Browser Extension Security Risks for Small Business

Your endpoint security doesn't look at browser extensions, and your staff have installed dozens of them. An extension with the right permissions reads every page your team opens, including the ones behind your login.

It doesn't need to break anything to steal from you. You approved it.

The short version:

• Extensions run inside the authenticated browser session, after MFA and after the login, where endpoint tools don't look.
• Vetting once doesn't hold — LayerX found a third of extensions expanded their permissions in the past year.
• The fix is allowlisting through Chrome or Edge enterprise policy. It takes an afternoon.

What does a malicious extension actually steal?

Everything the logged-in user can see. In January 2026, researchers at OX Security found two Chrome extensions with roughly 900,000 combined installs that were exfiltrating complete conversation content from ChatGPT and DeepSeek sessions, along with every open tab URL, to a remote server every 30 minutes.

Both were AI assistant sidebars. Both asked for permission to collect "anonymous, non-identifiable analytics data."

Think about what your team pastes into an AI chat window. That's the exposure.

Why your security stack misses it

An extension is a blind spot, not a threat you'd catch. There's no malware on disk to detect.

It runs inside the browser session — after your VPN, after MFA, after the login. If it holds permission to read and change data on all sites, it sees what the logged-in user sees: client records in your CRM, email in Outlook Web, invoices in QuickBooks.

Endpoint tools treat that as normal browser activity, because it is.

How many extensions is your team running?

Nearly all of them are running at least one. LayerX's Enterprise Browser Extension Security Report 2026 — vendor research, so read it as directional — found that 99% of enterprise users have at least one extension installed and about 25% have more than ten.

Its 2025 edition put 53% of enterprise users as having installed extensions with "high" or "critical" permission scopes.

Ask your IT provider for a list of extensions installed across your fleet. If they can't produce one, that's the finding.

Vetting an extension once doesn't hold

Extensions change after you install them. The same LayerX 2026 report found that 34% of extensions increased their permissions in the past twelve months, and AI extensions were six times more likely to have done so.

Permission creep is the mechanism that matters. A note-taking extension you approved in March with read access to one domain can quietly ask for more in June.

Ownership changes too. A popular extension gets sold, the new owner pushes an update, and the whole install base gets malware in a routine version bump.

Has this hit real businesses?

Repeatedly, at scale.

• April 2025. Researcher John Tuckner of Secure Annex documented 57 Chrome extensions with about 6 million users carrying hidden tracking capability. Many were unlisted — not searchable in the Chrome Web Store, installed only by direct link, which is exactly how they'd reach a workforce through a shared message.
• July 2025. Koi Security disclosed the RedDirection campaign: 18 extensions across Chrome and Edge affecting 2.3 million users, capable of hijacking the browser.
• January 2026. The AI sidebar extensions above, shipping ChatGPT conversations to a command-and-control server.

None of these required a user to do anything wrong. They installed something from an official store that later turned on them.

Three controls, in increasing order of protection

Pick the one you'll actually maintain.

• Inventory and prune. The minimum. Chrome Enterprise Core browser reporting and the Intune discovered-apps report both give you the list; pull it, remove anything nobody can justify, and repeat quarterly. Across the SMB fleets we manage, the first pass usually removes about half of them.
• Blocklist the categories you'll never allow. VPN extensions, "free" PDF converters, screen recorders, and coupon or shopping tools have poor risk-to-value ratios in a business. Block the category, not the individual extension, so you're not playing catch-up.
• Allowlist. The real fix. Set ExtensionInstallBlocklist to * and list what you approve in ExtensionInstallAllowlist, and everything else is blocked. Both Chrome and Edge support it, and if you already manage devices with Intune you can push that policy today.

Allowlisting annoys people for a couple of weeks, then nobody notices. It's the single highest-value control on this list.

How do you decide whether to approve one?

Four questions, in order:

1. What permissions does it ask for? "Read and change all your data on all websites" is the one to argue about. Extensions scoped to a single site are a different risk class entirely.
2. Who publishes it? A named company with a real support address and a privacy policy beats an anonymous developer with 4.8 stars.
3. Does the business already have this? A lot of extension requests are people working around a tool you already pay for.
4. What happens to the data? If it processes documents or chat content off-device, that's a vendor decision, not an install decision.

AI extensions deserve extra scrutiny, and not because AI is inherently dangerous. They need broad page access to be useful, they change fast, and they route content to third-party servers by design.

That's the same pattern we covered in shadow AI data leaks and, from the other direction, in AI browsers at work.

Extension questions we get from clients

Aren't extensions in the Chrome Web Store already reviewed by Google?

They are, but review is periodic and sampled, not a live guarantee. Google reviews every submission, updates included, and says existing items are also reviewed periodically for compliance. All three incidents above — OX Security, Secure Annex, and Koi Security — involved extensions that passed that process and shipped malicious behavior anyway.

Does our antivirus or EDR catch a malicious extension?

Usually not. An extension isn't a file executing on the endpoint in a way most tools flag — it's code running inside the browser's own process, using permissions the user granted. Some browser security and managed detection tooling covers it. Standard endpoint antivirus doesn't.

Can we just tell people not to install extensions?

You can, and some will anyway. Policy without enforcement doesn't survive a deadline and a helpful-looking PDF tool. If the risk matters to you, push the browser policy through device management so the answer is technical rather than social.

What about extensions on personal devices used for work?

That's the harder case, and it's the argument for not letting unmanaged devices reach company data at all. Conditional access can require a managed, compliant device for anything sensitive. Our BYOD guide covers where to draw that line.

How often should we review installed extensions?

Quarterly if you're allowlisting, monthly if you're not. The permission-creep data is the reason: an extension that was fine when you approved it often isn't now. Once the reporting exists, the review itself takes minutes.

Does this apply to Firefox and Safari too?

The risk does. The controls are weaker. We've scoped this post to Chrome and Edge because that's what most SMB fleets run and because both have mature enterprise policy. If you have Firefox or Safari in the mix, the practical answer is usually to standardize the browser first.

Find out what's installed

Most companies have never looked. The NetSys Group will inventory the extensions running across your fleet, flag the ones holding permissions they don't need, and set up allowlisting if it makes sense for you. We do this for businesses across New York, New Jersey, Connecticut, Pennsylvania, and Southwest Florida. Get in touch for a free risk assessment.