This article explains exactly what each Claude Code data-use account setting permits, so you can choose the privacy level that fits your work. You adjust these settings in your Anthropic account console, and they control whether your prompts, outputs, and conversation logs are used to train AI models. The three main tiers are “Don’t train,” “Training with low retention,” and “Training with standard retention,” and each has distinct rules about data storage, human review, and model improvement.
Background: Why Claude Code Has Data Settings
Claude Code is Anthropic’s command-line tool that lets developers run AI coding tasks directly in their terminal. Because it processes source code, prompts, and outputs, Anthropic gives users granular control over how that data is used. The settings exist to balance two needs: improving the AI model through real-world usage data, and protecting users’ intellectual property and privacy. Understanding these settings matters because a wrong choice could either leak sensitive code into training data or prevent you from contributing to model improvements that benefit the community.
How the Settings Work
When you enable the “Training” option in your Anthropic console, Claude Code may use your conversations (prompts and outputs) to train future models. The key differentiator is the retention period and whether a human reviewer can see your data. Here is what each setting allows:
- Don’t train: Your prompts and outputs are not used for training. They are still stored temporarily to provide the service, but they are deleted after 30 days and are never reviewed by humans.
- Training with low retention: Your data may be used for training, but it is automatically deleted after 30 days. Human reviewers may see it only to improve safety, and only if you explicitly opt in to that review.
- Training with standard retention: Your data may be used for training and is retained for up to 5 years. Human reviewers may access it for safety and model improvement, and you may be contacted if they spot a critical issue.
Note that even when training is off, Anthropic still processes your data to run the service, but it does not use it for model improvement.
How to Choose the Right Setting
Your choice should depend on the sensitivity of your code and your comfort with data sharing. Here are practical guidelines:
- If you work with proprietary code, trade secrets, or personal data, choose Don’t train to minimize exposure.
- If you want to help improve the model but are concerned about long-term storage, choose Low retention – it limits the window of exposure.
- If you are using Claude Code for personal projects or non-sensitive work, Standard retention gives Anthropic more data to improve the tool, which can benefit everyone.
You can change the setting at any time, but note that data already collected under a previous setting will be handled according to the old rule until it expires.
Common Problems and Misconceptions
A common misconception is that turning off training means your data is never stored. In reality, Anthropic must retain data temporarily to provide the service, but it is deleted within 30 days. Another issue is that some users think the setting applies to all Claude products; actually, it applies to Claude Code and the Claude app, but other products (like the API) have separate data handling policies. Also, if you use Claude Code through a third-party platform, that platform may have its own data policy that overrides your console setting.
Comparison with Other AI Coding Tools
Other AI coding assistants, such as GitHub Copilot or Amazon CodeWhisperer, have similar data-use toggles. For example, GitHub Copilot allows you to block training on your code, and CodeWhisperer lets you opt out of training. However, the retention periods and review processes differ. Anthropic’s low-retention option is relatively unique because it offers a middle ground that many tools lack. Always check the specific documentation for each tool, as policies change frequently.
Costs and Safety Implications
The data-use setting itself has no direct cost – it does not affect your subscription price. However, there are indirect costs: if you choose “Don’t train,” you lose the potential benefit of model improvements that come from aggregated usage. From a safety perspective, enabling training with human review can help Anthropic catch harmful outputs, but it also introduces a human element that some users may find uncomfortable. If you handle sensitive data, the safest choice is always “Don’t train” plus the 30-day deletion guarantee.
Alternatives and Additional Controls
If you want even more control, you can avoid sending sensitive code to the cloud entirely by using Claude Code’s local mode, if available, or by redacting sensitive parts of your code before using the tool. Additionally, Anthropic offers a data retention policy for API users that is separate from the consumer app setting. For enterprise users, there may be custom agreements that provide additional guarantees. Always review the latest privacy policy on Anthropic’s official website, as these details can change.
Frequently Asked Questions
Does turning off training delete my data immediately?
No. Anthropic retains your data for up to 30 days to provide the service, even if training is off. After that, it is permanently deleted. You cannot manually delete it earlier through the settings.
Can I use Claude Code without creating an account?
No, you need an Anthropic account to use Claude Code. The account is where you manage your data settings and billing.
Does the setting affect the quality of Claude Code’s responses?
No. The data-use setting only affects whether your data is used for training. It does not change the model’s behavior or the responses you receive.
What happens if I change my setting from “Don’t train” to “Training”?
Your future conversations may be used for training, but past conversations that were collected under “Don’t train” will still be deleted after 30 days and will not be retroactively used.
Are my code snippets visible to other users if I enable training?
No. Training data is used in aggregate to improve the model. Individual snippets are not displayed to other users, but they may be viewed by Anthropic employees and contractors for safety and quality purposes, under strict confidentiality agreements.
Related
Formula used
sel privacy | Privacy setting | Don't train=0, Low retention=1, Standard retention=2 | 0 out level | Training allowed | | privacy | 0 out storage | Storage detail | | privacy | 0
Each line is one input (in, sel, list) or one result (out: label, unit, formula, decimals). How the calculators work
Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.
More cards in this drawer
- Unbox ASMR Codes: What They Are & How to Use5.00 min/km
- Repurposed Old Motorola Phone for Homelab?
- How to Evaluate AI Vendors for Data Security, Privacy & Compliance
- Best Ergonomic Mouse: Top Picks & Buying Guide
- Captions in Google Slides: How to Add & Format Them
- Use Mac Type in Japanese? Full Guide