Tomorrow at 10 am" triggers a violence content filter block (400) — gpt-4.1, default DefaultV2 guardrail

Hi all,

A plain appointment-booking phrase is being blocked by the input content filter on our Azure OpenAI deployment.

Repro (Foundry playground, no system prompt, no tools, single user message):

• User message: tomorrow at 10 am
• Result: request blocked by the content filter

Error: Interaction blocked This interaction was blocked by a safety and security control in this asset's Foundry guardrail. Risk type: Violence (High) is detected at User Input.

Environment

• Model: gpt-4.1, version 2025-04-14

Guardrail/content filter: Microsoft.DefaultV2 (default, unmodified)

-> I tried testing this with both medium and low threshold for all content safety risk types and still facing the same issue.

This affects several production chatbots that book appointments, so users hit it during a normal booking conversation.

Questions

1. Is a benign date/time phrase expected to be classified as violence at severity "high"? This looks like a false positive.
2. Any recommended mitigation other than relaxing the input filter?

Thanks!