The office is a browser tab

The laptop is not the office anymore#

I still hear remote work described as a laptop, a VPN, and a hope that the endpoint agent is installed. That was a fair picture in 2016. It is a bad picture now.

The applications left the laptop. Payroll, the CRM, the ticket queue, the board pack, the HR system, the thing finance uses instead of a shared drive: a lot of them were never installed. They are websites with a login. The person working from a kitchen table is not "on the network". They are in a tab.

Omdia, cited by Sophos, puts about 85% of the modern workday inside a web browser. Gartner said in April 2025 that fewer than 10% of organisations had deployed a secure enterprise browser, and that this would reach about 25% by 2028. Read those two numbers together. The work moved. The control mostly did not.

An enterprise browser is the unglamorous answer to that gap. Not a new perimeter with a new three-letter acronym. A browser that knows which tabs are work, and is allowed to say no.

SaaS moved the data, not just the login#

A file server had a shape you could draw. The file lived on a disk you owned. Getting it out meant crossing something: a share permission, a USB port, a mail gateway, a proxy that could see the filename. Imperfect, but the loss had a moment you could point at.

SaaS does not work like that. Once the session is open, the data is already on the screen. Copy is a local action. Paste into another tab is a local action. Upload into a second SaaS product is an HTTPS post that looks like every other HTTPS post. Nothing malicious executes. The endpoint agent has nothing to quarantine. The firewall sees encrypted traffic to a destination you probably allow, because you allow the internet.

This is why "we have DLP" so often means "we have DLP for files we can see". A payroll export copied out of the finance system and pasted into a personal Gmail draft never becomes a file on the way out. A customer list dropped into a chatbot to "tidy the columns" never hits the mail gateway. The exfiltration already happened, inside the browser, before any gateway had a chance to be clever.

The AI bit is not a future problem. The KPMG and University of Melbourne 2025 study found that 58% of employees intentionally use AI at work on a regular basis, mostly free public tools rather than whatever the company rolled out. Of the people using AI, about 48% said they had uploaded company information, financial, sales, or customer data, into a public AI tool. Almost half also said they had used AI in ways that break company policy. A ban did not tidy this up. Among employees using AI, inappropriate use was more common where generative AI had been banned (67%) than where there was no policy at all (33%). Staff route around a block. I have written about that before: a rule people cannot live with is not a rule, it is a countdown to a workaround. Same point as the security quality-of-life problem.

LayerX's 2025 browser telemetry tells the same story from the session, not the survey. They saw 77% of employees pasting data into generative AI tools, and more than 40% of files uploaded into those tools containing personal or payment-card data. Take the exact percentages as one vendor's view of its own traffic. The shape of the problem does not depend on them. If the sensitive thing is text in a tab, your file-based controls are watching the wrong door.

What "stop the exfiltration" actually means#

You cannot answer this by blocking SaaS. The business is the SaaS. The job is narrower, and more annoying to design: for each application, what is allowed to leave.

The paths that matter, in the order I actually see them argued about:

• Paste. Out of the sanctioned app, into a personal account, a chat tool, or an AI prompt. This is the one network DLP is worst at, because nothing crossed the network as a document.
• Upload and download. A CSV out of the HR system, then up into a personal drive. Or the reverse, a random file into a corporate app that then becomes the system of record.
• Screen capture and screen share. Legitimate all day on a customer call. Also a clean way to take the whole record with you.
• Print to PDF. Old, boring, still works.
• The personal login of the same product. Corporate OneDrive in one tab, personal OneDrive in the next. Same icon. Different tenant. The data does not even change product.
• Extensions. A consumer browser will run whatever the user clicked. Some of those extensions can read the page.

An enterprise browser is built to put a policy on those actions, per app, for the signed-in user. Copy out of the finance system can be blocked while copy between two approved work apps still works. Downloads can be allowed, scanned, or turned off. A screenshot of the customer record can be refused without turning off screen share for the sales call. Sophos has had to add that distinction explicitly, because a blanket "no screenshots" rule breaks the working day and then gets switched off.

That last point is the whole implementation. If the boundary is clumsier than the job, people will do the job on a phone. The control has to survive a Tuesday.

Phishing, at the moment they type#

The other half of this is not data leaving. It is the wrong person getting in, using the same tab.

The phishing that hurts a remote workforce is rarely a weird attachment. It is a page. It looks like Microsoft, or Okta, or the company VPN, and it asks for the work email and the password. Sometimes it is cruder than that. Sometimes it is an adversary-in-the-middle proxy: the user really does complete MFA, against the real identity provider, through a site the attacker is sitting in the middle of. I went through why "we have MFA" is not the end of that story in Identity risk is rising. The session cookie leaves with the attacker, and the mailbox is the payload.

Endpoint antivirus often stays quiet through all of this. Nothing ran. A page rendered, and a person typed.

The browser is the one place that can see the typing. A useful enterprise browser knows your corporate domains and your real sign-in pages. If someone starts entering a work email or a work password on a site that is not those pages, it warns them or it blocks the submission. It does not wait for the password to arrive at a server you do not own and then try to clean up.

Sophos shipped that as a specific control in October 2026. Protected Browser can watch for corporate domains and email addresses being entered on suspicious, high-risk, or low-reputation sites, and the policy can warn or block. Island describes the same class of defence as native to the browser: phishing, session hijacking, man-in-the-browser, the exploit that targets the browser itself rather than the file. Sophos also hangs DNS Protection off the browser, separate from whatever DNS the laptop happens to be using, so a known-bad domain can die before the page paints. From October 2026 that DNS policy can apply to the browser's own lookups even when the device DNS is something else.

Be precise about the limit, because the marketing will not be. This stops the phish that depends on a person typing a work credential into the wrong origin. It does not stop a phone call to the helpdesk, an MFA prompt the user approves because they are tired, or a lookalike domain that has already stolen a session some other way. A camera pointed at the monitor still works. The browser is a very good place to break the typed-password attack. It is not a theory of everything.

The situations where this earns its keep#

I am wary of "ten use cases" pages. These are the ones that come up when someone is actually trying to buy this, rather than collect a category.

The app you rent, and the data inside it. Salesforce, Workday, the finance suite, Microsoft 365 in a browser. You will not get a useful control from each vendor's own admin centre, and you will not get it from a proxy that cannot tell a paste from a page load. The browser can group those apps and apply one boundary: who can open them, what can be copied out, what can be downloaded, what can be uploaded in. Sophos calls this SaaS application control plus data boundary controls, and the newer builds log file uploads and downloads so you can see the movement instead of guessing it. Island's version of the same idea is an application boundary: data moves between approved work apps, and stops at the edge of them.

The contractor who needs one system for six weeks. Shipping a laptop, imaging it, and standing up a VDI seat for a person who needs a browser is a slow way to draw a box around Chrome. The enterprise-browser version is: they install a browser, they authenticate with your identity provider, they see the apps you provisioned, they cannot walk off with the export, and when the account dies the access dies with it. Island has built a lot of the product around this, including handing over a shared login without ever showing the contractor the password. Sophos points the same pattern at third parties, mergers, and guests. Day one of an acquisition is the version I like best. You do not have their laptops, you do not want their laptops, and you do need them in two applications by Friday.

The home PC you will never manage. BYOD is where device agents go to fail. You are not putting your full stack on a partner's machine, and you should not. A browser is a smaller ask, and if the data is not allowed to land on the disk, the unmanaged device matters less. This only works if personal browsing stays personal. Island is explicit about that split, with an on-screen indicator so the person can see when the work policy is in force. A browser that quietly records Saturday is a browser people will refuse to install. Fair.

The VDI estate that only exists to host a browser. A lot of remote-access spend is a virtual desktop whose only job is to open web apps without leaving data on the endpoint. That made sense when the alternative was a VPN and a prayer. It is an expensive way to render HTML. Gartner's note on secure enterprise browsers lists this directly: complement or reduce the reliance on VPN and VDI where the work is cloud-delivered. I would not rip out a virtual desktop that runs a fat client. I would stop buying new seats for applications that are already websites.

Shadow SaaS, and shadow AI. You cannot govern a tool you cannot see. The first month of an enterprise browser is often just the report: which apps are actually in use, which files moved, who is pasting into a chatbot. Then you write the policy. This group can use the tool. This kind of data cannot go in. Sophos is selling Workspace Protection partly on that visibility, Shadow IT and Shadow AI in the same console as the rest of the stack. The KPMG finding is the reason the visibility matters. People are already in the public tools. Pretending otherwise is how you get the 48%.

The admin console. Privileged access is a browser session too. The Azure portal, the firewall, the finance back end. Island treats this as its own case: tighter posture, authentication stepped up, the session recorded in enough detail to say who did what. Sophos ties the same idea to device health. If the device looks compromised, access to the important app can be pulled, rather than waiting for someone to notice the laptop and file a ticket.

The private app that is not SaaS at all. Some of the estate is still an internal web app, an RDP host, an SSH box. The traditional answer is a full-tunnel VPN, which puts a remote laptop on the network and then hopes. Both products put zero trust network access in the browser instead. Sophos is the more specific of the two here: ZTNA is part of the Workspace Protection bundle, the browser can open RDP and SSH itself, and the private app stays off the internet. Island's version drops the separate ZTNA agent and reaches private apps from the browser. This is not the paste problem. It is the reason a browser project sometimes replaces a remote-access project, and it is worth knowing which of the two you actually bought it for.

Sophos Workspace Protection, without the brochure#

Sophos announced Workspace Protection in January 2026, from Oxford, and started giving customers access in February. It showed up in Sophos Central in March. The pitch is deliberate: an alternative to standing up a heavy SASE stack when what you needed was control of the workspace. I think that is sometimes exactly right, and sometimes a shortcut past a network that still has a job to do. Branch offices did not vanish because Chrome got a policy engine. Buy it for the tab. Do not throw the firewall out to celebrate.

The bundle is four things, and you can take them separately:

• Sophos Protected Browser. Chromium, built with Island, or a lighter extension for the browsers people already have. The extension is the realistic rollout. The full browser is what you force for the applications that need the harder controls, including protection from attacks on the browser itself. There is an MSI, and it deploys with Intune, Group Policy, Jamf, or Google's enterprise tools. You can also email someone a setup link, which is how a contractor install actually happens.
• Data boundary controls. Copy and paste, screen capture, uploads, downloads, redaction. The October 2026 profiles are the ones that show they have met a real user: one lets you copy out of an app while blocking pastes in from outside, and one keeps the data inside the policy but still allows screen sharing and recording. File upload and download events are logged, with a report, so GenAI and unsanctioned apps are visible rather than theoretical.
• ZTNA, with RDP and SSH in the browser, and device posture in the policy. A bad health state can cost someone the app until the device is clean. Existing ZTNA licences picked up the rest of the Workspace Protection components, which is the sort of licensing detail that decides whether a trial is a project or a Tuesday afternoon.
• DNS Protection and an email monitoring add-on beside Google or Microsoft mail. DNS is the "don't even load it" layer. The mail piece is the other phishing door, the one that is not a browser.

Managed in Sophos Central, next to the endpoint and the firewall if you already have them. That is the actual reason to pick this shape rather than a browser on its own. The logs and the policy live where the rest of the Sophos estate already lives. If they do not, you are mostly buying the browser.

What Island does when it is the product#

Island is not a footnote under Sophos. Sophos Protected Browser is powered by Island. Mike Fey, Island's chief executive, is on the Sophos announcement for that reason. If you deploy Sophos, you are deploying Island's browser technology with Sophos wrapped around it: Central, ZTNA, DNS, the email add-on, the endpoint story.

If you go to Island itself, you get the browser without that bundle, and a wider surface.

The full browser runs on Windows, macOS, Linux, and Chromebook. There is an extension for Chrome, Edge, Safari, and Firefox, and they also list the newer AI browsers. Mobile is there: iOS, iPadOS, Android. That matters if "remote work" includes a phone, which it does, and a laptop-only control is going to miss the paste that happened on the train.

The security ideas are the ones above, stated as product:

• Last-mile controls on print, download, screenshot, and copy/paste, including paste that leaves the browser. They also put policy on desktop apps they name outright, Zoom, Slack, Teams, WhatsApp, which is further than "websites only".
• A boundary between work apps, so data can move inside the set and not out of it.
• Work and personal browsing kept apart, with a visible indicator. Personal sites are not the monitoring project. This is the difference between a tool people tolerate and a tool they disable.
• Anti-phishing and the other web attacks that happen in the rendering, not in a file.
• Conditional access decided in the browser: identity, device, network, location, application.
• A built-in password manager, zero knowledge, no extension to farm out. Shared credentials can be used without being revealed.
• ZTNA to private applications without a second agent.
• The activity log shipped to your SIEM. You get the session detail without decrypting everyone's traffic on a proxy and pretending that scales.

Island organises the pitch around BYOD, mergers, contractors, VDI reduction, SaaS, and privileged users. That list is the market. The interesting part, to me, is the privacy indicator and the extension. The indicator is how you avoid turning the browser into spyware. The extension is how you avoid a flag-day cutover from Chrome. Force the full browser only where the extension cannot do the job. Sophos has the same split, and it is the right one.

What I would turn on first#

If I were doing this for a company that is remote-heavy and SaaS-heavy, I would not start by replacing the browser for everyone. I would start with the tabs where a mistake is expensive.

1. Finance, HR, the CRM, and the admin consoles. One data boundary. Paste between them is fine. Paste out, and download of the export, is not, unless someone has a reason and a ticket.
2. The public AI tools, allowed on purpose for the teams that need them, with company data blocked from the prompt and the upload. A ban will lose. A boundary might not.
3. Corporate password entry, blocked everywhere except the real sign-in domains. Turn this on before you argue about watermarks. It is the phishing control, and it is the one users only notice when it saves them.
4. Contractors and anyone from an acquisition. Browser, identity, the two apps they need. Not a laptop project.
5. The logs, in a place a person looks. A paste that was blocked should be findable. A control you cannot investigate gets switched off the first time it blocks the wrong thing.

None of this retires the endpoint agent, and none of it retires identity work. A phone still photographs a screen. A helpdesk can still reset a factor for a caller who sounds rushed. Those are different doors. I would rather close the one that is currently wide open, which is the tab, than pretend a VPN profile is still the design.

References#

Sources for the figures and product details above.

• Sophos, Workspace Protection announcement, January 2026, including the Omdia figure that about 85% of the workday happens in a browser
• Gartner, 25% of organisations will use a secure enterprise browser by 2028
• KPMG and the University of Melbourne, Trust, attitudes and use of artificial intelligence, 2025
• LayerX, Browser Security Report 2025
• Sophos Protected Browser
• Sophos Protected Browser release notes, including phishing protection and data boundary profiles
• Island Enterprise Browser
• Microsoft, adversary-in-the-middle phishing and business email compromise