AI Usage Blocked By Guardrail

Updated: August 11, 2026

Description

Severity: Medium

A guardrail was invoked and blocked an employee's AI usage.

The content of the interaction matched a guardrail configured by your organisation - for example PII detection, credential detection or a restricted topic - and the guardrail prevented the interaction from completing. This indicates that employees are attempting to share content with AI services that your organisation has classified as unacceptable.

Example Attack

An employee pastes a database connection string containing credentials into an AI chatbot while asking for help debugging an error. The credentials-detection guardrail blocks the prompt before it reaches the AI service, preventing the secret from being exposed to a third party.

Remediation

Review the blocked interactions to understand what content triggered the guardrail. If the guardrail is working as intended, follow up with affected employees to reinforce the acceptable use policy. If legitimate workflows are being blocked, refine the guardrail configuration to reduce false positives.

Security Frameworks

Prompts must be scored by content classifiers against configurable thresholds and rejected or sanitized before reaching the model context.

AI interactions, safety filtering and policy decisions, and inference telemetry must be logged to a structured, interoperable schema to support audit and incident response.