Amazon Bedrock¶

What Is Amazon Bedrock?¶

Amazon Bedrock is a fully managed generative AI service that allows organizations to build and scale AI applications using foundation models from AWS and third-party providers.

Bedrock supports use cases such as:

AI assistants
chatbots
document summarization
content generation
security analysis
automation
retrieval-augmented generation (RAG)

Bedrock provides access to multiple foundation models without managing infrastructure.

Think of Amazon Bedrock as:

A managed generative AI platform for building secure AI-powered applications.

Why Amazon Bedrock Matters for Security¶

Generative AI introduces major security concerns around:

sensitive data exposure
prompt injection
model misuse
unauthorized access
logging and monitoring
data governance

Security teams must understand how to secure:

prompts
AI-generated responses
model access
knowledge bases
API usage
IAM permissions

Organizations increasingly use Bedrock for:

security copilots
automated investigations
threat analysis
security summarization
operational automation

Core Concepts¶

Bedrock provides managed foundation models
applications interact using APIs
IAM controls model access
prompts are sent to foundation models
knowledge bases support RAG architectures
Bedrock integrates with AWS security services
customer data is not used to train foundation models by default

Common Security Use Cases¶

Security AI Assistants¶

Organizations can build AI assistants for:

SOC analysts
operational teams
incident responders
internal support systems

Security Investigation Summarization¶

Bedrock can summarize:

GuardDuty findings
CloudTrail activity
incident reports
investigation timelines

Threat Intelligence Analysis¶

Used to analyze:

indicators of compromise
threat reports
suspicious activity
security findings

Automated Security Reporting¶

Bedrock can generate:

incident summaries
executive reports
compliance documentation
operational insights

Retrieval-Augmented Generation (RAG)¶

Organizations can securely query internal knowledge bases using:

Bedrock Knowledge Bases
vector search
OpenSearch integrations

Security Automation¶

Bedrock can integrate with:

Lambda
EventBridge
Security Hub
ticketing systems

for automated workflows.

How Amazon Bedrock Works¶

Basic Workflow¶

User or application sends a prompt
Bedrock processes the request
Foundation model generates a response
Optional knowledge base retrieval occurs
Results are returned to the application
Logging and monitoring capture activity

Simple Architecture¶

User / Application
        ↓
Amazon Bedrock
        ↓
Foundation Model
        ↓
Knowledge Base / Data Source
        ↓
AI Response

¶

Example Use Case: Secure AI-Powered Security Investigation Assistant¶

flowchart TD
    A[Security Analyst] --> B[Internal AI Security Assistant]

    B --> C[Amazon Bedrock]

    C --> D[Foundation Model]

    C --> E[Bedrock Guardrails]

    C --> F[Knowledge Base]

    F --> G[Amazon OpenSearch Service<br/>Vector Search]

    F --> H[Amazon S3<br/>Security Reports and Logs]

    H --> I[AWS KMS Encryption]

    C --> J[AWS CloudTrail]

    C --> K[Amazon CloudWatch]

    B --> L[AWS Lambda Automation]

    L --> M[Amazon EventBridge]

¶

Important Components¶

Foundation Models¶

Bedrock provides access to models from providers such as:

Anthropic
AI21 Labs
Meta
Cohere
Amazon Titan

Prompts¶

Prompts are inputs sent to foundation models.

Sensitive prompts should be protected carefully.

Knowledge Bases¶

Knowledge Bases support:

retrieval-augmented generation
document querying
enterprise search

Guardrails¶

Bedrock Guardrails help control:

unsafe responses
harmful content
sensitive information exposure

Embeddings¶

Embeddings convert data into vector representations for semantic search and RAG workflows.

Agents¶

Agents can automate:

workflows
reasoning tasks
API interactions
operational actions

Important Integrations¶

AWS IAM¶

IAM controls:

model access
API permissions
administrative access

Amazon OpenSearch Service¶

OpenSearch is commonly used in Bedrock architectures for:

vector search
embeddings storage
semantic retrieval
retrieval-augmented generation (RAG)

This allows AI applications to securely search internal knowledge bases and security datasets.

Amazon S3¶

Used for:

document storage
training datasets
knowledge base content

AWS Lambda¶

Used for:

automation
orchestration
AI workflows

Amazon EventBridge¶

Can trigger:

AI workflows
automation
notifications

AWS CloudTrail¶

CloudTrail logs:

API activity
model invocation
configuration changes

Amazon CloudWatch¶

Provides:

monitoring
metrics
logging
alarms

AWS KMS¶

KMS helps encrypt:

stored documents
vector databases
AI-related data

AWS Security Hub¶

Security findings and alerts can integrate into AI-driven operational workflows.

Security Features¶

Principle of Least Privilege for AI Workloads¶

The IAM role used by AI applications should only access:

approved Bedrock models
required Knowledge Bases
authorized S3 buckets
permitted APIs

Example:

An internal security assistant may require:

bedrock:InvokeModel
access to a specific Knowledge Base
read access to a specific S3 bucket

but should not have unrestricted access to all Bedrock resources or all S3 buckets.

Least privilege access is critical for: - sensitive AI workflows - internal security assistants - enterprise RAG systems

IAM-Based Access Control¶

IAM policies should restrict:

model invocation
knowledge base access
administrative permissions

Generative AI Guardrails¶

Bedrock Guardrails help protect AI applications from:

unsafe responses
toxic content
prompt abuse
sensitive data exposure
restricted topic generation

This is important for implementing secure generative AI applications.

Guardrails are major security controls for:

AI assistants
internal copilots
RAG systems
enterprise AI applications

Encryption¶

Data can be encrypted using:

AWS KMS

for storage and integrations.

Protecting Knowledge Base Data¶

Knowledge base data stored in services such as Amazon S3 should be protected using:

AWS KMS encryption
bucket policies
IAM least privilege permissions
access logging

Even though Bedrock provides the AI capability, organizations must still secure the underlying data sources.

Customer Data Protection¶

By default, customer prompts and responses are not used to train foundation models.

Very important security concept.

Logging and Monitoring¶

CloudTrail and CloudWatch help monitor:

API usage
suspicious access
operational activity

Least Privilege Access¶

Applications should only access:

required models
approved knowledge bases
necessary APIs

Quick Service Identity Triggers¶

Need generative AI protections or output filtering? → Bedrock Guardrails
Need semantic vector search for AI assistants? → Amazon OpenSearch Service
Need AI-powered automation workflows? → Lambda + EventBridge
Need secure AI knowledge base storage? → Amazon S3 + KMS + IAM
Need centralized AI API logging? → CloudTrail + CloudWatch

Common Exam Scenarios¶

Scenario 1¶

A company wants to build a secure generative AI assistant without managing infrastructure.

Answer:

Amazon Bedrock

Scenario 2¶

A company needs semantic search over internal security documentation.

Answer:

Use Bedrock Knowledge Bases with vector search.

Scenario 3¶

A company wants to prevent unsafe AI-generated responses.

Answer:

Use Bedrock Guardrails.

Scenario 4¶

A security team needs centralized logging for Bedrock API activity.

Answer:

Use AWS CloudTrail and CloudWatch.

Scenario 5¶

A company needs to secure model access using least privilege permissions.

Answer:

Use AWS IAM policies.

5-Second Recall for AI Security¶

The Guardrail¶

If the scenario mentions:

filtering toxic content
preventing sensitive data leakage
restricting unsafe AI responses
GenAI OWASP protections

Answer:

→ Bedrock Guardrails

The Knowledge Base¶

If the requirement is:

securely querying internal documents
enterprise AI search
retrieval-augmented generation (RAG)
semantic retrieval over private data

Answer:

→ Knowledge Bases for Amazon Bedrock

Common integrations:

Amazon S3
Amazon OpenSearch Service

The Audit Trail¶

If the scenario asks:

who invoked the AI model
who modified Guardrails
who changed permissions
how to audit Bedrock API activity

Answer:

→ AWS CloudTrail¶

Common Exam Traps¶

Trap 1 — Forgetting Guardrails¶

Guardrails help reduce:

unsafe outputs
harmful responses
data leakage risks

Trap 2 — Ignoring Prompt Security¶

Sensitive information should not be exposed through prompts or AI responses.

Trap 3 — Assuming Customer Data Trains Models Automatically¶

By default, Bedrock does not use customer prompts or outputs to train foundation models.

Trap 4 — Overly Broad IAM Permissions¶

Applications should only access:

approved models
required resources
authorized knowledge bases

Trap 5 — Forgetting Data Protection Responsibilities¶

Even with Bedrock, organizations must still secure:

S3 buckets
vector databases
IAM permissions
encryption keys

Quick Revision Notes¶

Amazon Bedrock = managed generative AI service
supports multiple foundation models
commonly used for AI assistants and RAG
IAM controls model access
Guardrails help control unsafe responses
OpenSearch commonly supports vector search
CloudTrail logs Bedrock API activity
customer prompts are not used for model training by default
Bedrock integrates with Lambda and EventBridge
Knowledge Bases support enterprise AI search
KMS and bucket policies protect AI knowledge base data

Amazon Bedrock¶

What Is Amazon Bedrock?¶

Why Amazon Bedrock Matters for Security¶

Core Concepts¶

Common Security Use Cases¶

Security AI Assistants¶

Security Investigation Summarization¶

Threat Intelligence Analysis¶

Automated Security Reporting¶

Retrieval-Augmented Generation (RAG)¶

Security Automation¶

How Amazon Bedrock Works¶

Basic Workflow¶

Simple Architecture¶

User / Application ↓ Amazon Bedrock ↓ Foundation Model ↓ Knowledge Base / Data Source ↓ AI Response ¶

Example Use Case: Secure AI-Powered Security Investigation Assistant¶

Important Components¶

Foundation Models¶

Prompts¶

Knowledge Bases¶

Guardrails¶

Embeddings¶

Agents¶

Important Integrations¶

AWS IAM¶

Amazon OpenSearch Service¶

Amazon S3¶

AWS Lambda¶

Amazon EventBridge¶

AWS CloudTrail¶

Amazon CloudWatch¶

AWS KMS¶

AWS Security Hub¶

Security Features¶

Principle of Least Privilege for AI Workloads¶

IAM-Based Access Control¶

Generative AI Guardrails¶

Encryption¶

Protecting Knowledge Base Data¶

Customer Data Protection¶

Logging and Monitoring¶

Least Privilege Access¶

Quick Service Identity Triggers¶

Common Exam Scenarios¶

Scenario 1¶

Scenario 2¶

Scenario 3¶

Scenario 4¶

Scenario 5¶

5-Second Recall for AI Security¶

The Guardrail¶

The Knowledge Base¶

The Audit Trail¶

→ AWS CloudTrail¶

Common Exam Traps¶

Trap 1 — Forgetting Guardrails¶

Trap 2 — Ignoring Prompt Security¶

Trap 3 — Assuming Customer Data Trains Models Automatically¶

Trap 4 — Overly Broad IAM Permissions¶

Trap 5 — Forgetting Data Protection Responsibilities¶

Quick Revision Notes¶

`User / Application ↓ Amazon Bedrock ↓ Foundation Model ↓ Knowledge Base / Data Source ↓ AI Response`
¶