iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Sometimes. Whether an AI coding assistant may use repository-related data to improve models depends on the product, plan, account settings, feature, and route used. “Private repository” does not mean “never processed”: an assistant may need to process prompts and code context to answer while excluding the repository’s stored contents from training. Check the rules for the exact account and feature you plan to use, and distinguish model training from temporary processing and retention. The vendor policies discussed here were reviewed on October 25, 2026; they can change.
Can an AI coding assistant train on code from a private repository?
It can, depending on the service and circumstances. The important distinction is between code stored in a repository and the prompts, snippets, surrounding context, outputs, or feedback created when someone actively uses an assistant. A vendor may say it does not train on private repository content at rest while still processing code from that repository to provide assistance; interaction data may have separate rules.
GitHub makes that distinction explicit: it says Copilot processes private-repository code during active use, while its announced training program excludes private repository content at rest. But GitHub says that, beginning April 24, 2026, interaction data from Copilot Free, Pro, and Pro+ may be used to train and improve models unless the individual opts out. The defined interaction data includes inputs, outputs, code snippets, and associated context. GitHub says the change does not apply to Business and Enterprise plans or users who opt out. GitHub’s policy announcement explains the distinction and scope.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteThat means a prompt written while working in a private repository can be treated differently from the repository’s stored files. Do not treat a statement about “private repository content” as an answer about every prompt or code fragment produced during a coding session.
#1 Best Overall
What do “processing,” “retention,” and “training” mean?
These terms describe different stages of data handling. A privacy answer is incomplete unless it tells you which stage and which data it covers.
- Processing: The assistant or model receives information needed to respond, such as a prompt and relevant code context. Processing is part of providing an AI feature; it does not by itself establish that the data is used for training.
- Retention: Data is kept after a request, for some period or purpose. A retention term may apply to prompts and outputs, engagement or telemetry data, safety review, or deleted chats—and each can have a different period.
- Model improvement or training: A vendor may use eligible interactions to improve or train models, subject to its policy and settings. A training opt-out is not automatically a promise that no data is processed or retained for other stated purposes.
- Deletion: Removing a conversation from a history view and removing data from backend systems or training pipelines are not necessarily the same event. Read the applicable deletion terms rather than assuming one action does all three.
For example, Cursor says prompts and code context are sent to model providers when its AI features are used, while its documentation says Privacy Mode prevents code from being used for training by Cursor or those providers. That is a training restriction, not a claim that the feature runs without sending code context. Cursor’s privacy documentation describes the mode and its scope.
How do the policies differ by assistant and account?
The same vendor can apply different rules to an individual subscription, an organization workspace, an API route, or a particular model. Use this comparison as a starting point, then verify the current terms for the configuration you will actually use.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Rank #2
- Easy to read text
- It can be a gift option
- This product will be an excellent pick for you
| Service and account context | Training and data-use statement | Retention or control details |
|---|---|---|
| GitHub Copilot Free, Pro, and Pro+ | From April 24, 2026, GitHub says interaction data can be used to train and improve models unless the user opts out. The announced policy excludes Business and Enterprise users who opt out, and says private repository content at rest is not used for training. | Users can change the AI-training preference in Copilot settings under Privacy. GitHub says a previous opt-out of product-improvement data collection is retained. Policy announcement. |
| GitHub Copilot Business and Enterprise | GitHub says the announced individual-plan training change does not affect these plans; enterprise data is governed by its Data Protection Agreement. | GitHub’s product page says IDE chat and completion prompts and suggestions are not retained by default. For other Copilot access and use, it says prompts and suggestions are retained for 28 days; engagement data is retained for two years. These are stated defaults, not proof of an organization’s actual configuration. Copilot product page; GitHub changelog. |
| Claude consumer plans, including Claude Code used under those accounts | Anthropic says it may use chats and coding sessions for model improvement when the user permits it, when conversations are flagged for safety review, or when the user explicitly opts in. Its documentation treats commercial products separately. | Deleted consumer chats are removed from history immediately and from backend systems within 30 days. If model improvement is allowed, de-identified data may be retained in training pipelines for up to five years. Turning the setting off stops future training use, but does not remove material already in training runs or trained models; deleting a chat excludes it from future training. Training documentation; retention documentation. |
| Cursor | Cursor says Privacy Mode ensures code is not used for training by Cursor or model providers; it also says prompts and code context are sent to model providers when AI features are used. | Privacy Mode is enabled by default for teams, and an administrator can enforce it organization-wide. Verify the mode in the relevant account or team dashboard. Cursor privacy documentation. |
The periods in this table are not interchangeable service-wide retention windows. They refer to different data types, product surfaces, triggers, and account contexts.
How long do coding assistants keep prompts and code?
There is no single retention period for “your code.” Ask what data the period covers, which access surface it applies to, and whether it is a default, a contractual commitment, or a model-specific rule.
GitHub Copilot: access surface matters
For Business and Enterprise, GitHub’s product documentation distinguishes IDE chat and completions from other access and use. It says the former are not retained by default, while the latter are retained for 28 days; engagement data is retained for two years. These are vendor-stated defaults, not confirmation of an individual organization’s settings. Check the organization’s configuration and agreement before treating them as its actual terms. GitHub Copilot product documentation.
Rank #3
Claude consumer chats: deletion and improvement are separate
Anthropic says deleted consumer chats disappear from history immediately and from backend systems within 30 days. Separately, when model improvement is allowed, it may keep de-identified data in training pipelines for up to five years. The documentation says turning off model improvement stops future training use, but does not remove material already in training runs or trained models; deleting a chat excludes it from future training. These are consumer-account terms and should not be applied automatically to API or commercial-workspace use. Anthropic’s retention documentation.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Hosted models and special retention arrangements
Routing can introduce another policy layer. GitHub’s enterprise model-hosting page describes provider-specific arrangements: it says Anthropic has a zero-data-retention agreement for generally available Anthropic features, while noting exceptions for designated models or preview features. It also describes Google Cloud service terms as prohibiting training on GitHub data. Model availability and exceptions can change, so check the live page for the selected model and feature rather than treating these statements as universal. GitHub’s model-hosting documentation.
Anthropic’s documentation for designated “Covered Models” describes a 30-day retention requirement for prompts and outputs, including specified zero-data-retention arrangements. It does not apply uniformly to all models or to consumer Claude and Claude Code use; workspace and cloud-specific controls for Claude Code also matter. Do not infer a 30-day period for a model or deployment unless the stated coverage applies to it. Anthropic’s Covered Models documentation.
How can I check or opt out before using a private repository?
Start with the identity and exact feature that will make the request, not just the name of the coding assistant. An individual account used inside a company repository may not inherit the protections of an organization-managed plan.
- Identify the account and plan. Record whether the user will sign in with an individual subscription, a business or enterprise workspace, an API account, or a cloud-provider route. Note whether the account is personally purchased or organization-managed.
- Identify the data boundary. Check the policy for stored repository files separately from prompts, retrieved snippets, surrounding context, generated suggestions, feedback, and other interaction data. If a setting speaks only about repository content at rest, do not read it as a blanket promise about active sessions.
- Check the training control. For Copilot individual plans, GitHub identifies the AI-training preference under Copilot settings, then Privacy. For Claude consumer plans, consult the model-improvement setting and its effect on future use. For Cursor, verify Privacy Mode in the account or team dashboard; an administrator can enforce it for the organization.
- Check retention by feature. Determine whether the documented period applies to IDE chat, completions, web or other access, deleted conversations, engagement data, or a specific model. Confirm whether a “no retention” statement has named exceptions.
- Check the route and model. Establish whether the assistant uses the vendor’s hosted model, an API, or a cloud provider, and identify the selected model and whether it is generally available or a preview. Provider-specific terms may differ.
- Confirm administrative and contractual terms. Ask the administrator to verify enforcement, user permissions, the applicable data-processing terms, regional commitments, and subprocessors for the selected setup. A public general product page does not establish the exact commitments in an employer’s signed agreement.
- Test a low-risk setup first. Have an administrator verify the settings in a non-sensitive repository, capture the product, account type, model, and feature configuration, then review the matching documentation and contract before enabling access to sensitive code.
If an expected control is missing, do not assume it is enabled just because the service offers it. Confirm the signed-in identity, plan, organization enforcement, and feature being used. If the vendor documentation does not resolve a model or route exception, treat that as an unresolved configuration question and ask the account administrator or vendor before sending sensitive code.
What should an organization verify before approving an assistant?
A security review should document the complete path from developer to model, not stop at a product-level “private” or “enterprise” label. The practical record should let the organization reproduce the decision if a plan, model, or policy changes.
Best Value
- Made in USA - Proudly produced in Ohio by a Veteran-owned business
- Comprehensive Coverage: This BookFactory log book includes essential fields such as post/shift, time of change, date, weather conditions, and a designated space for detailed notes. This ensures that all relevant information is captured and easily accessible.
- Sturdy Cover: The trans-lux cover protects the log book from wear and tear, ensuring its longevity and maintaining the integrity of your recorded data.
- Essential Security Tool: This log book is an indispensable tool for any organization that values security and accountability. It helps to prevent misunderstandings, improve communication, and ensure a smooth transition between shifts.
- Wire-O with Trans-lux cover, 100 Pages, Dimensions 8.5" x 11" - (Security-Pass-Down) Reorder SKU: LOG-100-7CW-PP(Security-Pass-Down)
- Plan and identity: Which user identity accesses the repository, and is it covered by an organization-managed plan?
- Data categories: Which content is processed, retained, used for improvement, or excluded from those purposes?
- Feature and model: Which chat, completion, agent, or other feature is enabled, and which model serves it?
- Retention and deletion: What periods apply to prompts and outputs, telemetry, feedback, safety review, and deletion? Which periods are defaults versus contractual requirements?
- Controls: Is the relevant opt-out or privacy mode enabled, can an administrator enforce it, and can an end user change it?
- Route and provider: Does a third-party model provider or cloud service receive the context, and what terms apply to that route?
- Contract and geography: Which agreement and regional commitments govern this use, and do they match the selected configuration?
Policies reviewed on October 25, 2026 are snapshots of vendor documentation, not a substitute for the organization’s current settings or contract. Recheck the live terms when a plan, model, region, or feature changes.
Frequently Asked Questions
Does opting out undo training that has already happened?
Not necessarily. Anthropic says turning off model improvement stops future training use, but does not remove material already in training runs or trained models. The GitHub announcement describes the opt-out scope but does not establish that already completed training can be reversed, so do not assume it can.
Does deleting a consumer Claude chat remove it from every place immediately?
Anthropic says deletion removes the chat from history immediately and backend systems within 30 days. Its separate model-improvement documentation says deletion excludes the chat from future training, which is different from removing material already present in training runs or trained models.
Does a zero-data-retention label mean the model never receives my code?
No. Retention and processing are different questions: an assistant may send code context to a model to generate a response even where an applicable arrangement limits retention. Check the selected model’s actual coverage and exceptions rather than interpreting the label as local-only processing.
Do consumer, API, and company accounts follow the same rules?
Not automatically. Anthropic’s consumer Claude and Claude Code documentation treats commercial products separately, and organization contracts or API routes can have different terms. Confirm the account type and route that will actually handle the repository context.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
