How Standard Bank Group automates AI risk assessment for generative AI
Standard Bank Group automated its generative‑AI risk assessments by building an end‑to‑end pipeline that runs entirely in its own AWS account, using Amazon S3 for storage, the AWS Well‑Architected Tool for intake, AWS Step Functions for orchestration, and a foundation model accessed via Amazon Bedrock as an “LLM‑as‑a‑judge” to evaluate evidence against internal and international AI standards. The system produces automated reports and a dashboard that feed directly into the bank’s existing risk‑committee approval gates, delivering consistent, auditable evidence without manual review. This approach lets the bank maintain its strict security and compliance posture while accelerating the rollout of new generative‑AI use cases.
⚡ Key Takeaways
- The assessment pipeline stores documents, code, and test results in an Amazon S3 bucket in the bank’s AWS account.
- AWS Step Functions orchestrates the end‑to‑end workflow from intake to reporting.
- Amazon Bedrock provides the foundation model that judges evidence, with access controlled via AWS IAM and no prompt or output data shared with model providers.
- The AWS Well‑Architected Tool is used for the intake step, ensuring each use case is captured uniformly.
Engineers can now embed automated, policy‑driven AI risk checks into existing governance workflows, reducing manual review time and scaling compliance for rapidly evolving generative‑AI applications.
✅ Practical Steps
- Store all generative‑AI use‑case artifacts (documents, code, test results) in an Amazon S3 bucket within your AWS account.
- Configure the AWS Well‑Architected Tool to capture intake data for each new use case.
- Build an AWS Step Functions state machine that invokes an Amazon Bedrock foundation model to evaluate the collected evidence and generate a risk‑assessment report.
Want the full story? Read the original article.
Read on AWS ML Blog ↗