Opus 5.5 keeps flagging my non-security Claude Code sessions as [cyber]. Anyone else?
retained text
Hey all, I've been running into this a lot since switching to Opus 5.5 in Claude Code and wanted to check if it's just me. Mid-session I get this message: > "Opus 5.5's safeguards flagged this session... Opus 4.8 is answering instead, or you can edit and retry with Opus 5.5." > Details: \`\[cyber\]\` The weird part is that nothing I'm doing is security related. My setup is basically a personal productivity stack: Obsidian vault notes, Todoist and Google Calendar via MCP, an ATS connector for recruiting work, some Python scripts on launchd, and a Telegram bot. The session that got flagged just now had me asking it to draft a simple post. A few things I'm trying to figure out: 1. \*\*Triggers:\*\* Does anyone know what tends to trip the \`\[cyber\]\` classifier? My guess is the big system prompt (lots of MCP tools, scripts, API tokens, bash access), not the actual request. 2. \*\*Workarounds:\*\* Has anyone found that trimming MCP servers, skills, or CLAUDE.md content cuts down the false flags? 3. \*\*Stickiness:\*\* Once a session is flagged, does it stay on 4.8 for the rest of the session, or does editing and retrying actually get you back onto 5.5? 4. \*\*Feedback:\*\* Has anyone sent \`/feedback\` on this and heard back or noticed it improving? Not complaining too much since the fallback still works, but it's pretty disruptive when it happens halfway through a task. Would love to hear if others are seeing the same thing!
document fetched 2026-10-01 00:25 · canonical https://reddit.com/r/ClaudeAI/comments/1wtzkvn/opus_55_keeps_flagging_my_nonsecurity_claude_code