Sample question 1 — GenAI Application Implementation and Integration
A fintech company runs the same generative platform in development, staging, and production accounts inside one AWS organization. During a recent incident, engineers discovered that developers had invoked the production-grade model from a development account, producing charges and outputs that were never reviewed under the production change process. Leadership now requires a hard boundary so a development identity can never reach the production-approved model. Which control should the team put in place?
- A. Ask developers to sign a policy promising not to call production endpoints.
- B. Set lower temperature values so development outputs differ from production.
- C. Enable detailed billing reports so cross-account usage can be reviewed later.
- D. Use distinct IAM roles plus service control policies per environment.
Answer: D. Option D is correct because giving each account its own IAM role and layering service control policies creates an enforced boundary that blocks development identities from reaching the production-approved model, regardless of individual intent. Option A is incorrect because an honor pledge is not technically enforced and the incident shows it would be bypassed. Option B is incorrect because sampling values change output style but do not stop the call from succeeding. Option C is incorrect because billing reports reveal misuse after the fact and cannot prevent it.