You can use an AI coding assistant with proprietary code, but a single “don’t train on my data” switch is not a complete privacy plan. Before opening a work repository, separately check four things: whether your content may be used to improve models, how long prompts and outputs are retained, what telemetry is collected, and which files, credentials, network destinations, and actions the assistant can reach.
The exact controls depend on the assistant, account type, subscription, IDE, agent harness, repository owner, and administrator policies. Use an employer-approved account for work, record the effective settings, and confirm them again whenever a provider changes its terms.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Alice and Bob Learn Secure Coding | $30.25 | Buy on Amazon |
| 2 |
|
The Secure Vibe Coding Handbook: A Practical Guide to Safe and Secure AI Programming | $14.99 | Buy on Amazon |
| 3 |
|
Secure Coding in C And C++ | $29.99 | Buy on Amazon |
| 4 |
|
Secure Coding: Principles and Practices | $39.98 | Buy on Amazon |
| 5 |
|
Secure Coding in C and C++ (SEI Series in Software Engineering) | $71.99 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
Start with the four controls that are easy to confuse
“Privacy” is not one setting. Treat these as separate questions and obtain a clear answer to each before sending proprietary material to an assistant.
| Control | Question to answer | What it does not prove |
|---|---|---|
| Model training or improvement | Can new prompts, code snippets, responses, or task data be used to improve a provider’s models? | An opt-out does not prove that data is deleted immediately or that it is never logged. |
| Retention and logging | How long are prompts, responses, metadata, abuse-monitoring records, or task artifacts stored? | A short retention period does not prevent the assistant from reading more files than intended. |
| Telemetry | What does the IDE, extension, CLI, or agent report about usage, errors, features, and code suggestions? | Turning off IDE telemetry does not necessarily change provider-side retention or training policy. |
| Access and execution | Which files, directories, environment variables, credentials, network services, and commands can the assistant use? | A training opt-out does not sandbox the agent or stop it from transmitting data during a task. |
Write these as four lines in your deployment record. If any line is “unknown,” treat the setup as unapproved for sensitive code until the account owner or provider documentation resolves it.
#1 Best Overall
Identify the exact deployment before changing settings
Privacy behavior can change at every layer. Record the following before you configure anything:
- The assistant and version, including whether it is an IDE extension, command-line tool, web service, API integration, or autonomous agent.
- The signed-in identity and account owner: personal, company-managed, business, enterprise, or API organization.
- The subscription or plan, because individual and organization plans can have different defaults and controls.
- The IDE, client, agent harness, and repository owner. A workspace preference is not the same as an organization policy.
- Whether requests execute locally, in a provider cloud environment, or in a mixture of both.
VS Code’s Copilot setup guidance tells users to sign in through the Copilot interface and review telemetry and code-suggestion settings. That sign-in step matters: an accidentally active personal account can put a company repository under a different policy than the organization intended.
Use an employer-approved account for work repositories
For proprietary work, follow the employer’s approved assistant, identity, and repository policy. Do not copy a private repository into a personal account merely because its controls are easier to find. An administrator may have negotiated contractual protections, configured retention exceptions, or required a particular client version that a personal account does not provide.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Ask the repository owner or security team these questions before onboarding a project:
- Which assistant and plan are approved for source code, secrets, logs, and generated artifacts?
- Must the account use single sign-on, a managed domain, or a company API organization?
- Are cloud agents allowed, or must all inference and file processing stay on managed machines?
- Which directories are in scope, and which must be excluded by policy?
- Are generated patches, command transcripts, prompts, and audit logs considered company records?
- Who can change training, retention, telemetry, network, and sandbox settings?
Keep the answers with the project’s security documentation. A developer preference is not a substitute for an enforced organization policy.
Choose the account and plan deliberately
Individual OpenAI services and Codex
For individual OpenAI services, the Help Center describes controls that let users opt out of using new conversations and Codex tasks for model improvement. In ChatGPT, the relevant control is labeled Improve the model for everyone; some account surfaces present the choice as Do not train on my content. The setting applies to new conversations or tasks after it is changed, so do not assume it retroactively changes earlier submissions.
Codex has a separate Include environments control. Review it independently from the model-improvement preference: an environment may contain files, command output, configuration, or credentials that are not present in the visible prompt. Determine exactly which environment is included and whether the task can read or modify it.
OpenAI’s Help Center states: “By default, we don’t use inputs or outputs from ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, or our API to improve our models.” This is a statement about model improvement, not a promise of zero retention, no abuse monitoring, or no administrator-visible logging. The same guidance notes that API organization owners can enable data sharing, so check the organization’s active setting rather than relying on the plan name alone.
Individual GitHub Copilot plans
GitHub documents an individual Copilot interaction-data policy scheduled to take effect on April 24, 2026, with a personal setting to disable the described training use. Because the policy is date-sensitive, open the current individual subscriber settings and confirm the setting after signing in. Do not infer that a setting for an individual subscriber applies to a company-managed organization.
Business and enterprise accounts
GitHub says Business and Enterprise customer data is protected under its Data Protection Agreement. That contractual protection is relevant, but it does not answer every operational question: check the organization’s enabled features, supported clients, retention terms, telemetry configuration, and agent permissions. An enterprise contract also does not make an unrestricted agent safe; it still needs a bounded workspace and approval rules.
| Deployment | First setting to verify | Then verify |
|---|---|---|
| Personal chat or coding account | Model-improvement or training preference | Task/environment inclusion, retention, telemetry, and file access |
| Business or enterprise service | Organization policy and data-protection terms | Client coverage, administrator overrides, retention, audit logs, and sandbox limits |
| API integration | Organization data-sharing and endpoint policy | Abuse-monitoring retention, endpoint-specific storage, credentials, and network path |
| Local IDE or CLI agent | Signed-in account and local policy | Filesystem, shell, network, environment variables, approvals, and telemetry |
Review VS Code and assistant telemetry settings
Telemetry has a different purpose from model training. It can include product usage, diagnostics, feature events, or error information even when code is excluded from model improvement. Review both the IDE and the assistant extension.
Free tools Windows power users keep installed
One-click scans. No signup required.
Check VS Code telemetry
- Open VS Code Settings and search for
telemetry.telemetryLevel. - For a local opt-out, set the value to
off. Confirm whether the setting is user-level, workspace-level, or enforced by a device policy. - Review the assistant’s own telemetry and code-suggestion settings in its Copilot Settings page; these are separate from the general VS Code setting.
- If the organization does not permit built-in AI features, set
chat.disableAIFeaturesat the required scope. A workspace setting can affect one project, while a managed policy can cover supported clients.
VS Code documentation describes telemetry as enabled for its free tier and explains that users can disable VS Code telemetry with telemetry.telemetryLevel: off. Confirm the current labels and scope in the installed version instead of copying an old screenshot or settings file.
Rank #3
Check the effective policy, not just the editable preference
In managed environments, enterprise settings, device policies, organization rules, user preferences, and workspace values can have different precedence. Use the Command Palette and run Developer: Policy Diagnostics to inspect the effective values in VS Code. Save the output required by your organization’s change record, but treat it as a configuration check rather than proof that provider-side retention is zero.
Limit what the agent can read, execute, and transmit
An assistant can expose private code without using it for training. A prompt, diagnostic bundle, shell command, or cloud task may send content outside the intended repository. Reduce the agent’s authority before asking it to inspect a sensitive project.
Use a narrow workspace
- Open only the repository or subdirectory needed for the task, not a parent folder containing multiple projects.
- Keep production credentials, signing keys, customer exports, database dumps, and private certificates outside the workspace.
- Use ignore or exclusion rules for generated files, secrets, vendor directories, and large unrelated trees. Confirm that the client honors those rules for chat context, indexing, search, and agent actions.
- Prefer a clean checkout or disposable worktree when investigating an unfamiliar change.
Sandbox the shell and filesystem
GitHub documents local sandboxing for Copilot CLI and VS Code, as well as policy control over cloud sandbox access. Where the client supports it, allow read access only to the project directory, deny writes outside an approved output path, and block access to home directories, SSH configuration, cloud credentials, and operating-system secrets.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchNetwork permissions deserve the same treatment. A local agent may need package registries or a test service, but it usually does not need unrestricted outbound access. Use an allowlist or a managed proxy, and make cloud execution an explicit policy decision rather than an accidental default.
Require approval for consequential actions
- Require confirmation before deleting files, rewriting history, changing dependency locks, applying migrations, or publishing artifacts.
- Review commands that touch production systems, customer data, cloud accounts, or authentication services.
- Use short-lived, least-privilege credentials and avoid placing long-lived secrets in environment variables visible to the agent.
- Keep command, file-change, and approval logs where the organization can audit them without collecting more source code than necessary.
An OpenAI article describing how OpenAI runs Codex safely presents sandboxing, approval boundaries, credential controls, and agent-aware logs as internal governance measures. Use those categories as a threat-model checklist, not as a claim that an identical configuration is automatically available in every Codex deployment.
Understand API retention and endpoint-specific storage
If your assistant calls an API, read the data-controls documentation for the exact endpoint and features you use. OpenAI states that API abuse-monitoring logs may contain prompts, responses, and derived metadata and are retained for up to 30 days by default, subject to exceptions when longer retention is legally required or reasonably necessary to protect the service or others.
Rank #4
- Used Book in Good Condition
Eligible customers can apply for Modified Abuse Monitoring or Zero Data Retention, but approval is required and limitations apply. Some endpoints or features may still store state even when a customer has an approved retention arrangement. Therefore, record the endpoint, feature, approval status, and exceptions rather than writing “zero retention” as a blanket label.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →| Question | Evidence to record |
|---|---|
| What is retained? | Prompts, responses, derived metadata, files, conversation state, embeddings, and abuse-monitoring records listed for the endpoint. |
| How long? | The default period, any legal or service-protection exceptions, and the organization’s approved alternative. |
| Who can access it? | Provider operations, security reviewers, organization administrators, and application operators, as defined by the current terms. |
| Does the feature keep state? | Whether threads, files, fine-tuning data, vector stores, caches, or task artifacts persist independently of abuse-monitoring logs. |
Enforce controls across a team
Recommendations are fragile when a developer can switch clients or sign in with a different account. Map each control to the administrator surface that can enforce it.
Enterprise-managed VS Code settings
VS Code describes enterprise-managed settings for supported Copilot clients and separate device policies for managed VS Code installations. Confirm which clients are covered: a policy enforced in the desktop editor may not govern a standalone CLI, web interface, alternate editor, or personal machine.
GitHub organization and enterprise policies
GitHub organization and enterprise controls govern account-level and service-side behavior, including organization policies and third-party agent controls documented for Copilot. Assign an owner for each policy, document its scope, and test a newly created account or device to ensure the rule is actually applied.
Policy precedence and exceptions
Decide what happens when a user preference conflicts with an enterprise policy. The secure default is to make the restrictive value win, prevent unauthorized overrides, and alert administrators when a client is unsupported or operating outside managed policy. Verify the resulting value in the client’s diagnostics, not only in an administration console.
Recommended Free Tools
Run a harmless verification before opening private code
A non-sensitive test can reveal which account and client are active, but it cannot prove that every privacy control works. Use it together with policy diagnostics and administrator confirmation.
- Sign out of unrelated personal accounts and sign in with the approved identity.
- Record the assistant, client, version, plan, repository owner, and execution mode.
- Confirm the model-improvement or training preference, including the date it took effect.
- Inspect any Codex environment or project-context inclusion setting.
- Check VS Code telemetry, assistant telemetry, and Developer: Policy Diagnostics where applicable.
- Open a disposable repository containing no secrets and ask for a simple explanation of a deliberately generic function.
- Observe whether the agent requests files, network access, shell commands, or credentials beyond the test directory.
- Verify the sandbox boundary by attempting a harmless access that should be denied, such as reading a file outside the approved directory.
- Confirm the API endpoint’s retention and logging terms, including any organization approval for modified monitoring or zero-retention treatment.
- Save the configuration record and obtain security-owner approval before using proprietary code.
Common failure modes and fixes
“I disabled training, so the code is private.”
Training preference, retention, telemetry, and access are different controls. Recheck all four, then inspect endpoint-specific logs and environment inclusion.
“The workspace excludes secrets, so the agent cannot see them.”
Exclusion behavior can differ between indexing, chat attachments, search, shell commands, and cloud tasks. Remove sensitive files from the accessible filesystem and enforce permissions at the sandbox or operating-system level.
“VS Code says telemetry is off, so the provider stores nothing.”
That setting concerns VS Code telemetry. It does not automatically alter assistant-service logs, API abuse monitoring, account analytics, or organization audit records.
“Our organization has an enterprise plan, so every client is covered.”
Enterprise controls usually apply only to supported clients and identities. Check coverage, block unmanaged clients where possible, and use Policy Diagnostics to inspect the effective value.
“Zero Data Retention means no state is stored anywhere.”
Approved zero-retention arrangements can have endpoint and feature limitations. Review the exact API path and stateful feature documentation, and distinguish abuse-monitoring logs from application data your own system stores.
“The agent only generated a patch.”
Generating a patch may require reading source files, configuration, dependency manifests, test output, and environment variables. Review the complete task trace and permissions, not only the final diff.
A practical approval checklist
- Identity: approved work account, correct organization, and no personal-account fallback.
- Plan: documented account type and current provider policy.
- Training: model-improvement preference verified for new tasks.
- Retention: endpoint, logs, duration, exceptions, and stateful features documented.
- Telemetry: IDE and assistant settings reviewed at the effective scope.
- Access: repository, files, secrets, shell, network, and credentials limited to the task.
- Approvals: destructive, external, production, and publication actions require confirmation.
- Enforcement: organization or device policies cover every supported client in use.
- Verification: harmless test completed and diagnostics or administrator confirmation saved.
- Change control: provider policy and client versions are rechecked before material changes or new projects.
What a defensible setup looks like
A defensible deployment uses an approved organization account, a current plan policy, an explicit model-improvement choice, documented retention for the exact endpoint, and separately reviewed telemetry. The assistant runs against a narrow workspace in a sandbox with limited filesystem and network access; secrets stay outside that boundary; consequential commands require approval; and administrators can verify the effective policy across supported clients.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsNo setting justifies an absolute promise that private code “never leaves your machine” unless the selected product’s deployment model, network behavior, logging, and policy enforcement have been specifically verified. Treat privacy as an ongoing configuration and governance task, not a one-time checkbox.
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.
