Claude Code v2.1.278 moves auto-mode classifier server-side, removing billing overhead for API and enterprise users
Claude Code v2.1.278 runs the auto-mode safety classifier server-side by default, ending classifier overhead charges for API and enterprise users.
What changed in v2.1.278
Claude Code v2.1.278 changes where the auto-mode classifier runs. From this release, the classifier defaults to running server-side for Claude API and Enterprise users, and for sessions running on Amazon Bedrock, Google Cloud’s Agent Platform (Vertex), Microsoft Foundry, and compatible API gateways. The practical result: you are no longer billed for classifier overhead in any of those environments.
To check your installed version, run claude --version. The full changelog is published at code.claude.com/docs/en/changelog.
A quick word on what the auto-mode classifier actually does
Auto mode lets Claude Code work without prompting you to approve every action. Instead of asking “are you sure?” before each shell command or network request, a classifier runs a pre-flight safety check. If the action looks irreversible, destructive, or aimed outside your environment, auto mode stops it. Everything else proceeds without interruption.
Until v2.1.278, each of those checks was a separate model request, and that request was billed. Anthropic had already removed classifier overhead charges for Pro, Max, and Team plan users from August 2026. This release brings the same treatment to Claude API accounts, Enterprise plans, and the major cloud platforms.
The mechanism is straightforward: Claude Code now sends a safeguards field in the request body, and Anthropic’s API returns safeguard_results in the response. The check rides along with work the session was already doing, rather than firing as a separate billable call.
What this means for you
If you are an API or Enterprise customer who avoided auto mode specifically because of the classifier line on your invoice, that cost is now gone. Turn auto mode on and it will use server-side checks by default with no additional charge.
If you are on a Pro, Max, or Team plan, nothing changes for you. You were already getting free classifier overhead from August 2026.
If you run Claude Code through an LLM gateway or proxy, read the section below before assuming you are covered.
How to see which classifier your session is using
The /status command now includes an “Auto mode server” row. It reads Enabled when server-side checks are active and Disabled when Claude Code has fallen back to its own billed classifier. Check this row if you are unsure which path a session is on.
Gateway and proxy users: what you need to know
Server-side classification requires the safeguards request field to travel unchanged from Claude Code to Anthropic’s API, and safeguard_results to come back untouched. Any gateway or proxy that strips unknown body fields, rewrites tool-use IDs, or otherwise modifies the traffic will break that handshake.
When Claude Code detects this has happened, it warns you before the first action it would check:
“We’re changing auto mode to no longer charge for classifier requests in Claude Code. However, this session isn’t eligible because your requests go through [gateway address], which isn’t compatible with this update. Nothing breaks: auto mode keeps working, and its classifier requests are billed as before.”
If you see this message and there is no gateway in your path, the likely explanation is that server-side checks have not yet rolled out to your specific platform, region, or credential. Setting CLAUDE_CODE_AUTO_MODE_SERVER=0 suppresses the notice, but auto mode will continue to bill for classifier requests as it did before.
For gateway operators, the requirement is to pass request headers and body fields through unchanged, including fields the gateway does not recognise. The safeguards field specifically must be forwarded as-is.
Opting out
If you need to opt back out on Bedrock, Vertex, Foundry, or a gateway environment, set the environment variable CLAUDE_CODE_AUTO_MODE_SERVER=0. This returns the session to the previous behaviour and the associated billing.
Why Anthropic is confident in auto mode
Anthropic’s decision to roll auto mode out as a default is backed by substantial testing: internal red-teaming, third-party red-teaming, prompt-injection evaluations, and a controlled study with 1,053 paid testers. Across every measure tested, auto mode matched or outperformed manual review for catching dangerous actions. The server-side shift in v2.1.278 is an infrastructure change, not a change to how auto mode evaluates safety.
Full billing documentation is available at code.claude.com/docs/en/auto-mode-classifier-billing, and configuration options are documented at code.claude.com/docs/en/auto-mode-config.