Hopp til hovedinnhold

Claude stops asking permission: auto mode becomes the default in the apps for Pro, Max and Team

From next week, Claude on web, desktop and mobile runs in auto mode by default on Pro, Max and Team. An automated check replaces step-by-step approval. Team owners can turn it off.

Av Joachim HøgbyKilde: Anthropic

Anthropic is changing how Claude works for paying users. Starting next week, sessions in Claude on web, desktop and mobile will run in auto mode by default on the Pro, Max and Team plans. Claude will no longer ask the user for permission before each step. An automated check is meant to block actions it judges to be risky instead.

Claude announced the change on X on Thursday 22 October at 20:04 UTC. Users can switch back to manual approval at any time, according to the announcement. Enterprise customers are not on the list.

What auto mode does

Anthropic describes the mode in the Claude Help Center, where it is called "Automatically approve", or Auto. Claude keeps working without stopping at every step. Each action is first reviewed by Claude for safety, for example to catch data exfiltration and prompt injection, meaning hidden instructions in content Claude reads. Actions judged unsafe are blocked automatically.

When an action is blocked, Claude looks for a safer way to finish the task or asks the user directly. If Claude keeps running into blocks, it switches back to asking permission for each step, according to the help page.

Anthropic says the mode applies to all existing connectors, plugins, the built-in browser, Claude in Chrome and some Cowork actions such as fetching websites. For connector tools set to "Needs approval", read-only tools are approved automatically. For tools that write or delete, Claude decides. Blocked tools stay blocked.

Default for Pro, Max and Team, not Enterprise

The help page for Team and Enterprise customers describes two settings under Organization settings > Capabilities. The first controls whether members can use the mode at all. It is on by default. The second controls whether new tasks start in auto mode. It is on by default for Pro, Max and Team, and off for Enterprise. Members can switch modes within any session.

A separate setting decides whether members can skip per-task approval for write-capable connector tools. It is off by default, according to Anthropic.

The announcement on X gives no exact date, only "next week". It says nothing specific about Europe and nothing about pricing.

Anthropic's own caveats

Anthropic itself writes that no defence is perfect and no mode replaces the user's judgement. For work with real consequences, such as money, messages sent as the user and important files, the company recommends staying close or switching back to manual approval.

The company says auto mode was tested internally and with third-party red-teaming before it became the default in Claude Code in August. At that time Anthropic said the classifier caught 89 percent of dangerous commands in testing, against 14 percent for manual approval. Those figures apply to Claude Code and Anthropic's own tests. The company has not published comparable figures for the apps.

What leaders should do now

Organisations on the Team plan get the change without doing anything. It is a real shift in who approves what. Previously an employee checked each step. Now a model decides most of them.

First, make a deliberate choice before the rollout. Team owners can turn off the setting that starts new tasks in auto mode, or switch the mode off entirely.

Second, review which connectors Claude can reach. If Claude has write access to email, CRM, finance systems or file sharing, that is where auto mode matters most. Keep per-task approval for write tools unless there is a good reason to change it.

Third, give employees a simple rule. Anthropic itself says work that moves money, sends messages or changes important files should be watched closely. Organisations can adopt that advice as their own.

hogby.ai has not tested auto mode in the apps. The description above is based on Anthropic's announcement and help pages, checked on 23 October 2026.

Sources and media

Source: Anthropic.

Illustration: AI-generated with Grok Bot / hogby.ai

Mer om Anthropic

Oppskriften. Strategien. Hver dag.

AI-nyheter for ledere. Kuratert av en CIO som bygger det selv. Daglig i innboksen.