The AI Threat Vector Hiding in Plain Sight
Every enterprise I talk to has some kind of strategy for managing AI adoption. They're thinking about SaaS platforms, API integrations, data governance around LLMs. Good. But almost none of them are thinking about browser extensions.
A new report from LayerX lays out some numbers that should make any CTO or CISO uncomfortable. AI browser extensions are quietly becoming one of the most dangerous, and most ignored, threat surfaces in enterprise environments. They bypass your DLP. They don't show up in your SaaS logs. And they have direct access to everything your employees see, type, and stay logged into.
Let me walk through what the data actually says and what I think it means for how we govern AI in practice.
This Is Not a Niche Problem
The first instinct is to dismiss browser extensions as a fringe concern. Something a handful of power users install. That instinct is wrong.
According to the report, 99% of enterprise users have at least one browser extension installed. More than a quarter have over 10. This is not a long-tail issue. It's universal.
And roughly 1 in 6 enterprise users already has at least one AI-specific extension. That number is growing fast.
But here's the real problem: most organizations can't answer basic questions about this. Which extensions are running? Who installed them? What permissions did they grant? What data can those extensions read?
We've spent years building visibility into networks, endpoints, and identity systems. Browser extensions are the gap nobody closed.
AI Extensions Are Measurably Riskier
It would be tempting to assume that AI extensions carry the same risk profile as any other extension. The data from LayerX says otherwise. Compared to the average browser extension, AI extensions are:
60% more likely to have a known vulnerability (CVE)
3x more likely to have access to cookies, which means potential exposure of session tokens
2.5x more likely to have scripting permissions, enabling data extraction and manipulation
2x more likely to be able to manipulate browser tabs, which can facilitate phishing or silent redirection
6x more likely to have escalated their permissions in the past year
These aren't abstract risks. Cookie access exposes active sessions. Scripting permissions let extensions read and modify page content. Tab manipulation enables redirection attacks. And all of it installs in seconds, often without IT ever knowing.
AI extensions create an ungoverned layer of AI usage that bypasses your existing visibility and policy enforcement. Your DLP, your CASB, your SaaS security posture management tools: none of them see what's happening inside these extensions.
Extensions Are Not "Install and Forget" Safe
I see a lot of security teams treat extensions like static artifacts. Approve them once, put them on an allowlist, move on. But that's not how extensions work in practice.
Extensions evolve. They get updates, change ownership, and expand permissions over time. The LayerX report found that AI extensions are nearly 6x more likely to change their permissions compared to other extensions. Over 60% of users have at least one AI extension that modified its permissions in the past year.
This is a moving target. An extension that was safe when you approved it in January might be reading session cookies by March. Static allowlists can't keep up with that reality.
The Trust Signals Are Weak
How do you evaluate whether an extension is trustworthy? The typical signals are publisher transparency, install count, update frequency, and the presence of a privacy policy. But for AI extensions, these signals are often missing or unreliable.
Some numbers worth noting:
33% of AI extensions have fewer than 5,000 users
Nearly 50% of AI extensions have fewer than 10,000 users
About 40% of all extensions haven't been updated in over a year, suggesting they may be abandoned or unmaintained
A tiny user base means less community scrutiny. No updates means unpatched vulnerabilities sitting in your environment. And abandoned extensions can be acquired by new (potentially malicious) owners without anyone noticing.
Think about the scrutiny you apply to a SaaS vendor before procurement. Now compare that to the zero scrutiny most AI browser extensions receive. The gap is enormous.
What This Means for Your Organization
Here's my take on this. If you're a CTO or VP of Engineering, this is a real architecture and governance problem. Take it seriously.
Most organizations have invested heavily in controlling how AI enters the environment through APIs, SaaS integrations, and managed tools. But browser extensions are a completely parallel channel that sidesteps all of that work. Your developers, your product managers, your support teams, they're installing AI writing assistants, summarizers, and coding helpers directly into their browsers. Each one potentially has access to your internal tools, customer data, and active sessions.
It's the shadow IT problem all over again, but with AI capabilities and browser-level access bolted on.
Practical Steps to Take Now
The LayerX report lays out a sensible framework, and I agree with the direction. Here's how I'd prioritize it:
Get a full inventory. You cannot manage what you cannot see. Audit every browser extension across managed and unmanaged endpoints. Cover all browsers, all users. This is step zero.
Apply stricter governance to AI extensions specifically. Given their elevated risk profile, AI extensions deserve targeted policies. Control how they interact with enterprise sessions and sensitive data. Don't lump them in with ad blockers and dark mode toggles.
Move from static approval to continuous assessment. Allowlists are a starting point, not a solution. You need to monitor for permission changes, ownership transfers, and behavioral shifts over time. An extension approved six months ago might have a completely different risk profile today.
Establish minimum trust criteria. Extensions with very low install counts, no privacy policy, or stale update histories should be flagged or blocked by default. This is basic hygiene that most organizations aren't doing yet.
The Bigger Picture
Browser extensions have been treated as a convenience feature for years. A productivity tool. Something harmless. But the combination of AI capabilities, elevated permissions, fast adoption, and zero governance makes them a serious and growing attack surface.
I believe the core lesson here is one we keep relearning in technology: every new consumption channel eventually becomes an attack vector. APIs did. Mobile apps did. SaaS integrations did. Now AI browser extensions are following the same pattern, just faster and with less visibility.
If your AI governance strategy doesn't include the browser, it has a hole in it. A big one.
The full LayerX Extension Security Report is worth reading if you want the detailed data behind these findings.
Written by
Bruno Bonando
Fractional CTO and technology advisor. 23+ years shaping platforms for many companies across Europe and Latin America. Has had leadership roles at REWE, MediaMarktSaturn, Cazoo, and some others.
