Control adoption self-assessment

Track how many of the recommended AWS security controls you have adopted across the 6 AI capabilities. This measures control adoption and maturity, not compliance or certification.

ⓘ Your assessment stays in this browser only. Nothing is sent to a server. Export it to a file to back it up or move it between devices. Record evidence as references (ARNs, Config rule names), not secrets.

Coverage ⓘ

0 of 56 adopted

Implemented: 0

Adoption by maturity phase

Foundational

Enhanced

Advanced

Import

✓ Saved

Scope

These are optional labels only. They are not validated and are just saved with your assessment and included when you export it.

Organization Environment AWS accounts

AWS-managed controls are pre-marked as inherited (AWS AOC).

▶ 1. Model inference

c1.1 Scope model access with least-privilege IAM

c1.2 Encrypt inference data with customer managed KMS keys

c1.3 Filter inputs and outputs with Amazon Bedrock Guardrails

c1.4 Log every model invocation with CloudTrail and CloudWatch

c1.5 Authenticate users before they reach the model

c1.6 Protect the endpoint with AWS WAF and rate limiting

c1.7 Keep model access private with VPC endpoints

c1.8 Detect and redact PII in prompts and responses

c1.9 Validate model outputs before downstream use

c1.10 Enable AI-aware threat detection

c1.11 Isolate AI workloads across accounts with guardrails

c1.12 Evaluate models and monitor for drift

▶ 2. Model customization

c2.1 Encrypt the customization job, data, and model with a CMK

c2.2 Use a least-privilege service role for customization and import

c2.3 Protect training-data integrity with S3 Object Lock and versioning

c2.4 Run customization privately in a VPC with no internet

c2.5 Discover sensitive data in training sets with Macie

c2.6 Import models safely with Safetensors, not pickle

c2.7 Detect training-data leakage with canaries

c2.8 Secure the model promotion workflow across accounts

c2.9 Monitor customized models for drift and bias

▶ 3. Retrieval-augmented generation (RAG)

c3.1 Encrypt the knowledge base and vector store with a CMK

c3.2 Use a least-privilege service role for the knowledge base

c3.3 Enforce retrieval access control with metadata filtering

c3.4 Validate and scan documents before ingestion

c3.5 Classify and tag data at ingestion

c3.6 Keep the vector store on private connectivity

c3.7 Secure the vector store and embedding integrity

c3.8 Track data lineage and provenance across the RAG pipeline

c3.9 Ground and validate responses against retrieved sources

▶ 4. Tools

c4.1 Scope tool roles with least privilege and permission boundaries

c4.2 Broker tool credentials through AgentCore Identity

c4.3 Encrypt tool I/O and isolate tool sessions

c4.4 Centralize tools behind AgentCore Gateway with rate limits

c4.5 Filter tool inputs and outputs with Guardrails

c4.6 Keep tool integration on private connectivity

c4.7 Constrain Lambda-based tools

c4.8 Log tool invocations with user attribution and observability

▶ 5. Generative AI agents

c5.1 Give each agent its own scoped identity

c5.2 Isolate agent sessions and secure agent memory

c5.3 Prevent agent memory poisoning

c5.4 Grant agents least-privilege roles with no escalation path

c5.5 Require human approval for high-consequence actions

c5.6 Harden agent prompts and validate input with Guardrails

c5.7 Keep agents private and protect public agent apps

c5.8 Limit destructive operations and justify each tool

c5.9 Trace agent reasoning, memory, and tool chains

c5.10 Adversarially test and review agents

▶ 6. AI applications

c6.1 Centralize identity and least-privilege access for AI apps

c6.2 Encrypt user data and isolate sessions

c6.3 Filter user inputs and AI outputs with guardrails

c6.4 Protect the app interface with WAF and TLS

c6.5 Keep AI service access on private connectivity

c6.6 Store application credentials in Secrets Manager

c6.7 Monitor AI application usage with attributed audit logs

c6.8 Automate response to AI application security events