Standard Bank Group is automating risk assessment for generative AI projects: each new application automatically gathers evidence, runs tests and generates reports against international AI standards and internal policies, while the risk committee retains final approval authority. The governance bottleneck is therefore beginning to shift from manual document review toward automated evidence generation—but accountability has not been handed over to the model.

What this is

In the past, a generative AI application could be completed in days yet remain stuck in compliance approval for much longer, because reviewers had to read every document, test the system and check whether it met fairness, explainability and safety requirements. Standard Bank and AWS Professional Services have integrated this process into the bank's existing model risk management and change approval mechanisms.

The assessment system runs in the bank's own AWS account: documents, code and test results are stored in Amazon S3, AWS Step Functions orchestrates the workflow, and foundation models on Amazon Bedrock evaluate the evidence before the system produces reports and dashboards. One of the methods is LLM-as-a-judge, which uses a large language model to inspect the output evidence of another AI system against defined rules.

Industry view

We believe the value here lies not in “AI approving on behalf of people,” but in making the basis for approval more consistent. Models and tools change every three to six months, making continuous coverage through manual spot checks difficult. Automated assessment can at least turn evidence collection, testing and remediation gaps into a traceable process.

The risks are equally clear: whether automated scores remain stable, whether underlying models understand the subtle risks of financial scenarios, and whether the evidence itself is reliable all require independent validation. A high score from a model does not mean a project is safe, and the risk committee remains accountable for final approval. What the industry needs is auditable, reproducible automation—not governance responsibility repackaged as a one-time model judgment.

Impact on regular people

For enterprise IT: As companies deploy AI, governance capabilities will increasingly resemble testing and operations infrastructure. Budgets may shift more toward access controls, log retention, automated assessment and interfaces for human review.

For careers: AI projects involving customer service, marketing and internal knowledge management may reach production faster. At the same time, AI compliance, evidence management and process auditing will create demand for new roles.

For consumers: As banks adopt systems of this kind, AI-enabled services will become more common. But faster approval does not mean consumer risks disappear: explanations, appeals and data boundaries in sensitive scenarios will remain critical.