Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →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
The safest way to limit an AI agent is to control four separate layers: the data and apps it can reach, the tools and actions it can use, the operations that require your approval, and the computer environment in which it runs. Start with only the connections required for one narrowly defined task, choose read-only access where possible, require confirmation for consequential changes, and isolate code or browser automation in a sandbox. These controls apply differently to ChatGPT and Gemini; Claude’s current controls must be checked in Anthropic’s documentation rather than inferred from either product.
What “limit an AI agent” actually means
An agent is more than a chatbot that generates text. It may read connected files, call an API, browse sites, operate a browser, run code, edit records or send a message. Limiting it therefore requires more than turning down a creativity setting or writing a stricter prompt.
Use four questions for every agent deployment:
- Access scope: Which apps, files, websites, accounts, credentials and data can it reach?
- Action scope: Can it only read, or can it create, edit, delete, submit, purchase or communicate?
- Approval behavior: Which operations pause for a human confirmation, and can an administrator require that pause?
- Execution boundary: Where does code or browser automation run, and what filesystem, network and operating-system resources are available?
These layers are independent. A confirmation prompt does not remove an app’s underlying access. Conversely, removing an app connection does not merely make the agent ask more often; it removes that route to the data. OpenAI explains this distinction for connected apps: permission choices govern when ChatGPT asks before using access that the connected app already has, while disconnecting the app or having a workspace administrator disable it removes the access itself. See OpenAI’s connected-app guidance.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Start with least privilege
Define the smallest useful task before you connect anything. “Summarize the invoices in this folder” is bounded; “handle everything in my email” gives an agent unnecessary opportunity to encounter sensitive material and malicious instructions. OpenAI’s safety guidance recommends specific tasks instead of broad directions such as handling everything in email. Google’s guidance likewise recommends minimum-required permissions, trusted tools and human review.
#1 Best Overall
- Powerful Turbo Fan:CAGIWIRU Electric Air Duster have reached speed up to 130000RPM,Can efficient remove dust and debris.With LED light and Three Adjustable Speed to meet with different cleaning tasks.
- With Attached USB C Fasting Charging cable.Charing 3 hours,Enjoy up to 240 minutes of use on the lowest setting, For third speed can use for 25mins.with four charging options to meet with your needs.
- Wide Application: attached 5 different nozzles,making it suitable for a variety of scenes, such as car,pet, pc, keyboards, and other electronic devices. It also serves for office and home clean duster.
- Handy Design & Portable:Cordless air duster electric nozzle is Light weight and small size, design and comfortable to hold and space saving, convenient to carry around. You can put it in your room or inside the car without taking up space.
- Economical &Reusable: Built durable materials with a long-lasting battery,our CAGIWIRU air duster it's a sustainable alternative to disposable air cans.
A practical access checklist
- Name the exact files, folders, sites or records the task needs.
- Disconnect unrelated apps and accounts before the run.
- Prefer read-only connectors and tools. Enable writing only when the task cannot be completed without it.
- Use a separate, narrowly scoped account or credential for automation instead of a personal administrator account.
- Set an expiration or calendar reminder to remove access when the project ends.
- Ask a workspace administrator to disable a connector when it should no longer be available to anyone.
Google Cloud’s agent-use guidance puts the responsibility plainly: “You are responsible for provisioning appropriate agent permissions including whether to use available tools such as Agent Identity Access Management Roles, where available.” Read the complete guidance at Google Cloud’s Agent AI Use Guidelines.
Separate read, write and high-impact permissions
Classify each tool by what it can do, not by its brand name. A calendar tool that can list events is read-only; one that can create or cancel events is a write tool. A browser that can view a product page becomes high impact when it can enter payment details or place an order.
| Permission level | Typical capability | Recommended control |
|---|---|---|
| Read | Search, list, download or summarize information | Allow only the required source and review the data returned |
| Write | Create or edit a document, record, code file or calendar event | Require a preview and explicit confirmation before committing |
| External action | Send a message, publish, submit a form, invite people or change an account | Require confirmation that names the destination, content and scope |
| High impact | Delete data, make a purchase, change permissions, use credentials or modify sensitive records | Use a human approval gate, least-privilege credentials and an audit trail; consider disallowing it entirely |
Do not assume that a product’s default confirmation covers every operation. Read the current product documentation and test a harmless action in a non-production account.
How to limit ChatGPT agent access and actions
ChatGPT’s agent features can pause for confirmation, describe high-impact-action confirmations, monitor for prompt injection and use a watch mode for certain sites. The exact availability and controls can change, so consult the current ChatGPT agent help page when configuring an account.
1. Connect only the required apps
- Open ChatGPT’s connected-app or connector settings for your account or workspace.
- Review every connected service and remove anything unrelated to the task.
- Within each service, grant the narrowest scopes available and choose read-only actions where offered.
- After the task, disconnect temporary services or ask the workspace administrator to disable them.
A permission choice that makes ChatGPT ask before using an app is not an access revocation. The connected service may still be authorized until you disconnect it or an administrator disables it. This distinction matters when a credential has been exposed or an employee leaves.
2. Use Workspace Agents controls when available
For Business and Enterprise workspaces, OpenAI’s Workspace Agents documentation describes read-only app actions, confirmations for write actions and constraints on connector actions. Workspace administrators should review which connectors are available, which actions are read-only, and whether write operations require confirmation.
Personal connections can create an additional risk: OpenAI warns that a builder’s personal connection may expose that builder’s account access to users of an agent. Do not attach a personal account to a shared agent unless the resulting access is intentional, documented and limited.
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 #2
- Screen Cleaner is Your Screen’s New Bestfriend! - EVEO is proud to present a screen cleaner that’s perfectly suited for your TV. Meet your new screen cleaner, computer screen cleaner, laptop cleaner, iPad cleaner, Macbook,tv cleaner or any other electronic device. Our screen cleaner spray and wipe along with the included plush microfiber cloth, easily removes dust, fingerprints & other smudges on the screen’s surface. Clean effectively for a streak-free screen
- Gentle on your Screens - Our screen cleaner spray and wipes is compatible to be used even for the most sensitive LCD, LED, CRT, and OLED screens in proper use. Be at ease knowing that this product is compatible with all the major brands: LG, SONY, Samsung, Sharp, iPad, iPhone, Android screens, Kindle, PC monitors - we’ve got you covered! Can be used as LCD screen cleaner, tv cleaner, smart tv screen cleaner, and more
- Premium Super Soft Plush Microfiber Cloth - Included in this efficient screen cleaning kit. This screen cleaner spray and wipe has our signature plush microfiber cloth specially designed to comply with screens and monitors and leave them streak-free. Has been shown to effectively absorbs stubborn stains, zero streaks, and no fiber shedding. Always use this together with our screen cleaner for the best results
- Easy-to-Use Screen Cleaner Kit + microfiber cleaning cloth - The EVEO Screen Cleaning Kit will easily be your new favorite tool when it comes to taking care of the things you love most, quick and easy solution for a streak-free screen. Simply screw on the sprayer gun on the screen cleaner bottle. Spray the cleaning liquid onto the microfiber cloth, then wipe the screen until desired cleanliness & shine is achieved.
- Quality Screen Care Kit Worth Purchasing - Enjoy a convenient Screen Cleaner anytime, anywhere you need it. Your satisfaction is our top priority. We stand behind the quality of our products and that’s why, you shouldn’t be worrying at all. We manufacture the Best Screen Cleaner. and can be used for laptop cleaning kit, laptop screen cleaner, tv screen cleaner for smart tv. MacBook screen cleaner, monitor cleaner or screen cleaner spray for electronic devices with microfiber cleaning cloth
3. Confirm consequential actions yourself
Before approving a ChatGPT agent action, inspect the exact operation, destination, data being shared and resulting scope. Pause or reject the run when it would:
- send or publish a message;
- edit, delete or move records;
- submit a form or create an account;
- use a password, token or other credential;
- make a purchase or financial commitment; or
- change security or account settings.
Confirmation is a decision point, not a substitute for limiting access. If an agent repeatedly requests an action outside the task, stop it and remove the relevant connector.
4. Narrow the instruction
State the allowed sources, output location and stopping conditions in one request. For example: “Read only the files in the folder named Q3 invoices, extract supplier, date and total, and create a draft table in a new document. Do not send, share, delete or edit the source files. Stop and ask before opening any other folder.” Specific boundaries reduce the chance that unrelated instructions in an email or webpage will redirect the run.
How to limit Gemini agents
Google exposes different controls for Gemini in Chrome, Gemini API computer use, managed Gemini API agents and AI Studio. Do not treat a browser feature and a developer API as the same product.
Gemini in Chrome: review browser actions
Google’s Gemini in Chrome auto-browse help says Gemini seeks confirmation before certain actions, including sending communications, modifying data, submitting forms, scheduling events and accessing sensitive financial or health sites. It asks permission before using Google Password Manager assistance for sign-in.
When a confirmation appears, verify the target site, account, fields and final text. Do not approve a request simply because the page is familiar: a compromised page can contain instructions aimed at the agent. If the task does not require sign-in, refuse password-manager assistance and use a read-only browsing task instead.
Gemini API computer use: implement safety responses
The Gemini API computer-use documentation identifies safety categories that include financial transactions, sensitive-data modification, communication tools, account creation and data modification. Your client application is responsible for implementing user confirmation when the model’s safety response requires it.
Rank #3
- 【Large size】：Measuring 800 x 300 mm/ 31. 5×12 inches, which is wider and taller than others. Large mouse pad can perfectly fit your keyboard and mouse of any size, and there is also ample space for peripherals such as phones, tablets, etc.
- 【LED backlight mouse pad】：Blade Hawks RGB mouse pad provides 7 static modes and 7 dynamic modes will give you the options you are looking for. Power-off memory function remembers the last lighting mode you selected. The super glow fiber Chroma will give you a colorful gaming setup that you want.
- 【Anti-slip rubber base】：Made from Natural rubber, this gaming mousepad will firmly grip your desktop, allowing you to enjoy the ultimate precision in the games you are most passionate about. It also comes with a texture that enhances this ability even further.
- 【Smooth and waterproof surface】: Micro-textured cloth surface, for extremely precise mouse control and a smooth movement. When liquid splashes on the gaming mouse pad, it forms water droplets and slides down. Will not delay your work or gameplay.
- 【Plug and play】: This mouse and keyboard pad is powered by USB, plug and play, no driver required. One button control light modes. Press one time to change the lighting mode, press twice to change the brightness, press and hold about 3 seconds turn ON/OFF.
- Inspect the proposed computer action and classify it against the documented safety categories.
- Show the user what will happen, where it will happen and which data will be submitted.
- Wait for an explicit approval before executing a gated action.
- Log the decision and the action result for later review.
- Terminate the run when the model requests an unapproved tool, credential or destination.
Google recommends running computer-use agents in a sandboxed virtual machine or container. Keep the environment disposable and avoid signing in with an account that has broader privileges than the task requires.
Managed Gemini API agents: restrict the sandbox and network
The Gemini API agents overview describes an OS-level isolated sandbox. It also states that outbound network access is unrestricted by default and can be narrowed with an allowlist. Therefore, isolation alone is not a complete network boundary.
- Replace the default unrestricted egress with an allowlist containing only required domains.
- Limit mounted directories and keep secrets outside the agent’s filesystem.
- Use short-lived, scoped credentials and rotate them after a sensitive run.
- Review actions and outputs before relying on the agent in a sensitive workflow.
The managed-agent feature is labeled Public Preview in the overview. Availability, limits and labels can change; verify them in the live documentation before deploying it for production work.
AI Studio agents: choose trusted tools and human review
Google’s AI Studio agent guidance recommends least-privilege credentials, trusted tools, read-only modes where possible and human review. Apply those recommendations in the tool definitions themselves: omit destructive functions, expose narrowly scoped endpoints and make the application reject calls outside an approved schema.
What can currently be said about Claude
The available official material does not establish current, product-specific Anthropic menu paths or a complete list of Claude agent permissions, approval prompts or sandbox settings. ChatGPT or Gemini instructions should not be transferred to Claude by analogy.
Recommended Free Tools
For a Claude consumer product, Claude Code or a computer-use/developer product, verify the current Anthropic documentation for:
- which connectors, files, websites and credentials are reachable;
- whether tools are read-only or can write and delete;
- which operations pause for confirmation;
- how administrators revoke access and inspect logs; and
- what filesystem, process and network isolation is provided.
Until those details are confirmed for the exact Claude product and plan, use the general controls in this guide: connect nothing unnecessary, expose read-only tools first, require an application-level approval before external effects, and run code or browser automation in an isolated environment.
Rank #4
- 【Softer and More Comfortable】Vaydeer wrist rest has unique diamond pattern, which is the combination of softness and aesthetics. The materials of wrist rest are improved into higher quality memory foam and covered with silky smooth lycra. The computer wrist rest makes you as comfortable and cushiony as like rest your wrists on clouds.
- 【Ergonomic Wrist Saver】The wrist rests for keyboard and mouse comes with a 17.32×3.15×0.83 inch keyboard wrist pad and a 5.94×3.15×0.83 inch mouse wrist support. Based on ergonomic design, the unique concave shape is the perfect fit for your wrist joints. The wrist rest pad fits most computer keyboards and laptops, improve hand and wrist posture, release your wrist and arm stress.
- 【Non-Slip Rubber Bottom】Featuring an anti-skid silicone base on the bottom, this wrist keyboard support stays firmly in place on your desk, preventing the padding from sliding around, ensuring stable and consistent wrist support during extended computer sessions.
- 【Better Experience & Pain Relief】Our keyboard arm rest is beneficial to alleviate the soreness caused by direct contact and friction between your arm and a hard desk surface, reducing the risk of wrist fatigue or carpal tunnel. The soft texture of memory foam can evenly distribute the pressure around your wrists and provide good support with just enough give.
- 【Helpful in Multiple Scenarios】Whether you're working, studying, writing, typing, gaming, this keyboard and mouse rest combo is an essential accessory to add comfort and support to your hands and wrists. It’s also a great gift for men, women, family, friend, coworker, gamer, teacher, etc.
Control the execution environment
An agent that can run code or operate a browser needs a boundary around the machine, not just a careful prompt. Use a disposable virtual machine, container or vendor-managed sandbox where appropriate. Restrict three resources separately:
Filesystem
- Mount only the working directory required for the task.
- Make source directories read-only when no edits are needed.
- Do not mount home directories, password stores, SSH keys or production backups.
- Use temporary output directories and delete them after verification.
Network
- Deny outbound traffic by default for code that does not need the internet.
- For web tasks, allowlist the specific domains and ports required.
- Block access to cloud metadata endpoints and internal administrative interfaces.
- Log DNS requests and outbound connections so unexpected destinations are visible.
Credentials and processes
- Use a service identity with only the required role and a short lifetime.
- Keep secrets in a broker or environment mechanism that does not expose them in prompts or files.
- Prevent privilege escalation and block access to host processes.
- Revoke tokens when the run ends or when an approval is denied.
Google’s managed-agent documentation provides an example of OS-level sandboxing, while its default-unrestricted-network warning shows why filesystem isolation and network policy must be reviewed independently. OpenAI’s safety guidance likewise treats sandboxing as one layer in a broader defense rather than a guarantee.
Design approvals that a person can actually check
An approval dialog is useful only when it presents enough context to make a decision. Require the agent or calling application to display:
- the exact tool and operation;
- the account and destination;
- the records or files affected;
- the text, fields or payload that will be sent;
- the expected irreversible or financial consequences; and
- an option to cancel without continuing the run.
Use two-person approval for production deployments, bulk deletion, permission changes and high-value purchases. For low-risk edits, a single reviewer may be sufficient if the change is reversible and logged. Never auto-approve all prompts merely to make an unattended workflow finish.
Defend against prompt injection
Prompt injection occurs when instructions embedded in an email, webpage, document or other external content try to redirect the agent. The content may look like ordinary text but can tell the agent to reveal data, ignore your task or take an external action. OpenAI’s safety guidance states: “It’s safer to ask your agent to do specific things, and not to give it wide latitude to potentially follow harmful instructions from elsewhere like emails.” Read Understanding prompt injections.
Reduce the attack surface
- Tell the agent which sources are data and which, if any, are authorized instructions.
- Prohibit following instructions found in documents, messages or webpages unless you explicitly approve them.
- Keep browsing and data extraction separate from sending, editing or purchasing.
- Require a fresh confirmation when the requested destination, scope or data changes.
- Stop the run if external content asks for secrets, policy changes, unusual downloads or a new connector.
These measures lower risk but cannot guarantee correct interpretation or prevent every attack. Human review remains necessary for actions with external consequences.
Free tools Windows power users keep installed
One-click scans. No signup required.
A platform-neutral setup procedure
Use this sequence whether you are configuring ChatGPT, Gemini, Claude or an agent built with an API.
Best Value
- 【No Drill Mounted】Cable management tray can be assembled in as little as 3 minutes or less-simply. Side clips easily clampe the edge of desk, pass through the holes on the sides, and secure excess cables with baskets. Desk cable management supports inward or outward installation to meet your needs in different scenarios, and provides you with great convenience when collecting and organizing wires.
- 【Rubber Pad Protects Desk】Wire organizer under desk built-in soft rubber pads on the clamp protect your desk from scratches, so you will not damage your furniture or office.
- 【Sturdy and Beautiful】Cord management under desk is made of high-quality carbon steel and holds about 15 lbs. The wire basket has 2 holes on each side of the tray to enable your cables to pass freely. Cable baskets are suitable for desk thickness use from 0.4" to 2.4".
- 【Space Saving & Keep Organized】Desk cable management tray holds all your power cords, outlet strips, USB hubs and computer transformers hidden from your feet, and the data cables will not enter the vacuum cleaner to keep your desk clean and tidy. Desk wire organizer is also suitable for use in the kitchen and outdoors.
- 【Reduce Safety Risks】Cable tray under desk keeps all plugs away from your kids, no more worry about tripping. Using 2 holes wire tray saves extra space, helps you easily cabling and wire protection. The mesh design is not easy to accumulate dust, but avoids safety accidents caused by messy wires.
- Write the task boundary. Specify the goal, permitted sources, output location, prohibited actions and stopping conditions.
- Inventory access. List every connector, file mount, website, credential and tool the agent could reach.
- Remove extras. Disconnect unrelated apps, unmount unrelated directories and delete unused tool functions.
- Choose read-only first. Run the task without write permissions and inspect the result.
- Add one write capability at a time. Define the exact records or directory it can modify.
- Insert approval gates. Require confirmation before messages, publication, form submission, purchases, deletion, credential use or sensitive-data changes.
- Isolate execution. Apply filesystem, network and process restrictions in the VM, container or vendor sandbox.
- Test with harmless data. Use a non-production account and reversible records.
- Review logs and outputs. Check tool calls, data shared, destinations and changes made.
- Revoke temporary access. Disconnect apps, expire tokens and remove test data when the work is complete.
Troubleshooting common failures
The agent can still see an app after I changed confirmation settings
Confirmation settings control prompts, not the connected app’s underlying authorization. Disconnect the app or have the workspace administrator disable it, then verify that the connector no longer appears in the agent’s available tools.
The agent keeps asking for approval
Break the task into read-only and write phases, narrow the destination and remove unnecessary tools. Repeated prompts can indicate that the request is broader than the permitted scope or that an external page is attempting to trigger an unapproved action.
A sandboxed agent reached an unexpected website
Check egress policy. In Gemini’s managed-agent environment, outbound access is unrestricted by default unless you configure an allowlist. Deny the run, review DNS and network logs, and add only domains required for the task.
A document told the agent to ignore my instructions
Treat that text as untrusted data, stop the run and do not provide the requested credential or connector. Restart with explicit instructions that external content cannot authorize actions.
I cannot find the same controls in Claude
Do not assume the controls exist under the same names. Confirm whether you are using a Claude consumer feature, Claude Code or an Anthropic developer product, then consult the current official documentation for that product’s permissions, approvals and isolation model.
Comparison by control layer
| Platform or product context | Access scope | Action and approval controls | Execution boundaries | Evidence and qualification |
|---|---|---|---|---|
| ChatGPT agent | Connected apps can be reviewed and disconnected; enable only needed apps. | May pause for confirmation and describes high-impact-action confirmations; permission prompts do not revoke app access. | Use the product’s available safeguards and apply separate sandboxing for developer workloads. | Current behavior and availability are described in OpenAI’s agent help; details can change. |
| ChatGPT Workspace Agents | Administrators control available connectors; personal connections may expose the builder’s account access. | Read-only actions and write-action confirmations are described by OpenAI. | Connector action constraints depend on workspace configuration. | See Workspace Agents help. |
| Gemini in Chrome | Browser reach depends on the signed-in profile and sites used. | Confirmation is described for communications, data changes, form submission, scheduling and sensitive sites; password-manager assistance asks permission. | Browser isolation and account permissions remain important; do not treat a confirmation as a sandbox. | See Gemini in Chrome help. |
| Gemini API computer use | Application-defined tools, sites and credentials. | Safety categories include financial transactions, sensitive-data modification, communications, account creation and data modification; the client must implement required confirmations. | Google recommends a sandboxed VM or container. | See Gemini computer-use documentation. |
| Managed Gemini API agents | Agent tools and credentials should follow least privilege. | Review actions and outputs before sensitive use. | OS-level sandbox; outbound network unrestricted by default unless narrowed with an allowlist. | Feature is labeled Public Preview in the agents overview. |
| Claude | Current product-specific controls are not established here. | Do not infer approval behavior from ChatGPT or Gemini. | Verify the exact Anthropic product’s sandbox, filesystem and network model. | Consult current official Anthropic documentation before deployment. |
Final operating checklist
- Is the task narrow enough to describe in one sentence?
- Can the agent complete it with read-only access?
- Have unrelated apps, files, websites and credentials been disconnected?
- Are send, publish, submit, purchase, delete, account-change and sensitive-data operations gated by a human?
- Are code and browser runs isolated with restricted filesystem and network access?
- Can you see exactly what data will be shared before approving an action?
- Do logs show the tools called, destinations reached, approvals given and changes made?
- Have you checked the vendor’s current documentation for the exact product, plan and date?
If any answer is no, reduce the agent’s scope before adding capability. Least privilege, explicit approvals and an isolated execution environment work together; none is a complete substitute for the others.
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.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errors
