Is Manus AI safe? What companies should know
Manus is a general-purpose AI agent that plans and carries out whole tasks: research, files, websites and, with its browser extension, work inside your staff's logged-in sessions. The question for a company is not whether it is clever. It is what it can reach, and on whose terms.
Manus at a glance
| Type | General-purpose autonomous agent |
| Launched | 6 March 2025, invitation-only beta |
| Business plan | Team plan with SSO, roles and audit logs |
| Minimum age | 18, under its platform terms |
| Apps | Web, macOS, Windows, iOS, Android |
| Browser extension | Manus Browser Operator for Chrome and Edge |
| Main domain | manus.im |
Next step
AI Agent Blocklist: keep Manus and other autonomous agents off unmanaged devices
The Manus extension, desktop app identifiers and domains, plus every other general agent, in formats for browser policy, MDM, DNS and firewalls.
From $99 a month for one organisation
How Manus works, and what each mode can reach
Manus takes a goal, breaks it into steps and works through them with its own tools. Where it runs decides what it can touch. Three modes matter for a security review.
In the cloud
Tasks run on Manus's side, including an isolated cloud browser. They keep running when the user's laptop is closed.
Reaches: what users upload or connectWith connectors
Users can connect accounts and services so the agent can read and act in them as part of a task.
Reaches: connected mail, files, appsIn your browser
Browser Operator lets Manus work in the user's own Chrome or Edge, in a dedicated tab, with the logins already there.
Reaches: every signed-in sessionThe mode that changes the risk most is Browser Operator. A cloud agent sees what you give it. An agent in the local browser starts from everything the user is already logged in to, on your network and from your IP address.
What staff ask Manus to do, and where the risk sits
Most requests are harmless research. The risk climbs when a task needs company files, a connected account or the user's own browser. Use this table to talk to teams about their actual use.
| Typical task | What the agent needs | Risk for the company |
|---|---|---|
| Market or competitor research from public sources | The public web | Low, as long as the prompt itself holds no confidential plans |
| A report or slide deck from internal documents | Uploaded company files | Confidential files leave your systems for the vendor's |
| Lead lists and outreach | A connected mailbox or CRM | Messages sent in the company's name, personal data processed |
| Booking, ordering or form filling on websites | Browser Operator with signed-in accounts | Purchases and submissions made with real accounts |
| Building and publishing a website or app | Hosting on the vendor's side | Company content published where you did not plan it |
| Recurring reports on a schedule | Stored access to sources | Standing access that outlives the project |
What Manus can do with company data
Every task moves some company data to the vendor's side. Follow one task from start to finish and decide at each step what your staff may put in and what may come out.
1. The prompt
Goals and instructions often hold plans, names and figures. Treat a prompt like an email to an outside supplier.
2. Uploaded files
Documents, sheets and decks are copied into the task. Decide which kinds of data may never be uploaded.
3. Connected accounts
A connector lets the agent read and act in mail, files or apps. Each one widens what a single task can reach.
4. Browser Operator
Pages open in signed-in sessions become task input. What the user can see in that tab, the agent can read.
5. Cloud work
The agent works through its steps in its own cloud environment and browser, and keeps going when the laptop is closed.
6. Outputs and replays
Results, files and task replays stay in the workspace and can be shared. Check who can open a shared link.
7. Published sites
Websites and apps the agent builds can be hosted on the vendor's side. Decide who may publish, and under which name.
8. Scheduled tasks
Recurring tasks keep stored access to their sources. Review them when people change roles or leave.
Why it matters even if you never approve Manus: staff may already use personal accounts with company files. An inventory of who uses it, and with which account, is the first step either way.
Which data may go into an agent task
Give staff one simple table. It works for Manus and for any other autonomous agent you review.
| Kind of data | Personal account on a work device | Approved Team workspace |
|---|---|---|
| Public information and web research | Not on company devices | Allowed |
| Internal documents and plans | Not allowed | Allowed after review of the data terms |
| Customer or employee personal data | Not allowed | Only with a signed data processing agreement and a named person responsible |
| Contracts, prices and financial results | Not allowed | Only for named roles, with a person checking every output |
| Passwords, keys and payment details | Never | Never |
| Student records | Never | Never |
Is Manus AI safe? Six questions that decide it for your company
There is no single yes or no. These questions turn "is it safe?" into answers your security, privacy and legal teams can sign off.
Which account are people using?
A Team workspace you control, or personal accounts? Personal accounts get none of your SSO, roles or audit logs.
Is Browser Operator installed?
If yes, the agent can act in signed-in work sessions. Decide whether that is ever acceptable on company devices.
What is connected?
Every connected mail, file or app account widens what one task can read or change.
Which data terms apply?
Team and individual terms differ. Read the current privacy policy and data processing terms for training, retention and transfers.
Who reviews the work?
Autonomous runs can publish, send or change things before anyone looks. Decide which outputs need a person to check.
Who is allowed at all?
Manus's platform terms set a minimum age of 18. Schools should treat it as off-limits for students.
The general risks of autonomous agents, with examples, are on our page about AI agent security risks.
Manus Team plan: the controls Manus describes
If a team wants Manus, a managed workspace is the only version you can govern. Manus lists these controls for its Team plan; confirm the current scope with the vendor and in the admin console.
| Control | What Manus says it offers | Your question to the vendor |
|---|---|---|
| Sign-in and provisioning | SSO and SCIM | Can SSO be enforced so personal sign-ins with work email are refused? |
| Roles and sharing | Roles and sharing controls | Can admins stop sharing of tasks and files outside the workspace? |
| Workspace-only access | A Team Only setting that keeps supported tasks, replays, files and internal sites inside the workspace | Which items are not covered? |
| Audit | Audit logs | Can logs be exported to your SIEM, and how long are they kept? |
| Training | Model providers are barred from training on Team customer data | Is this in the contract, and does it cover every provider? |
| Certifications | SOC 2 Type II, ISO 27001 and ISO 27701 | Can we see the current reports and certificates? |
Keep the browser part separate. Team controls govern the workspace. They do not decide whether the Browser Operator extension may run on your devices. That is a browser policy decision, made by you.
Manus Browser Operator: the part that touches your network
Uses local logins
Manus documents that the extension works with the user's existing logged-in sessions in their desktop browser, in a dedicated tab.
Per-session consent
The user authorises each session and can stop it by closing the tab or take over by clicking into it. Manus says actions are logged.
Started from elsewhere
Tasks can be started and watched from mobile while the desktop browser stays online, so the office browser can work while the user is away.
For a security team this means an autonomous agent working inside authenticated work sessions, from a company IP address, possibly while nobody sits at the desk. Few roles need that. Block it by default and allow by exception.
Manus also documents limits of its own: drag and drop and long multi-step forms may not work perfectly, and sites with strong anti-bot measures may need a person to step in.
Expect users to take over tabs halfway through a task. That blurs the line between what the person did and what the agent did.
That blur matters for audits. If an order or a message goes wrong, your logs will show the user's account, not the agent. Decide in advance who is accountable for work an agent started.
Signs Manus is already in use
Before you decide, find out who uses it today. Each source below shows a different piece, so check them together.
| Where to look | What to search for | What it tells you |
|---|---|---|
| Google Admin apps and extensions usage report | The Browser Operator extension ID | Managed Chrome profiles where the agent can act in signed-in tabs |
| Defender browser extensions inventory | The same ID on Windows devices | Devices, users and whether the extension is turned on |
| Intune discovered apps and Mac MDM inventory | The Manus desktop app and its bundle ID | Laptops with the desktop client installed |
| DNS filter or resolver logs | The Manus domain | Who uses the web app from your networks, including personal devices |
| Identity provider consent logs | Grants to Manus for mail, files or calendars | Work accounts connected to cloud tasks |
If you approve a Team workspace: set it up before anyone works in it
Sign-in
Connect SSO and SCIM so joiners and leavers follow your directory. Ask whether SSO can be enforced for every member.
Sharing
Turn on workspace-only access where available, and restrict public sharing of tasks, replays and files.
Connectors
Start with none. Add one service at a time after review, with the narrowest account that does the job.
Logs
Confirm what the audit log records and how you will export it. Schedule a monthly look.
Tell staff plainly: "Use Manus only through the company workspace, never with a personal account, and do not install the browser extension unless IT has approved it for your role." Short rules get followed.
Vendor review checklist for Manus
Use the same checklist for any autonomous agent vendor. Our AI agent risk assessment turns the answers into a decision.
- The legal entity you contract with and the governing law of the contract
- Where data is processed and stored, and how transfers are covered
- List of subprocessors, including model providers
- Training and service-improvement use of content, for Team and individual accounts
- Retention for tasks, files, replays and connected-account data
- Enforced SSO, SCIM and the ability to block personal sign-ups with work email
- Audit log contents, export and retention
- Which connectors can be disabled or limited by admins
- Whether Browser Operator can be switched off for the workspace
- Deletion on exit, and how quickly it is confirmed
- Current certifications and the latest reports
- How changes to terms, subprocessors and data locations are notified to customers
How we solve it: block Manus where it is not approved
If you have not approved Manus, or only approve a Team workspace, block the pieces that run on your devices and network. Our list carries these identifiers along with those of every other autonomous agent; every entry is checked before it enters the list, and the lists update every day.
| Piece | Identifier from our list | Where to block it |
|---|---|---|
| Browser Operator extension | cecngibhkljoiafhjfmcgbmikfogdiko | Chrome extension blocklist in Google Admin, Intune or Group Policy; Edge blocklist where Chrome Web Store installs are allowed |
| Desktop app for macOS | im.manus.desktop, Homebrew cask manus | Jamf, Kandji or Mosyle app restrictions |
| Desktop app for Windows | Microsoft Store ID 9PHV7M7V4S5L | Intune app rules, Microsoft Store restrictions |
| Web app and cloud tasks | manus.im | DNS filter, firewall, secure web gateway |
Steps for a mixed Windows and Mac fleet
- Browser first. Add the extension ID to your Chrome blocklist. In the Google Admin console use Devices > Chrome > Apps & extensions > Users & browsers and set the installation policy to Block. Guides: Google Admin and Intune.
- Then the apps. Restrict the macOS bundle ID in your Mac MDM and block the Store app in Intune. Our Intune guide for AI tools covers the Windows side.
- Then the network. Add the domain to your DNS filter or firewall where Manus is not approved at all. See blocking AI with DNS.
- Then the category. Manus is one general agent among many. Load the whole agent category so the next one does not take its place; see blocking AI browser extensions and blocking Genspark, Fellou and BrowserOS.
What a network block does not stop: tasks already running in Manus's cloud continue on their own. Blocking stops new sessions from your network and devices. Revoke connected accounts if you need running tasks to lose access.
Three sensible positions on Manus
Approved Team workspace
- Team plan with enforced SSO
- Connectors limited to reviewed services
- Browser Operator blocked except for named users
- Personal accounts blocked
Short pilot
- One small group, non-sensitive data only
- No connections to mail or HR systems
- Review after 30 days with the checklist above
Not approved
- Extension, apps and domain blocked
- Users told why and how to request
- Weekly check for new installs
Whichever you choose, write it into your inventory and policy. Unapproved installs found later belong in your shadow AI agent review. In schools, see students and AI agents.
Block every general-purpose agent, not just one
- 2,600+ AI agent domains, including general autonomous agents and agent platforms
- 5,300+ device identifiers: extension IDs, bundle IDs, Windows package IDs and process names
- Risk grade and category on every entry
- Updated every day; download access set up within one business day of your order
What the files contain: the AI Agent Blocklist. Broader approach: how to block AI agents.
Ask us or order a plan
Tell us what you need to block and where your rules live. We reply within one business day.
- Files for firewalls, DNS filters, secure web gateways, MDM and browser policies
- Updated every day as new AI browsers, agents and extensions appear
- Plans for one organisation, for MSPs and for product makers, or a hosted filter we run for you
Manus AI safety: questions companies ask
Is Manus AI safe for company use?
It depends on how it is used. Manus works with the accounts and browsers you connect to it. A Team plan with SSO, roles and audit logs can be reviewed like any SaaS vendor. Personal accounts on work devices sit outside your controls, so most organisations block them.
What data does Manus see during a task?
It sees what the user gives the task: the prompt, uploaded files, the content of connected accounts and, with Browser Operator, the pages open in the user's signed-in browser. Results, files and replays are then stored in the user's workspace, where they can be shared.
What should we check in the Manus privacy policy?
Read the current privacy policy and data processing terms before you approve it. Check where data is processed and stored, which subprocessors and model providers are involved, how long tasks and files are kept, and how content is used to improve the service.
Does Manus train on our data?
For Team customers, Manus states that its model providers are barred from training on Team customer data. For individual accounts, read the current privacy policy on how content is used to improve the service.
What is Manus Browser Operator?
A Chrome and Edge extension that lets Manus act in the user's local browser with existing logins, in a dedicated tab. Users authorise each session and can stop it by closing the tab.
How do we block Manus?
Block the extension ID cecngibhkljoiafhjfmcgbmikfogdiko in browser policy, the desktop apps by im.manus.desktop and Store ID 9PHV7M7V4S5L, and add manus.im to DNS or firewall rules where it is not approved.
Do Manus tasks keep running after we block it?
Tasks that already run in the vendor's cloud can continue, because a network block only stops new sessions from your devices and networks. To cut off running tasks, revoke the connected accounts and tokens in your identity provider and in the services the user connected.
Can students use Manus?
Manus's platform terms say users must be at least 18. Schools that block AI agents on student devices usually include it in the agent category of their block lists.
Manus is a trademark of its owner. Google, Chrome, Android and Google Admin are trademarks of Google LLC.
Microsoft, Windows, Edge, Intune and Microsoft Store are trademarks of Microsoft Corporation. Apple, macOS and iOS are trademarks of Apple Inc.
Genspark, Fellou, BrowserOS, Jamf, Kandji and Mosyle are trademarks of their respective owners. Names are used only to identify the products discussed. AI Agent Blocklist is not affiliated with, sponsored or endorsed by these companies.