Claude Code Made Auto Mode the Default on August 14, and Now on Every Plan

Auto mode is the default permission experience in Claude Code. It became the built-in default for new Pro, Max, and Team sessions on August 14, 2026, and with Claude Code v2.1.283 or later, Anthropic's permission-modes reference lists auto as the built-in starting mode for interactive terminal and VS Code sessions on every plan and provider, including Enterprise, Console API keys, and the cloud integrations. A pinned or administrator-managed default still wins. (Claude by Anthropic; ClaudeDevs on X)

Auto mode is not a blanket permission bypass. Claude Code routes each tool call through a separate classifier that evaluates whether the action is irreversible, destructive, or aimed outside the user's environment. When the classifier blocks something, Claude can usually look for a safer path or ask the user to approve the action directly. (Claude by Anthropic)

What Changed by Plan

If you already pinned a default permission mode, Anthropic says nothing changes. A user who configured a different default may see a one-time prompt, while a Team administrator's managed default takes precedence. Users without a configured default will see an in-product notice as the new behavior arrives. (Claude by Anthropic; ClaudeDevs on X)

The announcement said that "in the coming month, working with our cloud partners, we plan to make it the default across all of these and no longer charge for classifier overhead." The default half landed in late September. Anthropic's changelog records v2.1.283 (September 25) moving interactive sessions on third-party providers or with telemetry off to auto mode, and v2.1.284 (September 28) moving interactive terminal and VS Code sessions "on every plan and provider"; the permission-modes reference dates the every-plan default to v2.1.283 or later. The billing half has not landed in the docs: on Enterprise plans and on Claude API, Claude Platform on AWS, Bedrock, Agent Platform, and Foundry accounts, "classifier calls count toward your token usage," except where the server reviews actions as part of the session's model requests.

Who Starts in Auto Mode Now

Anthropic's permission-modes reference, checked October 9, 2026, decides the starting mode in this order: the --permission-mode flag, then permissions.defaultMode in a settings file, then the built-in default below, where the first matching row applies.

Your first session after an install or upgrade can start in a different mode than the table gives, because Claude Code may choose before its feature flags arrive.

Two further rules from the same page. If the built-in default selects auto but auto mode is unavailable to the session (a settings file turned it off, the model does not support it, or Anthropic has temporarily disabled it server-side), "Claude Code starts the session in Manual instead." And the first time the built-in default starts one of your sessions in auto mode, Claude Code shows a notice: "In a terminal, once, at the top of the session" and "In the VS Code extension, as a card on the new-conversation screen that stays until you dismiss it."

Availability is wider than the default. The docs list auto mode's requirements as "Plan: All plans"; on Team and Enterprise "auto mode is available by default" until an administrator sets permissions.disableAutoMode to "disable" in managed settings; and by provider it is "available by default on the Anthropic API, Claude Platform on AWS, Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry, and signed-in Claude apps gateway sessions." The model floor differs by provider: Opus 4.6 or later, Sonnet 4.6 or later, Haiku 5.5, or a Fable model on the Anthropic API and Claude Platform on AWS; only Sonnet 5 or later, Opus 4.7 or later, Haiku 5.5, and the Fable models on Bedrock, Agent Platform, Foundry, and gateway sessions.

How Auto Mode Makes Permission Decisions

The classifier runs on every tool call. It is intended to catch dangerous actions without forcing the user to review every routine read, edit, or test command. Anthropic says the classifier is designed to block data exfiltration, destructive operations, and actions that move work outside the user's environment. (Claude by Anthropic)

The mode has explicit fallback behavior. Claude Code switches back to manual approvals after three blocks in a row or twenty blocks across a session, according to Anthropic's write-up. Permission rules still run before the classifier, except for broad allow rules that would grant arbitrary code execution, such as python:*; those broad rules are set aside while auto mode is active so they cannot bypass the classifier. (Claude by Anthropic)

Auto mode also screens content returned from external sources for prompt-injection attempts. That safeguard is not a guarantee: Anthropic says the system relies on classification and does not eliminate risk, so high-stakes production changes still deserve direct review. (Claude by Anthropic)

The Classifier Is No Longer Charged on Three Plans

The classifier uses a small number of extra tokens per tool call. As of August 7, 2026, Anthropic says Claude Code users on Pro, Max, and Team plans are no longer charged for that classifier overhead. That is narrower than a promise of higher quotas: the announcement discusses classifier-token treatment, not a general increase to included usage or rate limits. (Claude by Anthropic; ClaudeDevs on X)

In August, Anthropic said it planned to stop charging classifier overhead on Enterprise and the listed API and cloud-platform surfaces when auto mode became the default there, and that Enterprise administrators would be notified first. The default has since arrived on those surfaces, but the permission-modes reference still says classifier calls count toward token usage there, except where the classifier review runs server-side. (Claude by Anthropic)

What Administrators Can Control

Organizations do not have to accept one global behavior without controls. Anthropic documents a managed defaultMode setting for pinning the default, and disableAutoMode for turning auto mode off entirely. Administrators can also add hard-deny rules for actions that should never be approved automatically. (Claude by Anthropic)

The announcement describes a few examples of those boundaries: auto mode should deny attempts to send code or secrets to an external destination, inspect the destination of a git push or pull request, and account for the current repository state before a destructive command such as git reset --hard. Those checks are there to preserve context that a simple allowlist cannot see. (Claude by Anthropic)

How to Switch Back to Manual Review

In the CLI, press Shift+Tab to change modes. In the desktop app, use the mode dropdown. A pinned personal or managed setting may control the starting state, but Anthropic's documentation still describes mode switching as available to the user. (Claude by Anthropic)

For a cautious rollout, start auto mode on a repository where the test and deployment boundaries are clear. Keep hard-deny rules for production and data-export operations, and inspect the first few long-running sessions rather than treating the new default as an unattended approval to run anything.

What Shipped After August 14

The Pro, Max, and Team rollout ran as described. For people on Pro, Max, or Team who had pinned a different default in ~/.claude/settings.json with no other settings file overriding it, the docs now say "Claude Code asks once, in the terminal or in the VS Code extension, whether to change the setting to auto mode. If you decline, your setting stays as it is." Four post-launch changelog entries are worth knowing if a session did not behave as this article predicted:

The change is behavioral rather than a new model or a new command: a fresh interactive Claude Code session, now on any plan, continues through routine tool calls without pausing for a prompt. The classifier becomes the review layer, manual mode remains available, and existing pinned or administrator-managed defaults continue to win. (Claude by Anthropic)

The broader implication is that Anthropic is making longer-running Claude Code work the default workflow while keeping a safety boundary around destructive and out-of-environment actions. Enterprise and API teams that relied on manual mode being the starting state should now decide which repositories are appropriate for auto mode and verify that managed settings express the exceptions they care about.

Bottom Line

Claude Code auto mode became the default for new Pro, Max, and Team sessions on August 14, 2026, and with v2.1.283 or later it is the built-in starting mode for interactive sessions on every plan and provider. It uses a per-tool-call classifier instead of asking for every permission decision, falls back to manual review after repeated blocks, and does not charge Pro, Max, or Team for classifier overhead. The practical choice for teams is not simply “on or off”: set managed defaults and hard denies for sensitive work, then use auto mode where fewer interruptions are worth the trade-off. (Claude by Anthropic)

Corrections and Updates

October 9, 2026: This page said Enterprise plans, Console API keys, Bedrock, Agent Platform, Foundry, and Claude Platform on AWS still start in manual mode; with Claude Code v2.1.283 or later, interactive sessions there start in auto mode, and from v2.1.285 so do claude -p and Agent SDK sessions that don't fetch feature flags.

September 6, 2026: the switch happened on schedule, and the docs now record it as the built-in default.

Sources

• Claude by Anthropic, “Auto mode is now the default in Claude Code for Pro, Max, and Team plans” (August 7, 2026): https://claude.com/blog/auto-mode-default-in-claude-code (re-checked September 6, 2026; the body still reads "Starting on August 14, new sessions on Pro, Max, and Team plans will run in auto mode" and "For Enterprise users and users who access Claude Code via the Claude API, auto mode remains opt-in for now")
• Anthropic, Claude Code docs, "Choose a permission mode": https://code.claude.com/docs/en/permission-modes (checked October 9, 2026; source for the built-in default table, the Manual fallback when auto mode is unavailable, the first-session notice, the one-time switch prompt, the availability requirements by plan, organization, provider, and model, and the classifier billing note for Enterprise, API, and cloud accounts)
• Anthropic, Claude Code release digest, "Week 32, August 3-7, 2026": https://code.claude.com/docs/en/whats-new/2026-w32 (checked September 6, 2026; source for "Starting August 14, auto mode is the default permission mode for new sessions on Pro, Max, and Team plans"; the Week 33 and Week 34 digests, checked the same day, carry no auto mode entry)
• Anthropic, Claude Code changelog: https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md (checked September 6, 2026; source for the 2.1.236, 2.1.243, and 2.1.251 entries in the post-launch table)
• Anthropic, Claude Code docs changelog: https://code.claude.com/docs/en/changelog (checked October 9, 2026; source for the 2.1.283, 2.1.284, and 2.1.285 entries and their dates)
• Verification status of the three ClaudeDevs posts cited below, September 6, 2026: x.com returns a JavaScript shell without the post text to non-browser clients, so the posts could not be re-read. Nothing on this page rests on them alone; every claim they support is also in the Anthropic blog post or the docs above.
• ClaudeDevs on X, “Starting August 14, auto mode will be the default permission mode in Claude Code” (August 7, 2026): https://x.com/ClaudeDevs/status/2085794862608318627
• ClaudeDevs on X, “Auto mode runs a classifier on every tool call” (August 7, 2026): https://x.com/ClaudeDevs/status/2085794868102889473