AWS prescriptive-guidance documentation change
Summary
Added glossary definitions for AI-related terms including A2A, Agent, Agent Ops, Citizen Developer, FM gateway, guardrails (AI), human-in-the-loop, MCP, Shadow AI, and tool.
Security assessment
The changes add definitions for security-related concepts like AI guardrails (safety mechanisms), Shadow AI (unauthorized AI applications), and FM gateway (centralized access control for foundation models). These are general security documentation updates, not fixes for a specific vulnerability or incident.
Diff
diff --git a/prescriptive-guidance/latest/agentic-ai-security/apg-gloss.md b/prescriptive-guidance/latest/agentic-ai-security/apg-gloss.md index 5ff2fdf75..5de65bbda 100644 --- a//prescriptive-guidance/latest/agentic-ai-security/apg-gloss.md +++ b//prescriptive-guidance/latest/agentic-ai-security/apg-gloss.md @@ -38,0 +39,5 @@ Seven common migration strategies for moving applications to the cloud. These st +A2A (Agent-to-Agent) + + +A stateful protocol for agent-to-agent collaboration supporting task delegation and state transfer. + @@ -63,0 +69,10 @@ A database migration method in which the source and target databases are kept in +Agent + + +An AI system that can autonomously reason, plan, and take actions using tools to achieve goals. + +Agent Ops + + +Operational practices for building, testing, deploying, and running AI agents in production at scale. + @@ -262,0 +278,5 @@ A categorization process that helps generate predictions. ML models for classifi +Citizen Developer + + +A business user who creates AI applications using no-code/low-code platforms without specialized technical skills. + @@ -662,0 +683,5 @@ A large deep-learning neural network that has been training on massive datasets +FM gateway + + +A centralized intermediary that controls and normalizes access to foundation models. Also known as an _LLM gateway_. + @@ -699,0 +725,5 @@ A high-level rule that helps govern resources, policies, and compliance across o +guardrails (AI) + + +Safety mechanisms that filter, validate, and constrain agent inputs and outputs to help ensure responsible and safe AI behavior. + @@ -726,0 +757,5 @@ A portion of historical, labeled data that is withheld from a dataset that is us +human-in-the-loop (HitL) + + +A workflow pattern where agent execution pauses for human review and approval at critical decision points. + @@ -927,0 +963,15 @@ See Migration Acceleration Program. +MCP + + +See Model Context Protocol. + +Model Context Protocol (MCP) + + +A stateless protocol for agent-to-tool communication. + +MCP server + + +A service that exposes one or more tools through the Model Context Protocol. + @@ -1492,0 +1543,5 @@ A model describing the responsibility you share with AWS for cloud security and +Shadow AI + + +Unauthorized AI applications built or used outside of governed channels within an organization. + @@ -1589,0 +1645,5 @@ To provide data for your ML model to learn from. The training data must contain +tool + + +A function or API that an agent can invoke to perform operations in external systems. +