Every managed fleet in India keeps an asset register. Laptops, phones, servers, licences, all of it tracked and reconciled quarterly. Very few of those registers record what is installed inside the browser, even though the browser is where most work now happens and where the session tokens live. MSPs that have started auditing extension inventories on behalf of clients are returning with counts nobody had budgeted for.
The reason this slipped through is architectural rather than careless. An extension sits between the user and every web application they open, and its permissions can include reading and modifying page content across all sites. That is wider access than most EDR agents hold. It arrives through a store, installs in seconds, updates itself silently, and never touches the CMDB.
Independent privacy research is direct about what those permissions make possible. Documentation from VPNOverview research on the privacy risks of browser extensions describes embedded keyloggers capturing login credentials and financial information, browsing activity logged and resold to marketing agencies, and extensions that install malware outright. The VPNOverview team also flags a point IT managers routinely miss: an extension that was clean at install can turn later. A developer account gets compromised and pushes a malicious update, or the extension is quietly sold to another company that converts it into adware. Because extensions update automatically, neither event produces a prompt.
Both scenarios played out publicly in 2025. Security firm Koi found that Urban VPN Proxy, a Chrome extension with more than six million users and a Google “Featured” badge, was intercepting conversations with AI chat platforms and forwarding prompts, responses, and session metadata to servers the publisher controlled. Seven further extensions from the same publisher, including ad blockers and browser security tools, shared the same backend. Somewhere above eight million users across Chrome and Edge were in scope. The harvesting code arrived in version 5.5.0, through an update nobody clicked on.
What the audits keep turning up
Run an extension inventory across a client fleet for the first time and the results tend to fall into recognisable groups. There is duplicate function from unknown publishers: three PDF converters, two screenshot tools, an ad blocker nobody remembers approving. There are dormant installs from a project that ended two years ago, still holding all-sites permissions and still updating on schedule. There is personal profile sync, where staff signed into Chrome with a private Google account on a corporate laptop and pulled their home extension set into the work profile.
Then there are the AI wrappers, the summarise-this-page tools that read page content because reading page content is the entire feature, including on the CRM and the ticketing system. And in any fleet above a few hundred seats, there is usually one extension that turns out to be actively malicious.
Permission scopes, not marketing copy
Extensions are not equivalent to one another, and audits stall when teams treat them as though they are. What matters is what an extension can technically do once installed.
| Permission | What it grants | Typical enterprise exposure |
| <all_urls> host access | Read and modify content on every site the user visits | CRM records, internal dashboards, webmail bodies, AI chat prompts |
| cookies | Read and write cookies for permitted domains | Session token theft and account takeover without ever needing a password |
| webRequest / declarativeNetRequest | Observe or alter network requests leaving the browser | Traffic redirection and injection into API calls |
| tabs | See the URL and title of every open tab | Reconnaissance on internal tooling and client systems |
| storage plus a remote endpoint | Persist collected data and ship it outward | Exfiltration with no file event for the endpoint agent to catch |
| debugger | Attach to the browser’s debugging protocol | Effective control of the page |
Store review does not close this gap. The Arcanum study from Georgia Tech, presented at USENIX Security 2024, found thousands of extensions pulling sensitive data out of web pages without disclosing the collection in their privacy policies or store listings. Some of that data sat inside emails, banking pages, and medical records rather than in browsing history.
Making the audit a service line
For MSPs and integrators, none of this requires a new practice. Most UEM platforms already surface installed extensions, and the security vendors have been moving toward the browser for a while. IT Voice’s coverage of CrowdStrike positioning the endpoint as the epicentre of AI security describes an acquisition specifically to extend runtime protection into the browser, alongside shadow SaaS and AI discovery. That tells you where the vendors expect the next few years of endpoint work to happen.
A workable sequence is to pull the inventory, sort by permission scope rather than by name, allowlist the handful that survive a real review, and enforce the result through Chrome Enterprise or Edge policy instead of an email asking people to be careful. The DPDP Act is the reason this eventually becomes non-optional. An extension reading customer records off a CRM page is a personal data disclosure the client is answerable for, and “we did not know it was installed” is not a defence anyone wants to write into an incident report.
The inventory is the part that cannot be skipped. Everything else can be sequenced around client budgets, but until somebody knows what is actually running in those browsers, the rest of the endpoint programme is describing a fleet that does not exist.
