[Bug] Cyber safeguards trigger on benign prompts such as "hi" across multiple models

Bug Description
Claude Code is triggering the cybersecurity safeguard on completely benign prompts.
For example, in a fresh session, sending only:
hi
immediately returns an API error with:
Details: [cyber]
I reproduced this on multiple models, including Opus 4.8 and Opus 4.6. The behavior is consistent and does not appear to depend on cybersecurity-related prompt content.
Steps to reproduce

1. Start Claude Code.
2. Open a fresh session.
3. Send hi.
4. Observe that the response is blocked by the cyber safeguard.
5. Switch to another available model and repeat.

Expected behavior
A benign greeting such as hi should receive a normal response and should not trigger the cybersecurity safeguard.
Actual behavior
The request is blocked with a cybersecurity safeguard error:
Details: [cyber]
Environment

• OS: Linux / Kali Linux
• Claude Code version: 2.1.287
• Terminal: xterm-256color
• Account type: Claude Max
• Models reproduced on: Opus 4.8, Opus 4.6
• MFA: enabled
• Identity verification: completed

Feedback ID
9dc57384-a8ab-4648-ab1e-706b77e35cd7
Example request IDs

• req_011CfcXpYb77BSkxLPZRiuF4
• req_011CfcXqdkRjiGcZ7H2J76VL

I have also contacted Anthropic Support with the same reproduction details

Environment Info

• Platform: linux
• Terminal: xterm-256color
• Version: 2.1.287
• Feedback ID: 9dc57384-a8ab-4648-ab1e-706b77e35cd7