Claude Auto Mode Is Now the Default: Who Approves the Form Your Agent Publishes? (2026)

Claude auto mode becomes the default this week across the Claude web, desktop and mobile apps for Pro, Max and Team plans. Anthropic announced the change on October 22 in an update to its auto mode post and on X. In its words: "Rather than asking you to approve each step, an automated check blocks risky actions. Switch back to manual anytime."

If Claude builds and publishes forms for you, the obvious question follows. When the agent stops asking, who approves the form before it goes live? Here is what changed and what still asks.

What Anthropic announced

Auto mode started in Claude Code in August. Instead of a permission prompt on every tool call, a classifier reviews each action. Safe actions go ahead. The classifier blocks actions that look irreversible, destructive or aimed outside your environment. Then Claude looks for another route or asks you directly. Anthropic's post says it tested the classifier with internal and third party red teaming, prompt injection evaluations and a controlled study with 1,053 paid testers.

The October 22 update extends that default beyond Claude Code. From the week of October 26, new sessions in most Claude apps on Pro, Max and Team plans run in auto mode. Enterprise admins get an email before it becomes the default for their organizations, and anyone can switch back to manual. Anthropic also notes that it stopped charging Pro, Max and Team users for the classifier's extra tokens.

The company is careful about the claim. In its own words, auto mode "reduces risk, it doesn't eliminate it". It is a classifier, not a guarantee.

Why this matters if your agent publishes forms

Approval prompts did two jobs. They stopped dangerous commands, and they gave you a moment to look at what the agent was about to do. Claude auto mode keeps the first job and hands it to a classifier. The second job is now yours to design, because a classifier does not know whether a form is ready for the public.

Publishing a form is a good example. It is not destructive in the sense the classifier watches for. Nothing is deleted. However, it replaces what your visitors see, and a wrong form on a live link costs you leads or candidates. So the question is not whether Claude will ask about it. It is whether the tool Claude uses asks.

This is where the connector matters. In dForms, the form builder for AI agents, the MCP connector marks publishing as a destructive action in its own right. Before Claude publishes, reverts, discards a draft, deletes a form or applies a brand kit, the connector tells you what will change. Then it waits for your yes. That happens at the tool level, so it holds in auto mode too. Edits stay in a draft until you publish, and you can revert a published form.

With an API key instead of the connector, there is no such step: an agent with a key can publish. Ask it to leave forms in draft if you want to approve them yourself. The create a form with Claude guide covers both paths step by step.

Try it with dForms

Connect Claude through the dForms connector at mcp.dforms.ai, then paste this in a new chat:

Create a form in dForms called "Webinar sign-up" with name, work email,
company size (1, 2 to 10, 11 to 50, 51 to 200, 200+) and one open question:
what do you want to learn? Score company size 11 to 200 as 10 pts, 2 to 10 as 5, others as 2.
Leave it in draft and give me the preview link. Do not publish until I say so.

Claude builds the draft and returns the link. When you say publish, the connector states what will change and waits for your confirmation, Claude auto mode or not. The same approval step protects the lead qualification form or any other form your agent runs.

Frequently asked questions

What is Claude auto mode?

Claude auto mode replaces per step approval prompts with an automated check. A classifier lets safe actions through and blocks actions that look irreversible, destructive or aimed outside your environment. It became the default in Claude Code in August 2026, and Claude auto mode reaches most Claude apps the week of October 26, 2026.

Can I turn Claude auto mode off?

Yes. Anthropic says you can switch back to manual at any time, and a default you pinned yourself does not change.

Will Claude publish my dForms form without asking in auto mode?

No, not through the dForms connector. The connector marks publishing as a destructive tool, so Claude explains what will change and waits for your yes. Through an API key, an agent can publish directly, so ask it to leave forms in draft.