On November 20, 2026, AWS Quick launched enterprise guardrails in preview for Professional and Enterprise plan customers, delivering configurable controls for content safety and sensitive information within user prompts directed at AI agents. Administrators create policies in the Quick administration console encompassing content filters, denied topics, blocked words and phrases, prompt-attack detection, and controls for selected personally identifiable information (PII) types. Policies apply at three hierarchical levels—individual user, enterprise identity provider (IdP) group, or account default—with user-level policies taking precedence over group policies, which in turn override account defaults. When multiple group policies apply, AWS Quick combines their restrictions and enforces the stricter setting for overlapping controls. A detect-only mode enables evaluation of prompt-filtering rules against production traffic without blocking requests; matches are logged for impact assessment and policy refinement prior to enforcement. Once enforced, matching prompts are blocked and a message displayed to the user. All detected and blocked events are reviewable in Amazon CloudWatch Logs. The feature addresses AI agent risks including data leakage, inappropriate content generation, and prompt injection, providing granular governance suitable for diverse organizational sensitivity requirements.

Read original