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driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteTo keep an AI coding assistant away from sensitive code and credentials, control more than whether a provider trains on submitted data. Approve the models and product surfaces employees can use, prevent sensitive material from entering the assistant’s reachable context, keep production credentials out of its runtime, restrict its tools and network access, and review consequential changes. Treat file exclusions and provider privacy commitments as useful safeguards—not as a complete security boundary—and verify every control for the exact client, model, plan, and agent mode.
What access do you need to control?
An assistant can encounter sensitive material through more than source files. Its context may include prompts, project instructions, repository contents, build artifacts, issue text, logs, connected tools, and credentials made available to its execution environment. Its authority also depends on what it can do after receiving that context: read other files, reach external services, run commands, change code, or trigger workflows.
- Model and product access: which models and features people can use, and through which service or hosting route.
- Information access: which repositories, paths, issues, logs, and other inputs are visible to each feature.
- Credential access: which keys, tokens, and other secrets are present in prompts or the agent’s runtime.
- Action access: which tools, commands, network destinations, writes, and deployments the assistant can use.
These are separate controls. A provider’s data-handling terms do not prevent an authorized assistant from reading a file, and a file exclusion does not stop an agent from using a credential explicitly provisioned to its runtime.
How should you set the boundary?
Classify information before enabling features
Inventory sensitive repositories and paths as well as less obvious sources such as generated artifacts, issue content, and logs. Classify credential types too. For each category, decide whether it may be processed by an external hosted model, an internally hosted model, or no AI tool. Where a particular codebase must not be transmitted to a provider, design the environment so the assistant cannot read or send it; do not rely only on instructions in a prompt.
#1 Best Overall
- POWERFUL SECURITY KEY: The Security Key C NFC is the essential physical passkey for protecting your digital life from phishing attacks. It ensures only you can access your accounts.
- WORKS WITH 1000+ ACCOUNTS: Compatible with Google, Microsoft, and Apple. A single Security Key C NFC secures 100 of your favorite accounts, including email, password managers, and more.
- FAST & CONVENIENT LOGIN: Plug in your Security Key C NFC via USB-C and tap it, or tap it against your phone (NFC) to authenticate. No batteries, no internet connection, and no extra fees required.
- TRUSTED PASSKEY TECHNOLOGY: Uses the latest passkey standards (FIDO2/WebAuthn & FIDO U2F) but does not support One-Time Passwords. For complex needs, check out the YubiKey 5 Series.
- BUILT TO LAST: Made from tough, waterproof, and crush-resistant materials. Manufactured in Sweden and programmed in the USA with the highest security standards.
Inventory every way people and automation use AI
Include IDE completion and chat, edit and agent modes, CLI tools, cloud agents, web chat, MCP-connected tools, and automated workflows. A policy that works in one feature may not cover another. Record the client, model, plan, feature, and hosting route for each approved use, then test controls on the actual surfaces employees run.
Approve models deliberately
Use enterprise model defaults and enablement controls where available, and disable models or features the organization has not approved. Availability and administrative controls can differ by model, plan, and product surface, so confirm the effective settings rather than assuming a central policy covers every entry point.
Rank #2
- POWERFUL SECURITY KEY: The YubiKey 5C NFC is the most versatile physical passkey, protecting your digital life from phishing attacks. It ensures only you can access your accounts
- WORKS WITH 1000+ ACCOUNTS: Compatible with popular accounts like Google, Microsoft, and Apple. A single YubiKey 5C NFC secures 100+ of your favorite accounts, including email, password managers, and more
- FAST & CONVENIENT LOGIN: Plug in your YubiKey 5C NFC via USB and tap it, or tap it against your phone (NFC), to authenticate. No batteries, no internet connection, and no extra fees required
- MOST SECURE PASSKEY: Supports FIDO2/WebAuthn, FIDO U2F, Yubico OTP, OATH-TOTP/HOTP, Smart card (PIV), and OpenPGP. That means it’s versatile, working almost anywhere you need it
- PRIMARY & SPARE KEYS: Just like having a spare house key, we recommend buying two YubiKeys - one for daily use and one as a spare. That way you’ll never get locked out of your accounts
Can file exclusions keep sensitive code out of an assistant?
They can reduce exposure on supported surfaces, but they are not a universal boundary. GitHub documents Copilot content exclusion for specified paid organization plans; excluded files do not inform supported suggestions and responses. Its documentation also describes limitations: exclusion is unsupported in some IDE Edit and Agent modes, indirect semantic information may remain available, and symlinks or remote filesystems can present exceptions. Check GitHub’s current support information for the exact client and mode in use.
- Remove secrets from source trees instead of treating exclusions as a substitute for secret hygiene.
- Apply repository or path exclusions where supported, then test them with the specific IDE and agent features employees use.
- For code that cannot be exposed to a provider, enforce the boundary through repository permissions, environment design, or a deployment architecture that prevents access and transmission.
GitHub also warns that its cloud agent can access code and sensitive information and that information could be exposed accidentally or through malicious input. Security validation, secret scanning, internet restrictions, and review controls are mitigations, not proof that leakage is impossible.
Outdated 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 matchPC 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 & 11Rank #3
- POWERFUL SECURITY KEY: The YubiKey 5 NFC is the most versatile physical passkey, protecting your digital life from phishing attacks. It ensures only you can access your accounts
- WORKS WITH 1000+ ACCOUNTS: Compatible with popular accounts like Google, Microsoft, and Apple. A single YubiKey 5 NFC secures 100+ of your favorite accounts, including email, password managers, and more
- FAST & CONVENIENT LOGIN: Plug in your YubiKey 5 NFC via USB and tap it, or tap it against your phone (NFC), to authenticate. No batteries, no internet connection, and no extra fees required
- MOST SECURE PASSKEY: Supports FIDO2/WebAuthn, FIDO U2F, Yubico OTP, OATH-TOTP/HOTP, Smart card (PIV), and OpenPGP. That means it’s versatile, working almost anywhere you need it
- PRIMARY & SPARE KEYS: Just like having a spare house key, we recommend buying two YubiKeys - one for daily use and one as a spare. That way you’ll never get locked out of your accounts
How do you keep credentials out of agent reach?
Do not paste keys into prompts, project instructions, issues, or logs. Keep production credentials and broad-scope tokens out of agent environments by default. A credential manager or secret store does not make a value inaccessible once it has been provisioned to the agent.
GitHub documents that configured cloud-agent Agents secrets are available as environment variables during setup and task execution. Give an agent a credential only when the task requires it, and then use the narrowest permissions and repository scope available. Prefer short-lived credentials when the platform supports them, and revoke task-specific access when the work is finished.
Rank #4
- POWERFUL SECURITY KEY: The Security Key NFC is the essential physical passkey for protecting your digital life from phishing attacks. It ensures only you can access your accounts.
- WORKS WITH 1000+ ACCOUNTS: Compatible with Google, Microsoft, and Apple. A single Security Key NFC secures 100 of your favorite accounts, including email, password managers, and more.
- FAST & CONVENIENT LOGIN: Plug in your Security Key NFC via USB-A and tap it, or tap it against your phone (NFC) to authenticate. No batteries, no internet connection, and no extra fees required.
- TRUSTED PASSKEY TECHNOLOGY: Uses the latest passkey standards (FIDO2/WebAuthn & FIDO U2F) but does not support One-Time Passwords. For complex needs, check out the YubiKey 5 Series.
- BUILT TO LAST: Made from tough, waterproof, and crush-resistant materials. Manufactured in Sweden and programmed in the USA with the highest security standards.
For workflows that need a sensitive credential, GitHub’s Agentic Workflows guidance describes keeping it in a downstream job outside the agent runtime. This separates the agent’s code-generation role from the later step that uses the credential.
How should you constrain what an agent can do?
Limit runtime authority independently of what files the assistant can read. Start with read-only access where practical, isolate execution from developer home directories and production systems, and allow only the tools and outbound network paths a task needs. Put human approval in front of consequential writes, workflow execution, and deployments.
Free tools Windows power users keep installed
One-click scans. No signup required.
Best Value
- POWERFUL SECURITY KEY: The YubiKey 5 is a versatile physical passkey that protects your digital life from phishing attacks. It ensures only you can access your accounts.
- WORKS WITH 1000+ ACCOUNTS: Compatible with popular accounts like Google, Microsoft, and Apple. A single YubiKey 5 secures 100+ of your favorite accounts, including email, password managers, and more.
- FAST & CONVENIENT LOGIN: Plug in your YubiKey 5 via USB and tap it to authenticate. No batteries, no internet connection, and no extra fees required.
- MOST SECURE PASSKEY: Supports FIDO2/WebAuthn, FIDO U2F, Yubico OTP, OATH-TOTP/HOTP, Smart card (PIV), and OpenPGP. That means it’s versatile, working almost anywhere you need it.
- BUILT TO LAST: Made from tough, waterproof, and crush-resistant materials. Manufactured in Sweden and programmed in the USA with the highest security standards.
GitHub’s workflow guidance describes read-only defaults, validated write outputs, isolated execution, internet restrictions, and approval gates. These controls can lower risk, but they do not remove the need to review generated changes. For cloud-agent use, review available session logs and scan repositories and proposed changes for exposed secrets where the platform supports those controls.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How do provider privacy and retention terms fit?
Check terms for each provider, model, feature, and hosting route. Record what applies to prompts and outputs: training use, retention, abuse monitoring, logging, and any eligibility requirements for data controls. Do not transfer one provider’s commitment to a different integration or model.
- GitHub Copilot: GitHub’s documentation describes provider- and model-specific differences and exceptions. Avoid treating a general statement about Copilot data handling as an evergreen guarantee for every model or surface.
- OpenAI API: OpenAI distinguishes abuse-monitoring logs from Modified Abuse Monitoring and Zero Data Retention controls. Those controls have eligibility conditions; verify whether the particular account and endpoint qualify.
- Anthropic: Anthropic’s notice states that prompts and outputs for designated covered models are retained for 30 days from June 9, 2026, within the scope of specified arrangements. That is not a universal retention term for every Anthropic model or use.
Provider privacy terms address handling by the service; they do not decide whether your own access controls should let an assistant read a sensitive repository or use a credential.
How can you compare candidate tools or deployment patterns?
Use the same security questions for every option. The table is a comparison framework, not a claim that one product or deployment pattern is inherently safer.
| Control area | What to verify |
|---|---|
| Repository and file boundaries | Can access be restricted at repository and path level? Which clients and modes honor exclusions, and what exceptions apply? |
| Policy coverage | Does policy cover IDE, CLI, cloud-agent, web, and automation use, or only selected surfaces? |
| Credentials | Which credentials can enter prompts or the runtime? Can access be scoped to a task and repository, and are secrets needed only in a downstream step? |
| Execution and network | Is execution isolated? Which tools and outbound connections are available, and can they be limited? |
| Changes and approvals | Are outputs validated? Can writes, workflow runs, and deployments be held for human approval? |
| Data handling | What retention, training, abuse-monitoring, logging, and hosting terms apply to the exact model and route? Does a claimed data-control option require eligibility? |
What should an implementation rollout include?
- Document data boundaries. List sensitive repositories, files, artifacts, issue sources, logs, and credential classes; assign an approved processing route to each.
- Set model and feature access. Configure enterprise defaults and enable only approved models and product features. Inventory every user-facing and automated surface.
- Enforce context boundaries. Remove secrets from code, apply supported exclusions, and test the actual clients and modes. Use an architecture that denies access when exclusion behavior is insufficient for critical material.
- Minimize credentials. Keep production secrets out of prompts and agent runtimes. Provision narrowly scoped access only for tasks that need it, and revoke it afterward.
- Limit runtime authority. Isolate execution, restrict tools and egress, begin with read-only permissions, and require approval for material changes or deployments.
- Record data-handling terms. For each approved route, note provider, model, feature, hosting, retention, training use, abuse monitoring, and any Zero Data Retention qualification. Recheck when models or products change.
- Test and monitor. Verify exclusions, permissions, and network restrictions on each supported surface. Review logs where available and scan source and generated changes for exposed secrets.
Product settings, model availability, supported surfaces, and provider terms change. Revalidate the actual configuration and current vendor documentation whenever the toolchain changes and during periodic security reviews.
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.
