# Auto Mode in Claude Code: Free Security Checks and the “this session isn't eligible” Notice > Starting with version 2.1.278, Claude Code does not charge for auto mode security checks on the server. What “this session isn't eligible” means and how to check the status. Source: https://ailoot.ai/blog/en/claude-code-auto-mode-classifier/ Canonical: https://ailoot.ai/blog/en/claude-code-auto-mode-classifier/ Language: en Publisher: Ailoot (https://ailoot.ai/) Translation: guide-042 Revision: sha256-16692f35b61176b192461a9dbb3d2a3ef11815432d6220ebaf4760e14be12b7a If the notice “We're changing auto mode to no longer charge for classifier requests in Claude Code. However, this session isn't eligible” appears in the Claude Code terminal (in the Russian wording: “We are changing auto mode so classifier requests in Claude Code are no longer charged. However, this session is not eligible”), it is not an error. Starting with version 2.1.278, Claude Code asks the server to perform auto mode safety checks by default and does not charge for them. But this only works where server-side checks reach the session. When Claude Code switches to its own classifier requests, it warns you, and checks are charged as before. [Ailoot documentation: API, keys and connection setup](https://ailoot.ai/docs/). ## How does Claude Code auto mode check actions? In [auto mode](https://code.claude.com/docs/en/permission-modes), a classifier checks actions before they run: for example, shell commands and network requests. Claude Code's own classifier requests are billed as ordinary token usage, just as they were before version 2.1.278. From version 2.1.278, Claude Code asks the **server** to perform checks as part of the requests the session is already sending to the model by default. Checks performed by the server are not charged. The same version added an `Auto mode server` line to the `/status` command: it shows whether the classifier for this session is running on the server. Who this affects: - **Enterprise plans and Claude API accounts** — server-side checks are requested by default. - **Claude Platform on AWS, Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry** — also by default, but the availability of server-side checks depends on whether the feature is enabled on the particular platform and in the region. - **Pro, Max, Team** — according to Anthropic's documentation, the notice never appears on these plans. A separate restriction, unrelated to billing: on Bedrock, Google Cloud's Agent Platform, Foundry and signed-in Claude apps gateway sessions, auto mode only works with Claude Sonnet 5, Opus 4.7 and newer models, and with the Fable family. The permission mode is one of the settings that determines how the agent behaves in your project: what it does on its own and what requires confirmation. If you regularly build projects in Claude Code, check that your permission settings match the project's tasks. ## Why server-side checks do not reach the session According to the documentation, the most common cause is an LLM gateway or proxy between Claude Code and the API. The problem occurs if this intermediary removes or rewrites request headers, drops fields it does not recognize, or changes responses: for example, by rewriting IDs or removing keys from streaming events. The server then does not receive the request for a check, or Claude Code does not receive its result, and the session switches to Claude Code's own classifier, whose requests are charged as before. If Claude Code can identify the gateway, the notice names it in the text. The second cause is that server-side checks are not yet available for the session's platform, region or credentials. If there is no gateway or proxy in the chain but the notice appears repeatedly, the documentation identifies this as the likely cause. Auto mode continues to work. The notice appears only when server-side checks stop reaching the session for the rest of its duration, and this can happen on the very first action that is checked. An individual action that the server cannot check does not trigger the notice: Claude Code checks it itself and contacts the server again on the next request. ## What to press when you see the notice The notice pauses the action until you respond: - **Enter** — continue. The paused action and the rest of the session use Claude Code's own classifier, and its requests continue to be charged as token usage. After Enter, the notice does not appear again in this session. If it names a gateway, acknowledging it hides the notice on this machine for 24 hours. If no gateway is named, the notice returns when the next session switches to its own classifier again. - **Esc or Ctrl+C** — cancel. The paused action is not performed, the current turn stops, and the session remains in auto mode. Your response is not remembered, so the notice appears again before the next action that needs to be checked. If you want to leave auto mode, switch the permission mode with `Shift+Tab` after responding. Where the notice cannot wait for a response, Claude Code reports the same text and continues in auto mode. The exception is if a gateway notice has already been acknowledged on this machine in the last 24 hours: the text is not shown. In non-interactive mode with the `-p` flag, the text is written to the error stream (stderr); in `stream-json` output, it arrives as a system warning that applications using the Agent SDK can read from the message stream. In the VS Code extension, the message appears in a dialog and requires no acknowledgment. ## How to check the status You can check where the classifier is currently running without waiting for the notice. Run `/status` in a session using auto mode and look at the **Auto mode server** line: - `Enabled` — the server is still checking the session's actions; - `Disabled` — the session has already switched to Claude Code's own classifier. ## What to do if the gateway is the cause The documentation describes two configuration options. 1. **Ask your company administrator or gateway provider to pass requests and responses through unchanged.** Headers and request body fields, including those the gateway does not recognize (such as the `safeguards` field), must be passed through as they are. Responses and streaming events must be returned without removing keys such as `safeguard_results` or rewriting tool call IDs. See the [gateway compatibility guide](https://code.claude.com/docs/en/llm-gateway-protocol#feature-pass-through) for details. After this, new sessions use server-side checks again. 2. **Disable requests for server-side checks if you already know the gateway will not support them.** Set the variable before starting the session, either in the shell or through the `env` key in Claude Code settings: ```bash export CLAUDE_CODE_AUTO_MODE_SERVER=0 ``` Claude Code then always uses its own classifier, its requests are charged as before, and the notice does not appear. This variable is not read when connecting directly to the Anthropic API. If `CLAUDE_CODE_AUTO_MODE_SERVER` is not set but `CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1` is set, server-side checks are also disabled. The documentation describes `CLAUDE_CODE_AUTO_MODE_SERVER` as a temporary setting that may be removed in a future release, so it is worth checking it again after updates. If you do not own the gateway and cannot configure it, you can contact the gateway provider or choose the second option. ## If there is no gateway but the notice appears When there is no gateway or proxy in the chain but the notice appears again, the likely cause, according to the documentation, is that server-side checks are not yet available for your platform, region or credentials. You can confirm this with support or your company administrator, or report the problem with the `/feedback` command. ## What next You do not need to configure anything in advance: when the notice appears, auto mode continues to work, and the classifier's own requests are charged as before. If the notice appears, respond to it (Enter or Esc) and check `/status`. If necessary, you can configure the gateway, disable requests for server-side checks, or contact support. Read about other causes of token usage in [“Where tokens go in Claude Code”](https://ailoot.ai/blog/en/where-claude-code-tokens-go/) and [“How to avoid burning through Claude Code limits”](https://ailoot.ai/blog/en/claude-code-token-limits-saving/). And here is how to reduce unnecessary code in model responses and save tokens that way.
### Sources - [Claude Code: the GitHub CHANGELOG.md commit containing the version 2.1.278 entry](https://github.com/anthropics/claude-code/commit/bf7d404e26a5fb6167d21b46c93a2bf6c22ab274) — Anthropic's official repository - [Auto mode classifier request charges](https://code.claude.com/docs/en/auto-mode-classifier-billing) — Anthropic's official documentation on the notice and charges for checks
If you need access to AI models through Ailoot, take a look at [Ailoot](https://ailoot.ai/). ## Questions about Auto Mode in Claude Code: Free Security Checks and the “this session isn't eligible” Notice ### Does pressing Esc disable auto mode? No. Esc or Ctrl+C cancels the paused action and stops the current turn; the session remains in auto mode. Change the permission mode with Shift+Tab after responding if you want to leave auto mode. ### How do I distinguish server checks from paid classifier requests? Run /status in an auto-mode session and inspect Auto mode server. Enabled means server checks are still active; Disabled means the session has switched to Claude Code’s own classifier requests, billed as token usage. ## Ailoot · The company and its project portfolio Ailoot combines established operating experience, a large client base, stable service and ongoing KPI control. The company and its project portfolio: millions of customers; monthly sales in the billions. Ailoot brings that operating experience to API access for AI features in customer products, agents, applications and automations.