[Bug] Anthropic API Error: Content Blocked on Non-Security Inputs in Session

Bug Description
I can reproduce the issue with multiple non-cybersecurity inputs in the same session:
HI → Request ID: req_011Cd5UBoS8N5NrwdfCB6KUV
？ → Request ID: req_011Cd5UfYH5swuaCYHfEVS7k
Both inputs are plain greetings or punctuation only. Neither contains cybersecurity content, instructions, targets, code, or technical context.
This indicates that the block is likely being applied at the account, session, conversation-context, or API classification layer, rather than based on the submitted message itself.
Please investigate and reset or correct the affected classification/session state. The issue currently prevents any normal use of Sonnet 5, including starting a conversation with a simple greeting or punctuation.

Environment Info

• Platform: linux
• Terminal: xterm-256color
• Version: 2.1.208
• Feedback ID: 5d7699f4-18a5-4c6c-8e4a-1e1dc35a7e3f

Errors

[]