DEV Community: aicoding-guide

Is your code used to train Anthropic's models? Claude Code data usage by plan

aicoding-guide — Fri, 11 Sep 2026 19:14:43 +0000

Originally published at https://aicoding-guide.com.

"Will the code I let Claude Code read end up training Anthropic's models?" This question comes up before any team adopts the tool. The answer is that it depends on your contract and your settings. Personal and commercial plans have different defaults, and on personal plans you have to check the setting yourself.

This article separates the concepts that get conflated, lays out the default per plan, and gives you a checklist to run before using Claude Code on work code. Always confirm the exact terms on Anthropic's own pages, since policy can change after this article's verification date.

Key point
What you will learn

How data usage differs between personal plans (Free/Pro/Max) and commercial ones (API/Team/Enterprise)

How to check and change the training opt-in

What to decide before you use Claude Code at work

Training, retention and access are three different things

These get mixed up constantly, so separate them first.

Concept	What it means
Training	Using what you send as training data for future models
Retention	Storing what you send on the server for some period, for things like abuse detection
Access	The conditions under which Anthropic staff can view content, normally limited to investigating policy violations

Opting out of training does not reduce retention to zero. Conversely, data being retained does not mean it is used for training.

Defaults by plan

Contract	Training default	Changeable
Free / Pro / Max (personal)	Follows your setting; you are asked once and can change it later	Yes, in settings
API (Anthropic Console), pay as you go	Not used for training	—
Team / Enterprise	Not used for training	—
Via Amazon Bedrock / Google Vertex AI	Governed by that cloud provider's terms; training data does not flow to Anthropic	—

Claude Code follows the contract attached to the account you are logged in as. Log in with a Pro/Max account and personal terms apply; authenticate with an API key and commercial terms apply.

Using a personal plan for work
Pro and Max are personal contracts. If you handle company code, opting out of training is not enough on its own: check whether your organization permits work data on a personal contract at all. For team use, a Team/Enterprise or API contract is the expected starting point.

Check the setting on a personal plan

Log in to claude.ai in your browser.
Open Settings and go to the Privacy section.
Find the toggle equivalent to "Help improve Claude". On means your data can be used for training; off means it is not.
You do not need to log in again in Claude Code afterwards. The setting applies at the account level.

Running /status in a Claude Code session shows the account you are logged in as and the authentication method, which tells you which contract you are actually operating under.

Retention periods

Retention depends on the training setting. Anthropic documents a long period (measured in years) when training is on, and a short one (measured in days) when it is off. The exact number of days is revised from time to time, so read the current wording in the Privacy Center rather than relying on a number quoted elsewhere.

Commercial contracts may qualify for Zero Data Retention depending on the agreement. If your industry requires it, raise it with the sales contact.

Checklist before using it at work

[ ] Which contract are you on (personal / API / Team / Enterprise)?
[ ] On a personal plan, is the training toggle off?
[ ] Does your organization permit sending code to an external AI service?
[ ] Do you have settings that keep secrets (credentials, customer data) out of Claude's reach? See Stop AI coding tools from reading your API keys and secrets
[ ] Do you know what data your MCP servers forward to external services?
[ ] If you go through Bedrock or Vertex AI, have you checked that provider's data terms?

What Claude Code actually sends

Claude Code reads the files it needs for the instruction you gave and sends their contents to the API. It does not ship your whole repository on every turn, but anything it reads with grep or cat is sent. Files you never want read should be blocked explicitly with a deny rule. See Claude Code permissions in settings.json for how to write them.

Usage telemetry is sent separately from conversation content, and the environment variable that disables it is documented in the official docs.

Glossary
Telemetry: statistics such as feature usage counts and error occurrences. It is handled separately from code content, and there is a setting to turn it off.

How Codex and Gemini CLI compare

Codex (OpenAI) and Gemini CLI (Google) follow the same shape: personal and commercial tiers have different defaults. Gemini CLI's free tier, signed in with a personal Google account, carries a notice that input may be used to improve the product. Check each tool on its own terms rather than assuming one policy covers all three.

Summary

Training, retention and access are separate; the setting mainly controls training
API, Team and Enterprise are not used for training by default; Pro/Max follows your setting
On a personal plan, check the toggle in claude.ai settings, and follow your company's policy for work code
What gets sent is the content of files Claude reads, so block what you don't want read with deny rules
Policy wording changes, so verify against Anthropic's Privacy Center rather than a secondhand summary

Does # add to CLAUDE.md in Claude Code? What the current docs say

aicoding-guide — Fri, 11 Sep 2026 04:03:08 +0000

Originally published at https://aicoding-guide.com.

You saw a tip somewhere that typing # at the start of the Claude Code prompt appends what you type to CLAUDE.md. You tried it, and nothing happened. This article explains what the official documentation actually says.

Here is the short answer. As of September 11, 2026, the official documentation contains no description of # as a special input. The documented ways to add to CLAUDE.md during a session are the /memory command, asking Claude in plain language, and auto memory.

Key point
What you will learn

How # is treated in the current docs, and which pages were checked

The three documented ways to add content to CLAUDE.md

Why "remember this" does not end up in CLAUDE.md

The docs list five input prefixes, and # is not one of them

The interactive mode reference has a "Quick commands" table listing characters that carry special meaning in the prompt. These are the five entries.

Input	Meaning	What it does
`/` (at start)	Command or skill	Opens the list of built-in commands and skills
`!` (at start)	Shell mode	Runs a command directly and adds its output to the session
`@`	File path mention	Triggers file path autocomplete and pulls the file into context
`:`	Emoji shortcode	Inserts an emoji from a `:name:` shortcode
`?` (on empty input)	Help panel	Toggles the keyboard shortcut help panel

There is no # row. The commands reference, the memory page, and the best practices page were checked as well, and none of them describes # as an append shortcut.

Whether it ever existed is not something we can confirm
Whether # worked in an earlier version, or whether the tip confuses Claude Code with another product, could not be confirmed in the official documentation. This article sticks to the verifiable fact that the current docs do not mention it. If it happens to work in your installation, treat that as behavior specific to your version.

Three documented ways to add to CLAUDE.md

1. Open the file with /memory

Running /memory in a session lists your memory file locations, including CLAUDE.md and CLAUDE.local.md, and includes entries for files that do not exist yet. Selecting a row opens it in your editor, creating it first if needed.

/memory

GUI editors such as VS Code open the file in a separate window, so you can keep using the session while it is open. Terminal editors such as Vim take over the terminal until you exit.

2. Ask Claude in plain language

This takes the fewest keystrokes. The documentation says that to add instructions to CLAUDE.md, you should ask Claude directly, like "add this to CLAUDE.md", or edit the file yourself via /memory.

Add the rule we just agreed on to CLAUDE.md: run tests with pnpm test, not npm test.

CLAUDE.md is an ordinary Markdown file, so Claude appends to it with its edit tool. Name the section you want it under if placement matters to you.

3. Let auto memory handle it

Auto memory is the mechanism by which Claude writes notes for itself. It is on by default, and files live in ~/.claude/projects/<project>/memory/, with a MEMORY.md index plus one file per topic.

Why "remember this" never lands in CLAUDE.md

This is the part that trips people up. When you ask Claude to remember something, such as "always use pnpm, not npm", that note goes into auto memory, not CLAUDE.md. That is why the file looks unchanged afterwards.

The two systems divide up as follows.

	CLAUDE.md	Auto memory
Who writes it	You	Claude
What it holds	Instructions and rules	Learnings and patterns
Scope	Project, user, or org	Per repository, machine-local
Loaded into	Every session	Every session (first 200 lines or 25KB of `MEMORY.md`)

Rules your team needs go in CLAUDE.md and get committed to version control. Auto memory is machine-local and is not shared across machines or cloud environments, so a shared convention parked there only ever helps you. See Turn off Claude Code auto memory or move it for the settings that control it.

Check that what you added is actually loaded

Appending to a file does not guarantee the session loads it. The command that shows you is /context, not /memory.

/context

Look for the Memory files list in the output. If your file is not there, Claude cannot see it. To check whether your file is in a location that gets loaded, see How to write CLAUDE.md: the three levels and a ready-to-use template. Files in subdirectories load on a different schedule, covered in When does Claude Code load a CLAUDE.md in a subdirectory?.

If you have no CLAUDE.md yet, /init generates a first draft from your project structure. See What Claude Code's /init puts in CLAUDE.md, and what to change afterwards.

Summary

The current official documentation does not describe # as a shortcut for appending to CLAUDE.md
The documented input prefixes are /, !, @, :, and ?
To append mid-session, open the file with /memory or ask Claude to "add this to CLAUDE.md"
Asking Claude to "remember this" writes to auto memory, which stays machine-local
Confirm the result with /context and check the Memory files list