According to Computerworld, analysts at Gartner, including Dennis Xu, Evgeny Mirolyubov, and John Watts, issued a stark warning last week. They strongly recommend organizations block all AI browsers, like Perplexity Comet and OpenAI’s ChatGPT Atlas, for the foreseeable future. The reason is an “irreversible and untraceable” data loss risk that can’t be adequately mitigated. This isn’t a theoretical problem, as data from cybersecurity firm Cyberhaven in October shows 27.7% of organizations already have at least one user with Atlas installed. In some companies, up to 10% of employees are actively using it, with adoption highest in tech (67%), pharma (50%), and finance (40%). Gartner’s call is based on known risks and others yet to be discovered in this nascent tech.
The Stable Door Is Already Open
Here’s the thing: Gartner’s warning is absolutely correct, but it’s also hilariously late to the party. The data shows the horse has not only left the barn—it’s galloping across the field. 27.7% of organizations already have a user with this installed? That’s a massive foothold. And the sectors where it’s most popular are the ones with the most to lose: finance and pharmaceuticals. It tells you exactly what’s happening. Employees, especially in data-heavy roles, are desperate for productivity tools that can summarize and navigate complex information. They’re downloading and using these AI agents because they see a tangible benefit, security policies be damned. Trying to block them now is a massive game of whack-a-mole.
Why This Is Different From SaaS AI
You might think, “We already have policies for ChatGPT.com, what’s the big deal?” But an AI browser is a whole different beast. A web-based chatbot is a destination. An AI browser is an agent. It operates with the permissions and context of your entire browser session. It can read every page you have open, interact with your internal web apps, and potentially exfiltrate data in ways that are incredibly hard to detect. As Gartner puts it, the data loss could be “untraceable.” The browser is the frontier of your enterprise data, and you’ve just invited a new, opaque layer of software to sit right on top of it. The potential for accidentally pasting sensitive data into a prompt is one thing. An agent automatically harvesting it is another level of risk entirely.
The Impossible Balancing Act
So where does this leave IT and security teams? With a nearly impossible mandate. Blocking is the safe, prudent recommendation. But look at the adoption rates. A top-down block will likely just drive usage underground, making it even less visible and manageable. The real solution has to be twofold. First, yes, you need technical controls to detect and manage these browsers, treating them like any other unsanctioned shadow IT. But second, and more importantly, enterprises need to provide a sanctioned, secure alternative. Employees are voting with their downloads. If the business doesn’t offer a governed, secure version of this powerful productivity tool, people will find their own. The future isn’t about blanket bans; it’s about managed, secure adoption. The question is, can security teams move fast enough to build that before the risks materialize? I’m skeptical.
