AIToday
Large Language ModelsAI Coding AssistantsZenn AI/MLPublished: Sep 29, 2026, 22:00 JST

Claude Code bypassPermissions still blocked by classifier

Claude Code bypassPermissions still blocked by classifier

3 Key Points

  1. What happened

    With bypassPermissions and 100 allow rules set, a user's Claude Code still rejected some operations. One denied action targeted another creator's account; stopping another session was also once blocked, though it later passed.

  2. Why it matters

    Permission settings only control which tools may be used; the auto mode classifier separately judges what an operation intends, so “allow everything” is not truly everything, and config files won't reveal why a block occurred.

  3. What to watch

    The workflow hinges on logging refusals rather than routing around them. Watch the ok=0 record format: before logging began, the user hit the same blocked point three days in a row.

WHO IT HITSDevelopers and operators automating work through Claude Code are the ones affected: they must treat a refusal as a recordable outcome rather than a config bug, since permission-mode settings won't clear it.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

The user arrived at this by treating the refusal as a permissions problem first, spending half a day inspecting settings before concluding the cause was never there to find. The block came from the auto mode classifier, a layer distinct from tool-use permission. The refusals were not identical each time: an operation to stop another session was rejected once and later passed, so the same action does not always produce the same result.

Two workarounds were considered and abandoned: replacing the page-side confirmation dialog to remove the confirmation itself, and handing the same task to a different session. The user judged the first as removing the stopping mechanism without changing the operation's content, and the second as moving a refused operation to another executor — both of which avoid the intent of the refusal.

What the user settled on was record-keeping. Before they started logging refusals, they hit the same point three days in a row; the logged fact is what keeps the next day from repeating it. The broader point they draw is that seeing where automation stops may matter more than designing for never stopping at all.

FAQ
Why does Claude Code reject an operation even when everything is allowed?
Permission mode and allow rules control which tools may be used, but a separate auto mode classifier checks what the operation is trying to do. That second check cannot be changed through settings, so “all permitted” can still mean some actions are denied.
What does the denial message say?
It reads “Permission for this action was denied by the Claude Code auto mode classifier” with the reason “Blocked by classifier.” The user notes the word “classifier” points to the cause, not to the permission settings.
How did the user handle blocked operations in practice?
Their procedure says not to work around a refusal, but to record it with ok=0 and move on. The log is one line, for example recording a reaction to another creator on 2026-09-20 as refused by the auto mode safety check.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleGPT-6 Astra attacks likelier than older models: AISI