Standard Bank uses AWS to automatically assess generative AI risk
Understand this story
Standard Bank Group, in collaboration with AWS Professional Services, has deployed an automated risk assessment solution for generative AI use cases: each use case is scored against international AI standards and the bank's own policies, and reports and dashboards are generated. The entire assessment runs within the bank's own AWS account, with foundation models on Amazon Bedrock verifying evidence in an "LLM-as-a-judge" manner, while approval remains the responsibility of the risk committee. The assessment is optional during the proof-of-concept stage, allowing teams to identify potential gaps early; once in the build stage it becomes mandatory, and the approval process uses its results as formal evidence. The assessment is currently running in Standard Bank's production environment, completing assessments before use cases go live; after connecting to online workloads, assessments will be re-run on schedule or when code or configuration changes, to keep evidence current. The above practices were described by Standard Bank and AWS in a blog post and are their self-reported practices and arrangements, not independently verified results.
AI overview based on source reports · Updated 52 min ago
Coverage timeline
Follow the coverage to explore different perspectives.
- AWS Machine Learning BlogHow Standard Bank Group automates AI risk assessment for generative AI
Standard Bank Group, working with AWS Professional Services, deployed an automated assessment in its own AWS account that scores each generative AI use case against international AI standards and its own policies, using a Well-Architected Tool custom lens, Step Functions orchestration, and an Amazon Bedrock foundation model as an LLM-as-a-judge. The pipeline collects evidence via a pre-assessment agent, examines documentation, code, and live application responses, evaluates them against more than 300 checks, and produces reports and a CloudWatch dashboard, while review and approval remain with the bank's risk committee. The bank reports reducing a traditionally manual two-week review to a risk report generated in approximately two hours across six assessments, with a 14% improvement in use-case risk scores after remediation.
Activity for this story
Not enough continuous observations to show a trend yet.