The most recent wave of browser innovation promised to liberate us from the drudgery of repetitive web tasks. AI browsers like Perplexity’s Comet and OpenAI’s ChatGPT Atlas arrived with fanfare, offering autonomous agents that could book your flights, summarize sprawling research papers, and complete forms while you sipped your coffee. Yet beneath this veneer of convenience lies something more unsettling. Gartner, the research firm whose pronouncements can make or break enterprise technology adoption, has issued a thirteen-page advisory urging organizations to block AI browsers for the foreseeable future. The reason? These tools, according to analysts Dennis Xu, Evgeny Mirolyubov, and John Watts, are simply too risky for general adoption.
The Anatomy of an AI Browser
To understand Gartner’s alarm, you first need to grasp what distinguishes an AI browser from the familiar Chrome or Firefox experience. These are not merely browsers with chatbot plugins bolted on. They feature two critical components: an AI sidebar and something called agentic transaction capability. The sidebar can summarize, translate, and respond to queries about the content you’re viewing. The agentic feature, however, is where things get interesting and dangerous. It allows the browser to autonomously navigate websites, fill out forms, click buttons, and complete transactions, often within authenticated sessions where you’re already logged in.
AI browsers feature two critical components: an AI sidebar and something called agentic transaction capability
This autonomy is the innovation that sets AI browsers apart from third-party conversational AI sidebars and basic script-based browser automation. It’s also what makes them a cybersecurity nightmare. When your browser can act on your behalf without constant supervision, the potential for both accidental and malicious harm multiplies exponentially.
What Makes Gartner Block AI Browser Recommendations So Urgent
The Gartner analysts didn’t arrive at their recommendation lightly. Their research, which included limited testing using Perplexity Comet, identified a cascade of vulnerabilities that make AI browsers unsuitable for most organizational environments. The primary concern centers on how these browsers prioritize user experience over security by default. Sensitive user data, including active web content, browsing history, and open tabs, is routinely transmitted to cloud-based AI backends unless settings are deliberately hardened and centrally managed.
Gartner’s research identified a cascade of vulnerabilities that make AI browsers unsuitable for most organizational environments.
Consider what this means in practice. An employee working on confidential financial projections or reviewing proprietary research could inadvertently expose that data to an AI service simply by using the browser’s summarization feature. As Comet’s documentation acknowledges, the browser may process local data using Perplexity’s servers to fulfill queries, reading context like text and email to accomplish requested tasks. Users typically view far more sensitive material in their browsers than they would deliberately type into a generative AI prompt, amplifying the risk of unintended disclosure.
The Danger of Autonomous Actions
While data leakage represents one vector of concern, the agentic capabilities of AI browsers introduce entirely new categories of risk. The Gartner analysts identified several troubling scenarios. First, there’s the possibility of indirect prompt-injection-induced rogue agent actions. A malicious actor could craft a website that deceives the AI browser into performing unintended actions, such as navigating to a phishing site or executing unauthorized transactions. Since large language models remain inherently vulnerable to such attacks, there’s no immediate technical solution on the horizon.
Since large language models remain inherently vulnerable to such attacks, there’s no immediate technical solution on the horizon.
Then there’s the problem of inaccurate reasoning-driven erroneous agent actions. AI models, for all their sophistication, make mistakes. An AI browser tasked with ordering office supplies might misinterpret specifications and place an incorrect order. One instructed to complete a form might fill in fields with wrong or outdated information. These errors, while perhaps individually minor, can accumulate into significant operational and financial damage.
Most alarmingly, there’s the risk of credential abuse. If an AI browser is deceived into autonomously navigating to a phishing website, it could inadvertently hand over login credentials or session tokens. Traditional browser protections, designed to warn users about suspicious sites, can be bypassed by agentic transaction capabilities that operate without constant human oversight.
The Temptation to Automate Everything
Human nature compounds these technical vulnerabilities. Employees, faced with tedious mandatory tasks like cybersecurity training, might instruct their AI browsers to automate such assignments while they focus on more engaging work. This not only defeats the purpose of the training but also normalizes using autonomous agents for tasks they may not be equipped to handle safely. The same impulse that drives people to seek productivity shortcuts becomes a liability when the tools involved have access to sensitive systems and data.
Why Gartner Says Block AI Browser Use Now
Given these compounding risks, Gartner’s recommendation is unequivocal. Organizations should block all AI browsers through network and endpoint security controls, preventing employees from accessing, downloading, or installing them. This isn’t a temporary pause while patches are developed. The analysts emphasize that beyond the identified risks, there are likely other potential risks yet to be discovered, given that AI browsers represent a nascent technology still in its early evolutionary stages.
Gartner’s recommendation is unequivocal: Organizations should block all AI browsers through network and endpoint security controls, preventing employees from accessing, downloading, or installing them.
For organizations with higher risk tolerance willing to experiment, Gartner offers some guardrails. Such entities should limit AI browser use to tightly controlled, low-risk automation use cases with robust safeguards and minimal exposure to sensitive data. If Perplexity Comet is used in a pilot program, administrators should disable the “AI data retention” setting to prevent Perplexity from using employee searches to train its models. Users should also be instructed to periodically delete all memories stored by the browser to minimize data leakage risks.
The Educational Imperative
Even in carefully controlled scenarios, user education becomes paramount. Employees must understand that anything they are viewing could potentially be sent to the AI service backend whenever they activate the AI sidebar or request an autonomous action. This means avoiding exposure of highly sensitive data while the AI browser is active, a discipline that runs counter to how people naturally use web browsers as all-purpose tools for both mundane and confidential tasks.
What This Means for the Future
The Gartner advisory arrives at a peculiar moment in the evolution of artificial intelligence. We’ve grown accustomed to AI features appearing in every product category, often with the implicit assumption that innovation necessarily represents progress. AI browsers seemed like a logical next step 💬 (1) , combining the ubiquity of web browsing with the capabilities of large language models. Yet this collision of technologies has produced something that, at least for now, appears more dangerous than useful for most organizations.
The situation recalls earlier chapters in internet history when promising technologies had to be constrained or abandoned due to security flaws. ActiveX controls, Java applets, and Flash plugins..
The situation recalls earlier chapters in internet history when promising technologies had to be constrained or abandoned due to security flaws. ActiveX controls, Java applets, and Flash plugins all faced similar reckonings when their vulnerabilities became clear. Some technologies eventually matured into safer forms; others simply faded away. AI browsers may follow either path, but for the moment, they occupy an uncomfortable space where their potential cannot overcome their perils.
Navigating the Present
Organizations evaluating whether to block AI browser use face a choice between innovation and security. Gartner’s research suggests that for most entities, particularly those handling sensitive data or operating under strict compliance requirements, the decision should be straightforward. The risks cataloged in the thirteen-page advisory, from data leakage to credential abuse to erroneous transactions, are not hypothetical edge cases but predictable consequences of how AI browsers fundamentally operate.
There’s something almost poignant about the trajectory of AI browsers. They represent genuine technical achievement, a demonstration that autonomous agents can navigate the complex landscape of modern websites. Yet that very capability, untethered from adequate security controls, becomes their fatal flaw. Perhaps future iterations will solve these problems through better isolation, more sophisticated permission models, or AI systems less vulnerable to manipulation. Until that future arrives, however, the prudent course is clear. When Gartner tells you to block AI browsers, the wise move is to listen.
Honestly? I downloaded Comet the day it came out. Thought it was going to change everything about how I research for articles. Used it for maybe a week before this whole security thing blew up.