[Bug] Auto mode classifier returns no verdict, blocking Edit/Bash operations

Also hitting this on macOS (Darwin 25.4.0), Claude Code 2.1.283, CLI. Every Bash call failed with "classifier gave no verdict (error)", including echo ok and ls, which is allowlisted via Bash(ls:*). So allow rules don't get around the check in auto mode. Read-only tools kept working. Turning auto mode off fixed it.

Same here. Probably a server outage?

ye its a server side issue

Also seeing this on macOS (Darwin 25.5), Claude desktop app (Code tab), auto mode, Opus 5.5.

• Started around 13:10 UTC on 2026-09-28. Intermittent at first (about 6 failures among many successful calls), and by about 13:35 UTC nearly every call failed.
• Affected: Bash (including read-only grep, ls, git pull, gh auth status) and the built-in browser's navigate. File reads that don't go through the classifier kept working.
• Message: "The server-side auto mode classifier gave no verdict (error), so auto mode cannot determine the safety of Bash. This is a transient failure of the check…"
• Workaround: switching the session out of auto mode (manual approval) skips the classifier, and everything works again.

Same issue here today (2026-09-28).
In my case, Auto mode repeatedly fails on Edit with:
The server-side auto mode classifier gave no verdict (error), so auto mode cannot determine the safety of Edit.
After 10 consecutive responses without a verdict, Claude Code stops completely with:
Auto mode is unavailable — the server returned no safety verdict for the last 10 responses, so Claude stopped.
Retrying does not resolve it. Read-only operations still work, but operations requiring the Auto mode classifier are blocked.

+1, same today. Every Bash, Edit and Write call failed with classifier gave no verdict (error), and Read and Grep kept working. Like others here, the only fix was switching out of auto mode into default (manual approval). Retrying never cleared it, and unlike the original report it didn't recover on its own.

A couple of extra details:

• Stock Anthropic routing. No proxy and no third-party model routing. So the thinking-budget / truncation theory from #49535 doesn't seem to explain this one.
• Timing. It started soon after I changed a setting. I couldn't confirm the change caused it, and the reports here from lots of different setups at the same time point more to a server-side outage.
• Possibly related to the long-running #49535, which is the same symptom under the older <model> is temporarily unavailable wording.

Can confirm — hit this repeatedly in a session today, running in the VS Code extension in auto permission mode.

Same behavior as described: Bash calls fail with the "no verdict" error, retrying sometimes succeeds immediately and sometimes fails again with the identical message. In one stretch this happened ~10 times in a row (which per the documented behavior ended the turn), then cleared on its own a few minutes later with no action taken on my end.

Notably it reproduces even on trivial, side-effect-free commands (echo hello, uptime) — so it doesn't seem related to the complexity or risk profile of the specific command, more like an outage/failure of the classifier service itself. Read-only tools (Read, Grep) were unaffected throughout.

Exact error text each time:

The server-side auto mode classifier gave no verdict (error), so auto mode cannot determine the safety of Bash. This is a transient failure of the check, not a judgment about the action: a later response may get a verdict. You may try the action again once, as-is. Repeated attempts are slowed by a growing delay, and after 10 responses in a row without a verdict the turn stops. If it keeps failing, continue with other tasks that don't require this action and come back to it later. Note: reading files, searching code, and other read-only operations do not require the classifier and can still be used.

Claude Code version: 2.1.284

i also have this issue