Was Claude Code overcharging you to monitor you? That changed
Claude Code 2.1.278 makes the auto mode safety classifier free for some users. See who it applies to and how to assess it.
Introduction
The change in Claude Code auto mode matters because it affects more than a technical setting: it can change how security and monitoring costs appear on an AI bill. Anthropic released Claude Code 2.1.278, and the auto mode safety classifier, which reviews commands before they run, now runs free on the server for pay-per-token users.
What changed
This update does not introduce a new feature. It changes who pays for an existing safety check. The classifier still reviews commands before execution, but the server-side processing is now free for users of the Claude API and Enterprise accounts, as well as platforms such as Bedrock, Vertex and Foundry.
Why it matters for business
In enterprise settings, small and hidden charges can become meaningful once they are spread across teams, workflows and usage volume. If a company uses automation, agents or command-driven processes that require pre-execution review, any safety-related charge can affect the total cost of ownership. The practical value of this change is accounting clarity: it helps separate core usage costs from security-control costs.
Where it applies and where it doesn't
Applies to
It applies to Claude API users and Enterprise accounts. It also applies in environments using Bedrock, Vertex or Foundry, as long as the request path supports server-side checks.
No change for
If you are on Pro, Max or Team, you never saw this charge in the first place. And if your traffic goes through a gateway or proxy that does not support server-side checks, the source says you still pay as before, with no warning. In other words, routing architecture can determine whether the billing change is visible at all.
Limitations and risks
The announcement does not say every deployment will benefit automatically. The main risk is assuming a pricing improvement applies universally when it actually depends on the plan, the consumption channel and the request path. Another risk is missing indirect charges if your organization relies on intermediaries that do not support the server-side verification.
Evaluation checklist
- Identify which plans different teams are using.
- Check whether usage comes through the Claude API, Enterprise, or platforms like Bedrock, Vertex and Foundry.
- Verify whether a gateway or proxy sits in the request path.
- Separate core usage charges from security or monitoring charges on the bill.
- Confirm server-side check support before expecting savings.
How to apply it in your business
Use Claude Code auto mode as a prompt to audit how your AI consumption is routed and billed. If your company pays per token, review whether security controls are being charged where they should not be. If intermediaries are involved, validate compatibility before concluding that the change has already reduced your total cost.