Four Users, Under Three Hours: Reports of Claude Code's [cyber] Flag Blocking Ordinary Prompts

Between 03:17 and 05:47 UTC on 7 November, four users filed GitHub bug reports saying Claude Code paused or refused requests they describe as ordinary or harmless, showing a "[cyber]" or "safeguards flagged this message" notice. The reports cite Claude Code versions 2.1.318 and 2.1.319.

Issue 108096 says a request to "Reply with one short sentence" was blocked in a normal Claude Code session and worked with --safe-mode; the same user says Claude.ai and the desktop app still worked. Issue 108093 reports "safeguards flagged this message" errors with the details "reasoning_extraction", and says every message failed after the first, including short ones, even after switching models. The user believes it is a false positive.

Issue 108102 is a user trying to flash official software onto their phone, blocked by the same [cyber] tag. Issue 108087 is a one-line complaint about the same flag, with almost no detail. These are one user's experience each; none establishes a cause. The issues were open when we checked, with no comments and no reply from Anthropic.

Key facts

• Issue #108096 says a request to "Reply with one short sentence" was blocked in a normal Claude Code session and worked with --safe-mode; the same user says Claude.ai and the desktop app still worked.
• Issue #108093 reports "safeguards flagged this message" errors with the details "reasoning_extraction", and says every message failed after the first, including short ones, even after switching models. The user believes it is a false positive.
• Issue #108102 is a user trying to flash official software onto their phone, blocked by the same [cyber] tag. Issue #108087 is a one-line complaint about the same flag.
• The reports cite Claude Code versions 2.1.318 and 2.1.319. The issues were open at the time of checking.