Skip to content
AI Dev Toolkit.
Esc
  • Loading…

Claude Code · Requests and models

Auto mode is unavailable — the server returned no safety verdict

Fix Claude Code’s auto mode stop, “the server returned no safety verdict”: send another message, check your gateway or proxy, or switch permission modes.

Message

The exact error

Claude Code
Auto mode is unavailable — the server returned no safety verdict for the last 10 responses, so Claude stopped. Send a message to try again, or switch out of auto mode.

Also appears as:

For a single action
The server-side auto mode classifier gave no verdict (timed out), so auto mode cannot determine the safety of <tool>.
For a single action, when the response ended early
The server-side auto mode classifier gave no verdict (the response ended before its verdict arrived), so auto mode cannot determine the safety of <tool>.

Meaning

What does “the server returned no safety verdict” mean?

In auto mode, a classifier on Anthropic’s side can review each action before Claude Code runs it, as part of the session’s model requests. When the server gives no verdict for an action, the action is denied rather than run unreviewed. After ten responses in a row without a verdict, Claude Code stops the turn rather than keep trying blind. It’s a safety stop, not a crash, and reads, searches and edits inside your working directory skip the classifier, so they keep working.

Causes

Why does it happen?

  • A temporary problem with the server-side classifier; on a direct Anthropic API connection, the check failing for an action, for example by timing out
  • An LLM gateway or proxy that cuts short or rewrites streaming responses, dropping the verdict

Fix

How do I fix “Auto mode is unavailable”?

  1. Send another message

    Claude tries again and the count starts over.

  2. If you use a gateway or proxy

    Check that it passes streaming responses through unchanged; see Anthropic’s gateway compatibility guidance.

  3. Use Claude Code’s own classifier requests

    Set CLAUDE_CODE_AUTO_MODE_SERVER=0 before starting Claude Code. On a direct connection to the Anthropic API the variable needs v2.1.281 or later, so update first if it seems to have no effect. On accounts where Claude Code’s own classifier requests are billed, it shows a notice about those charges.

    macOS / Linux
    export CLAUDE_CODE_AUTO_MODE_SERVER=0
    claude
    Windows PowerShell
    $env:CLAUDE_CODE_AUTO_MODE_SERVER = '0'
    claude
  4. Approve actions yourself

    Switch out of auto mode so Claude Code asks before each action. To skip prompts only for commands you trust, add allow rules with the settings.json builder, or gate risky commands with a hook from the hooks builder.

Explained

Why does auto mode check with the server before an action?

Under server-side classifier review, Claude Code asks Anthropic’s server to check the actions auto mode sends for review, instead of making separate classifier requests itself. The server’s verdict then decides each action. If the verdict never arrives, Claude Code denies the action rather than run it unchecked, which is why a temporary classifier problem shows up as denied commands or writes. The verdict is lost when the response ends before the review results arrive, or when they arrive in a form Claude Code can’t read, which an LLM gateway or proxy that cuts responses short or rewrites them can cause.

Explained

What does “gave no verdict” on a single action mean?

The server-side auto mode classifier gave no verdict (…), so auto mode cannot determine the safety of <tool>. is the per-action denial. The words in brackets name the reason when Claude Code can tell, such as timed out, and the rest of the message tells Claude whether one retry can help. Before some retries Claude Code waits; the spinner then shows Auto mode check unavailable with a countdown, and Esc interrupts the turn. Ten responses in a row without a verdict produce the “Auto mode is unavailable” stop. Before v2.1.280, Claude Code denied each such action immediately and never stopped the turn.

This is different from Output blocked by content filtering policy, which is a filter on Claude’s reply, not a missing safety verdict.

FAQ

Questions people ask

Why does Claude Code auto mode contact the server before writing a file?

Under server-side classifier review, Anthropic’s server checks each action auto mode sends for review before Claude Code runs it. If that check gets no answer, the action is denied rather than run unchecked, so a temporary classifier problem shows up as denied writes or commands. Reads, searches and edits inside your working directory skip the classifier and keep working.

What does “The server-side auto mode classifier gave no verdict” mean?

That is the per-action form: one action was denied because the classifier returned no decision, and the text in brackets names the reason when Claude Code can tell, such as timed out. Claude may retry on its own; the spinner can show “Auto mode check unavailable” with a countdown, and Esc interrupts. Ten responses in a row without a verdict end the turn.

Does this error appear in claude -p and subagents?

Yes. In a non-interactive -p run, the run ends with an execution error and, with the default text output, the message prints on stderr. A subagent that hits the limit stops before finishing, and Claude receives whatever it produced with a note that auto mode stopped it.

Related

Search all Claude Code errors

Keep exploring