News

Standard Bank Group details its Bedrock workflow for AI risk assessments

The bank says teams must run the assessment at the build stage and its risk committee uses the findings to decide whether applications can go live.

D
Nov 20, 2026 · 1 min read

On November 20, Standard Bank Group described how it uses Amazon Bedrock to assess generative AI applications before launch in a post co-authored with AWS. The bank said the assessment runs in production and supplies formal evidence for its existing approval process. The post says the bank’s risk committee still decides whether applications can go live.

Teams can run the assessment during a proof of concept and must run it at the build stage, the post says. It describes a software agent that drafts answers to about 40 intake questions from project code and documentation. The team building the application reviews the answers, corrects errors and adds missing evidence.

According to the post, a foundation model running on Amazon Bedrock uses documents, source code and outputs from live tests to answer more than 300 yes-or-no questions. The questions check for evidence of controls tied to the bank’s policies and external standards.

The post says the system rates risks using the share of required evidence the model did not find and a severity level the bank’s risk team can adjust. The system links each missing piece of evidence to a proposed fix and the relevant policy or standards clause.

The system presents the findings in an executive presentation, a checklist of fixes, a technical report and a dashboard, the post says. Reviewers confirm findings before an approval decision and can amend answers with a business justification, according to the post.

More news