Newer agentic browsers

Are Genspark, Fellou and BrowserOS safe for work?

Each of these browsers ships an agent that reads pages and takes actions for the user. On a company laptop that agent works inside your staff's signed-in accounts, outside every browser policy you have set.

Here is what each one does, what to ask before you allow it, and how to keep it off managed devices until you have.

Genspark AI Browser

Super Agent on every page, Autopilot mode, tool connectors

Fellou

Deep Action tasks, background workspace, computer use

BrowserOS

Open source, own model keys, driven by agents over MCP

Next step

AI Agent Blocklist

Blocks agentic browsers and browser agents, including BrowserOS and Genspark's service and extension, with domains, app identifiers and extension IDs in files your firewall, DNS filter and MDM already read.

From $99 a month for one organisation

The short answer: fine for a hobby, a problem on a work laptop

The question "is Genspark AI browser safe" or "is Fellou AI browser safe" has two answers. For one person on a private machine, it is a personal choice. For a company, three things change the moment the browser runs on a managed device.

It acts as your employee

The agent clicks, types and submits in sites where the user is already signed in: mail, CRM, cloud storage, admin portals. Whatever it does is logged as that person.

Your policies do not reach it

Your Chrome and Edge settings, extension rules and data controls apply to Chrome and Edge. A second browser installed next to them starts with none of that.

Nobody approved it

These browsers are installed by users, often without admin rights. Security, legal and procurement have not seen the terms, the data flows or where the agent's tasks run.

In December 2025 Gartner published advice titled "Cybersecurity Must Block AI Browsers for Now". The browsers on this page are in the same class of product. Our page on what Gartner's advice means for IT teams covers the reasoning.

What each browser is and what its agent can reach

Facts from each maker's own descriptions at our last check. Features in this field change month to month, so read the current terms before any decision.

Genspark AI Browser

From the maker of the Genspark agent workspace

Platforms
Windows and macOS, publicly released for both in September 2025.
What the agent does
A Super Agent works on whatever page is open, for example comparing prices or summarising a video. Autopilot mode lets the browser move through sites and finish multi-step tasks on its own.
Where work runs
Some language tasks can run on open-weight models on the device. The wider Genspark workspace also offers agents that run on a cloud computer set up for each user.
Connections
A built-in store of MCP tools links the browser to outside apps such as chat, code hosting and note services.
Main domain
genspark.ai

Fellou

Agentic browser built around Deep Action

Platforms
macOS builds for Apple silicon and Intel on the maker's download page. Check the site for other systems.
What the agent does
Deep Action turns one prompt into a chain of web and desktop steps, with a plan you can review first. Deep Search researches across sites, including accounts the user is signed in to.
Where work runs
Several tasks can run at once in a background workspace while the user browses other sites. Computer use lets it operate local apps and files once permission is granted.
Memory
An agent memory feature learns from browsing history and notes.
Main domain
fellou.ai

BrowserOS

Open source browser built on Chromium

Platforms
macOS, Windows and Linux. Licensed under AGPL 3.0.
What the agent does
Runs agents in parallel tabs, using the user's signed-in accounts, with a dashboard and replays of each session.
Where work runs
The browser and its backend run on the device. Models can be the user's own keys or local models. Sessions and screenshots are stored locally, per the project.
Connections
Any MCP client, including coding agents, can connect and drive the browser. A second product, BrowserOS neo, runs beside Chrome as a browser just for agents and can import Chrome logins.
Main domain
browseros.com

Side by side: what changes for IT

The same questions for all three. "Yes" here is not a criticism of the product. It marks where a company needs a control before the browser is allowed.

QuestionGenspark AI BrowserFellouBrowserOS
Agent can click, type and submit formsYes, Autopilot modeYes, Deep ActionYes, built-in agent
Works inside signed-in accountsWorks on the pages the user has openYes, including research across logged-in sitesYes, per the project
Runs tasks while the user does something elseMulti-step tasks in AutopilotYes, background workspaceYes, parallel agents in tabs
Reaches beyond the browserTool connectors through its MCP storeLocal apps and files after permissionOther agents connect to it over MCP
Remembers past activityCheck current termsYes, agent memorySession history stored on the device
Where model calls goOn-device models for some tasks, cloud for othersCheck current termsThe user's chosen provider or a local model
Follows your Chrome or Edge policiesDo not assume soDo not assume soDo not assume so

"Check current terms" means we could not confirm the point from the maker's public pages. Ask the vendor in writing before you allow the browser.

How an agentic browser reaches company data

The path is the same for all three. Nothing here needs a bug. It is how the product is meant to work, pointed at a work account.

1

Install

A user downloads the browser. On many machines that needs no admin rights.

2

Import

Bookmarks, passwords and sign-ins come across from Chrome or Edge in one step.

3

Read

The agent reads the open page, the user's tabs and any text a site puts in front of it.

4

Act

It clicks, fills forms and sends messages as the user, in every signed-in service.

5

Connect

Tool connectors, memory and other agents extend that reach beyond the browser.

Researchers showed in 2025 that text on a web page can steer the agent in an AI browser. This applies to the category as a whole, so judge each browser by the controls you can set around it. Our page on how prompt injection works in AI browsers explains the mechanism.

Questions to ask before you allow any of them

If a team asks for one of these browsers, put these questions to the vendor and to the team. A browser that cannot answer them stays on the block list.

  • Central management. Can IT set policies for all users, such as turning off the agent, limiting sites or blocking extensions?
  • Where tasks run. On the device, on the vendor's servers or on a cloud computer? Which country?
  • What is sent to models. Page text, screenshots, form contents, files? Is any of it kept or used for training?
  • Confirmation before actions. Does the agent stop and ask before it sends, buys, deletes or posts?
  • Site limits. Can the agent be kept away from mail, finance, HR and admin consoles?
  • Memory. What does the browser remember, for how long, and can users or admins clear it?
  • Connectors. Which outside tools can be linked, and can IT restrict that list?
  • Logs. Can you see which actions the agent took, on which site, and when?
  • Accounts. Does the browser need a personal vendor account, or can it use company sign-in?
  • Removal. When a user leaves or the approval lapses, how do you remove the app, its profile and its stored sessions?

For a wider view of these risks across every AI browser, see AI browser security risks for companies.

Signs these browsers are already on your devices

Most teams find agentic browsers through inventory and DNS logs, not through a request. These are the traces to look for. For BrowserOS, our list carries every package ID, process name and domain we track.

Windows inventory

Package IDs in software inventory, for example BrowserOS.BrowserOS and, for the agents-only browser, BrowserOS.BrowserOS.Neo.

Per-user installs often sit in the user profile, so check user-level inventory, not just Program Files.

Mac inventory

Application names and processes such as BrowserOS, and installs through a package manager, for example the cask browseros.

Fellou and Genspark appear as apps in the Applications folder of the user or the machine.

DNS and proxy logs

Lookups for genspark.ai, fellou.ai or browseros.com and their service hosts, especially from machines with no approved use.

Update checks keep appearing daily once an app is installed.

Extensions in managed browsers

Some vendors also ship extensions. Genspark's Chrome extension has the ID hmdkfngkncmbgiehonfebecighedjiak.

Your Chrome or Edge extension report shows who has it installed.

If a team already depends on one of them

Blocking overnight can stop real work. When a team has built a routine around one of these browsers, move in four short steps instead.

1

Day 1: Ask

Find out which tasks the agent does for them and which accounts it touches.

2

Day 2: Revoke

Sign the browser out of mail, storage and admin tools. Reset tokens for anything it reached.

3

Day 3: Replace

Move the task to an approved tool, or to a managed browser with your policies in place.

4

Day 5: Block

Apply app rules, extension blocks and domain blocks, and tell staff why.

5

Day 30: Review

Check DNS logs and inventory for reinstalls and for newer agentic browsers.

How we block them, browser by browser

The AI Agent Blocklist holds AI tools as typed entries: domains, app identifiers, process names and extension IDs, each with a risk grade. Coverage differs per browser, so each recipe says what comes from our list and what you add yourself.

Genspark

Service, extension and developer tools

  1. From our list: the Genspark domain for DNS and the firewall, its Chrome extension ID, its VS Code extension and its command-line packages.
  2. Add the extension ID to your Chrome and Edge extension block list.
  3. For the desktop browser itself, add an app rule in your MDM using the name and signer your inventory shows. Check the current admin console for where app rules live.

Fellou

Your own rules, then the list for the rest

  1. Add fellou.ai and its subdomains to your DNS filter and firewall as your own entries, so the agent cannot sign in or reach its service.
  2. Stop the Mac app launching, for example with a restricted software rule in Jamf Pro on the process name your inventory shows. Check the current admin console for the menu.
  3. Review who already has it and remove it with the steps in our removal guide. Then load our list so staff cannot simply switch to another agentic browser.

BrowserOS

Cover both browsers and the agent link

  1. From our list: block the package IDs of both browsers, the main browser and neo, plus the process names we list for BrowserOS and its helpers, on Windows and Mac.
  2. Add the download and update domains so it cannot be fetched again.
  3. Check developer machines: BrowserOS is often installed to be driven by a coding agent over MCP. Our guide to blocking MCP servers covers that side.

Block

  • All three browsers on managed devices
  • Their download and update domains
  • Agent extensions from the same vendors
  • Agents-only browsers such as BrowserOS neo

Monitor

  • DNS lookups for the vendor domains
  • New installs in user profiles
  • Requests from teams who want an agent browser
  • New entries in our daily list updates

Allow, with conditions

  • A named test group on separate devices
  • No access to mail, finance or admin tools
  • Written answers to the questions above
  • A review date to decide again

Want a single rule for every agentic browser rather than three? Our guide to blocking AI browsers in your organisation sets that up, and the list of AI browsers shows how many there now are.

One list for the agentic browsers that come next

New agentic browsers keep appearing, and existing ones add second browsers, extensions and connectors. Our list adds them as they appear and updates every day.

Every entry carries the tool name, its category, the type of identifier and a risk grade, so you can block every agentic browser or only the higher-risk ones.

From $99 a month for one organisation

  • 2,600+ AI agent domains, including AI browsers and browser agents, as a plain list, hosts file, PAC file and firewall external list
  • 5,300+ identifiers on devices: Windows package IDs, macOS bundle IDs, process names and extension IDs
  • CSV and JSON with every column, plus DNS filter import files
  • Chrome and Edge extension block lists for Group Policy, Intune and the Google Admin console
  • Daily updates; download access set up within one business day of your order

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

We use your details only to answer you. No newsletters.

Questions people ask about these browsers

Is the Genspark AI browser safe for work?

For one person on a private laptop it is a choice like any other app. On a company device it is an unmanaged browser with an agent that can browse and act inside signed-in sites and connect to outside tools, so most IT teams block it until they have reviewed it and can control it.

Is the Fellou AI browser safe?

Fellou runs multi-step tasks across web pages and, with permission, local apps and files, and it can run several tasks in a background workspace while you browse. That is a lot of reach for a tool IT does not manage. Treat it as unapproved software on company devices unless you have reviewed it.

What is BrowserOS?

BrowserOS is an open source browser built on Chromium with a built-in AI agent. It runs on macOS, Windows and Linux, can use your own model keys or local models, and can be driven by coding agents and other tools over MCP. The project also ships BrowserOS neo, a second browser meant for AI agents.

Do Chrome policies control BrowserOS or Genspark?

Do not rely on it. Chrome policies are written for Google Chrome, and a separate browser follows them only if its maker supports that. Block these browsers as apps on the device and by domain on the network, and keep your Chrome and Edge policies for the browsers you do approve.

Can we allow the Genspark website but block the browser?

Yes. The browser is an installed app, so you can stop it with an app rule in your MDM while you decide separately about the website. Our list marks the type of each Genspark entry, so you can apply the extension and developer tool entries and leave the domain entry out.

Does blocking the domains stop a browser that is already installed?

Partly. Without its service the agent cannot sign in or call its models in the cloud, but a local model or a cached session may still work. Block the app on the device as well, then remove existing installs and revoke the sessions the browser held.

Is Eko the same thing as Fellou?

No. Eko is an open source framework from the same team for building agents in natural language or JavaScript. Fellou is the browser. Blocking the browser does not stop developers from using the framework in their own code, which is a separate decision for your engineering policy.

Genspark, Fellou, Eko and BrowserOS are trademarks of their respective owners. Chrome is a trademark of Google LLC. Edge, Intune and Windows are trademarks of Microsoft Corporation. Mac and macOS are trademarks of Apple Inc. Jamf Pro is a trademark of Jamf.

Names are used only to identify the products discussed. AI Agent Blocklist is not affiliated with, sponsored or endorsed by these companies.