- Financial giant leverages AWS for secure AI scaling.
- New Agentic Reference Architecture standardizes development.
- Control Panel offers granular to enterprise-wide AI management.
- FinOps Circuit Breaker prevents runaway cloud costs.
S&P Global, a leading financial information and analytics company, is at the forefront of integrating AI into its operations to amplify essential intelligence for customers and boost internal productivity. Recognizing the inherent challenges in scaling AI agents, particularly in a regulated environment, the company has partnered with AWS to develop a robust framework for agent safety and cost management, primarily utilizing Amazon Bedrock AgentCore.
S&P Global's mission to amplify essential intelligence drives its excitement for AI. They apply AI both externally for customer products and internally to enhance workforce productivity. However, scaling agentic solutions presented significant challenges: ensuring compliance in a highly regulated industry, managing unpredictable cloud costs, and preventing human bottlenecks in the "human-in-the-loop" processes. These concerns underscored the need for a standardized and secure approach.
To address these challenges, S&P Global first established an Agentic Reference Architecture. This framework ensures developers build solutions consistently, considering not just models and prompts but also observability, FinOps metrics, and security from the outset. Their choice of AWS was strategic, leveraging managed infrastructure for scaling, availability, and patching, alongside inherent security features like IAM and Secrets Manager. This allowed them to build on existing cloud migration knowledge, rather than starting from scratch, accelerating development from months to days.
The core of S&P Global's safety strategy lies in two complementary tools. The Agent Safety Control Panel provides granular control, allowing operators to surgically isolate or terminate individual AI sessions, users, or even entire agents if unexpected behaviors occur. This human-in-the-loop system is designed to mature into programmatic control. Complementing this is the FinOps Circuit Breaker, a deterministic gate tied to cloud budgets. If an agent's spend exceeds predefined thresholds (e.g., 1x, 2x, 3x monthly budget), the circuit breaker automatically applies IAM policies to take specified AWS services offline, preventing runaway costs without deleting underlying systems, allowing for investigation and quick restoration.
This defense-in-depth approach, combining the flexible Control Panel with the decisive Circuit Breaker, provides S&P Global with the confidence to rapidly scale their agentic systems. It allows them to move fast while maintaining essential guardrails against compliance risks, cost overruns, and unpredictable agent behavior. The modular design of the Control Panel ensures it can adapt and grow with the company's evolving AI maturity, eventually enabling more automated responses to anomalies.
“We need a pattern that we could use to detect anomalies not from across a number of different sources. So, not one source, but many sources. We needed humans to remain in control.”
