AI safety, security, and governance cheat sheet
All 14 AIP-C01 flashcards for AI safety, security, and governance on one page, grouped by exam objective. Print it or save it as a PDF for a last-minute read.
Implement input and output safety controls
- Name four Bedrock Guardrails policy types.
- Content filters, denied topics, word filters, and sensitive information filters. Also the contextual grounding check and Automated Reasoning checks.
- What does the contextual grounding check score?
- Whether a response is grounded in the source passages and relevant to the query, with configurable thresholds.
- What is ApplyGuardrail for?
- Running guardrail checks on any text, including content not produced by Bedrock models.
- What is defence in depth for GenAI?
- Layered controls: input pre-filtering, guardrails in and out, output validation, and logging. No single layer is trusted.
- What is prompt injection?
- Untrusted text, user input or retrieved content, that tries to override the model's instructions.
- How do you make text-to-SQL safer?
- Read-only credentials, allow-listed tables, parameterised queries, and validation of generated SQL before execution.
Implement data security and privacy controls
- How do you keep Bedrock traffic off the internet?
- Interface VPC endpoints (PrivateLink) for bedrock-runtime and bedrock-agent-runtime plus endpoint policies and IAM conditions.
- Does Bedrock train base models on your prompts?
- No. Your inputs and outputs are not used to train base models or shared with model providers.
- Comprehend vs Macie for PII?
- Comprehend detects PII entities in text you send it. Macie discovers sensitive data stored in S3.
Implement AI governance and compliance mechanisms
- Is model invocation logging on by default?
- No. You enable it and choose CloudWatch Logs and/or S3 as destinations.
- CloudTrail vs invocation logs?
- CloudTrail records API activity for audit. Invocation logs capture request and response content.
- What is a model card for?
- Documenting a model's intended use, limitations, data and evaluation results for governance and transparency.
- How does the Glue Data Catalog help governance?
- It registers data sources and schemas, giving a record of where data came from (lineage and attribution).
Implement responsible AI principles
- How do you increase transparency for users?
- Cite sources, show reasoning or agent traces where appropriate, and report confidence honestly.