Claude Code keeps flagging harmless shell commands as [cyber] and downgrading me to Opus 4.8
retained text
Hi,
Since the Opus 5.5 launch, Claude Code constantly interrupts me with this:
> 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]`
It happens almost exclusively in **Claude Code**, not the web app (I've only hit it twice on claude.ai). Which makes me think the `[cyber]` classifier over-reacts to anything that looks like a shell command in a terminal context.
Concrete example. My Serena MCP server dropped its connection, so I re-ran the exact command from their docs to restart it:
uvx --from git+https://github.com/oraios/serena serena start-mcp-server --context ide-assistant
That's it. `uvx` fetches and runs the Serena MCP server in one shot, no permanent install. Serena is a well-known open-source tool that gives Claude persistent memory across sessions. I've used it for a while, it's completely benign. Re-running that one command triggered the `[cyber]` flag and bumped me to Opus 4.8.
Anthropic says these safeguards fire in "less than 5% of sessions" and are tuned conservatively on purpose. I get being cautious with a more capable model. But flagging a routine `uvx` command to start a memory MCP is a textbook false positive, and it happens often enough that the model I'm paying for feels like a coin flip in Claude Code.
Anyone else getting the `[cyber]` flag on normal dev commands? What sets it off for you?document fetched 2026-09-27 00:37 · canonical https://reddit.com/r/ClaudeCode/comments/1wqloz0/claude_code_keeps_flagging_harmless_shell