Nov 18, 2026: Log mode for prompt injection guardrails¶
Prompt injection guardrails in Cortex AI Guardrails
now support an optional mode setting in the AI_SETTINGS account parameter. Set mode: LOG to scan
and record possible prompt injections in CoCo, Snowflake CoWork, and Cortex Agents without changing
responses. Set mode: BLOCK, the default, to act on detections.
Use LOG mode to evaluate guardrails on your AI traffic before you block threats. Detections in LOG
mode appear in the CORTEX_AI_GUARDRAILS_USAGE_HISTORY view.
For more information, see Set the guardrail mode.