bedrock, certificate authority
AWS BedrockNative enforcement on Bedrock-hosted models; configurations auto-managed via API
Example Use Case

A global insurer runs 200+ AI services across Bedrock, OpenAI, and custom models. SentryWorks defines 'no unauthorized medical advice'. It auto-generates an EthicalZen Smart Guardrail for non-Bedrock traffic and a Bedrock denied-topic/content policy for Bedrock-hosted models. One policy yields two enforcement engines.

Implementation Notes

  • Represent policies in a portable intermediate format (policy IR) so multiple back-ends can be compiled consistently
  • Attach evidence requirements (logs, traces, signed configs) directly to each policy object
  • Provide policy simulation and preflight testing before promotion to production

Mode F: Observability Bridge

Unified compliance monitoring across multiple LLM providers. Even if enforcement remains provider-native, get a single pane of glass for all AI traffic.

ACVPS Gateway (transparent)
Bedrock + Guardrails
→ CloudWatch
OpenAI
(no native guardrails)
Groq / Custom
(no native guardrails)
Unified Metrics (Prometheus / OTel) → Single Dashboard

Role Split

ComponentPrimary Responsibility
EthicalZenTelemetry collection, dashboards, effectiveness scoring, drift detection, A/B comparison of enforcement decisions
AWS BedrockPrimary runtime enforcement for Bedrock traffic
ACVPS GatewayCapture request/response pairs and emit metrics/traces in a consistent schema
Example Use Case

An ML platform team runs five providers. Bedrock traffic uses native Bedrock Guardrails; ACVPS captures all request/response pairs for unified reporting and to identify where policies behave differently across vendors.

Implementation Notes

  • Adopt OpenTelemetry spans per request with consistent attributes (tenant, app, model, policy version, decision code)
  • Store aggregate metrics plus sampled payloads under strict data handling and retention policies
  • Use observability to drive continuous improvement: identify false positives/negatives and refine guardrails

Mode Comparison

Pick the right pattern by app portfolio, compliance tier, and vendor flexibility needs.

Dimension A: Complement B: Orchestration C: Migration D: Dual-Layer E: Governance F: Observability
Best for Add domain controls Manage Bedrock at scale Reduce provider lock-in Regulated workloads Policy-as-code Cross-provider analytics
Bedrock dependency High High Decreasing Medium Low Optional
EthicalZen role Supplementary Lifecycle manager Primary enforcer Pre + post checks Policy translator Telemetry hub
Provider agnostic No No Yes (Phase 4) Partial Yes Yes
Migration effort Low Medium Incremental Medium High Low
Primary value Specialized guardrails Single pane of glass Provider independence Defense in depth Unified compliance Drift detection

Recommended Approach

Start simple, expand as governance needs grow.

A

Start: Complement

"We add what Bedrock doesn't have." Fastest time-to-value with domain-specific Smart Guardrails.

D

Expand: Dual-Layer

"Defense in depth for regulated industries." Three validation layers for maximum safety.

E

Platform: Governance

"SentryWorks as the control plane." Policy once, enforce everywhere across all providers.

Integration Modes by Use Case

Use CaseTypical ConstraintsRecommended Modes
Regulated customer supportStrict refusal, audit evidence, PII redaction, injection resistanceD + E (plus F for analytics)
Healthcare guidance / triageMedical advice safety, crisis escalation, HIPAA PII patternsA → D, then E for scale
Enterprise internal copilotsLeakage prevention, IP/PII controls, cost limitsA + F, then B for lifecycle
Multi-provider platform teamVendor flexibility, uniform reporting, drift detectionF + C, then E
High-volume experimentationFast iteration, A/B tests, preflight simulationB + F, then E
📄

Download Full Integration Guide

Get the complete 12-page AWS Bedrock integration analysis as a PDF for offline reference and sharing.

Download PDF