How do I secure AWS Bedrock applications?

Integrate Wardstone Guard with your AWS Bedrock application to detect prompt injections, jailbreaks, harmful content, and data leakage. Wardstone validates inputs before they reach AWS Bedrock and screens outputs before returning them to users.

What security risks does AWS Bedrock have?

Model-Switching Attacks: Attackers can exploit differences in safety training between Bedrock models to find vulnerabilities. Knowledge Base Injection: Malicious content in RAG knowledge bases can be surfaced as trusted information. Agent Manipulation: Bedrock Agents can be manipulated to take unintended actions through prompt injection.

AWS Bedrock + Wardstone

Unified Security for Multi-Model AI

Secure your AWS Bedrock multi-model deployments with Wardstone Guard. Protect Claude, Llama, Mistral, and other foundation models with unified threat detection.

14 Supported Models· all protected with sub-30ms latency

Claude 3.5 SonnetClaude 3 OpusClaude 3 SonnetClaude 3 HaikuLlama 4 ScoutLlama 3.3 70B+8 more

Try the Playground Visit AWS Bedrock

Why Secure AWS Bedrock?

Model-Switching Attacks

High Risk

Attackers can exploit differences in safety training between Bedrock models to find vulnerabilities.

Knowledge Base Injection

High Risk

Malicious content in RAG knowledge bases can be surfaced as trusted information.

Agent Manipulation

High Risk

Bedrock Agents can be manipulated to take unintended actions through prompt injection.

Security Considerations

1
Bedrock Guardrails are model-specific, not unified across providers
2
Different models have different vulnerability profiles
3
Wardstone provides consistent security regardless of which model you use
4
Works with Bedrock Agents and Knowledge Bases

How to Integrate

Add Wardstone to your Lambda or container
Include the Wardstone SDK in your Bedrock-calling code.
Validate before InvokeModel
Screen all inputs before calling the Bedrock InvokeModel API.
Screen model responses
Validate outputs regardless of which foundation model was used.
Secure Knowledge Base content
Scan documents before ingestion into Bedrock Knowledge Bases.

Pricing Note

Bedrock charges per model per token. Wardstone provides unified security pricing across all Bedrock models.

Secure AWS Bedrock with Wardstone

# Step 1: Check user input with Wardstone
curl -X POST "https://api.wardstone.ai/v1/detect" \
  -H "Authorization: Bearer YOUR_WARDSTONE_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "User message here"}'
 
# Response: { "prompt_attack": { "detected": false, ... } }
 
# Step 2: If safe, send to AWS Bedrock (requires AWS SigV4 signing)
aws bedrock-runtime invoke-model \
  --model-id "anthropic.claude-opus-4-5-20251101-v1:0" \
  --body '{"anthropic_version":"bedrock-2023-05-31","max_tokens":1024,"messages":[{"role":"user","content":"User message here"}]}' \
  --region us-east-1 \
  output.json
 
# Step 3: Check Bedrock response with Wardstone before returning to user

Common Use Cases

Multi-model orchestration

Model comparison and A/B testing

AWS-native AI applications

Hybrid cloud deployments

Serverless AI functions

All Supported AWS Bedrock Models

Wardstone Guard protects all AWS Bedrock models with the same comprehensive security coverage. Whether you're using the latest releases or legacy models still in production, every API call is protected.

Claude 3.5 Sonnet

Claude 3 Opus

Claude 3 Sonnet

Claude 3 Haiku

Llama 4 Scout

Llama 3.3 70B

Llama 3.2 90B Vision

Llama 3.1 405B

Mistral Large

Mistral Small

Cohere Command R+

Amazon Titan Text