Executive Overview: A Paradigm Shift in Internet Browsing
The digital landscape is undergoing a seismic shift as the era of "passive browsing" gives way to "autonomous agency." According to the latest State of Agentic Traffic report from HUMAN Security, the internet is increasingly being navigated not by humans, but by AI-driven software agents. In October 2026, a staggering 90% of all agent-initiated requests on HUMAN-protected platforms originated from just two tech giants: Meta and OpenAI.
This development marks a transition from simple, script-based bot activity to sophisticated, goal-oriented "agentic" behavior. These agents—designed to research, browse, and eventually transact on behalf of their users—have seen explosive growth. Total agent request volume in October was approximately eight times higher than in August and 2.6 times greater than in September, signaling that AI-driven navigation is no longer a fringe phenomenon but a fundamental component of modern web traffic.
As these agents prioritize product research and search-based exploration, they are effectively reshaping how businesses manage their online presence. For website operators, particularly in the ecommerce and travel sectors, the challenge of distinguishing between valuable consumer intent and autonomous bot traffic has never been more urgent.
Detailed Chronology: The Meteoric Ascent of Meta’s Muse
The trajectory of AI agents in late 2026 has been defined by the rapid emergence and subsequent market consolidation around specific, platform-integrated models.
The Muse Phenomenon
Meta launched its "Muse" agent on September 8, and its impact was immediate. By October, Muse had captured a dominant 59% share of all agent requests, representing a 19-percentage-point increase from the previous month. In absolute terms, the volume of requests sent by Muse nearly quadrupled in just thirty days. Muse currently executes 3.7 times as many requests as its nearest competitor, OpenAI’s ChatGPT Agent.
The Contraction of Standalone Browsers
The market is showing a clear preference for agents tethered to established ecosystems. Standalone agent browsers, which were early favorites, have struggled to maintain relevance. For instance, Perplexity’s "Comet" agent, which accounted for a massive 41% of all agent requests in August, saw its influence crater to 9.4% in September and a mere 3.6% by October.
This trend suggests a "platform loyalty" effect: users are opting for the AI integrated into the services they already use, such as their primary search engines or social media platforms, rather than downloading specialized, standalone browser agents. This theory is further bolstered by OpenAI’s decision to discontinue its standalone "ChatGPT Atlas" browser in August, effectively funneling its user base into the broader ChatGPT ecosystem.
Supporting Context & Metrics: How Agents Interact with the Web
To understand the significance of this traffic, one must look at where these agents are spending their time. Unlike traditional scrapers, which often target site backends or specific data endpoints, these new AI agents are mimicking human-like browsing patterns.
The Browsing vs. Buying Divide
In October, 85% of all agent-initiated requests were directed toward product and search pages—an increase from 81% in September. This data point is critical for digital marketing teams: it indicates that while agents are heavily involved in the "top-of-funnel" research phase, they are not yet fully autonomous purchasing engines.
However, the "checkout" metric remains a vital signal. While requests to checkout and payment pages represent a small fraction of the total (dropping slightly from 2.4% to 1.9% in October), the absolute volume of agents hitting these sensitive pages is growing at a rapid, compounding rate. Analysts note that agents currently "browse far more than they buy," but the infrastructure for autonomous, agent-led transactions is clearly being stress-tested.
Sectoral Distribution
Ecommerce sites continue to bear the brunt of this traffic, maintaining a stable 60% share of total agent requests. Travel websites have emerged as the second-most-active category, surpassing media sites for the first time in the reporting series. This shift suggests that AI agents are becoming the preferred tool for comparing complex, high-value bookings like flights and hotels, where multi-tab research is a standard human behavior.
The Identity Crisis: Technical Challenges in Verification
As these agents proliferate, the question of "who" is visiting a website has become a technical and legal minefield. HUMAN Security employs a multi-layered approach to identification, combining behavioral patterns, user-agent strings, and proprietary "AgenticTrust" digital signatures to verify the source of a request.
The Complexity of Digital Signatures
The industry standard for agent verification is "Web Bot Auth," a system that allows agents to sign their requests, providing site owners with a verifiable identity. However, implementation remains inconsistent.
Meta’s Muse, for example, underwent a significant transition in its identity infrastructure. Between November 7 and November 9, Muse migrated its signing keys from Meta’s primary domain (meta.com) to its own dedicated domain (muse.ai). Reports from DataDome indicate that as of November 10, nearly 50% of Muse’s requests were being made without any digital signature at all.
This creates a paradox for site administrators. If a security system is configured to only allow "signed" traffic from verified sources, it may inadvertently block a large portion of legitimate traffic. If it allows unsigned traffic, it risks opening the door to malicious actors impersonating these popular agents.
The "Black Box" of Analytics
Perhaps the most troubling finding for web managers is that standard analytics tools are currently unable to distinguish between a human user and an AI agent. When an agent visits a page, it appears to the site’s internal tracking as a standard visitor. This leads to inflated traffic numbers and skewed conversion data, as businesses struggle to determine if a "visitor" is a potential customer or a machine performing market research for a competitor.
Official Statements and Industry Implications
The implications for the upcoming holiday shopping season—specifically Black Friday and Cyber Monday—are profound. As Meta and OpenAI solidify their duopoly, the burden of governance has shifted from the agent creators to the site operators.
In Meta’s own engineering documentation, the company notes that Muse’s browsing activity "will appear as your activity" to the sites it visits, essentially masking the machine nature of the traffic. This philosophy places the entire burden of detection on the website’s own security protocols.
Industry experts argue that the traditional robots.txt file, which has governed web crawling for decades, is becoming obsolete. "An agent like Muse gives robots.txt no name to match," noted Slobodan Manic in a recent analysis. "The decision about letting it in moves entirely to bot protection."
Future Outlook: The Road Ahead
The coming months will serve as the first major test for these agent architectures under high-pressure, high-traffic conditions. The focus of the industry will shift from "can we detect them?" to "how do we monetize or mitigate them?"
Anticipating the Holiday Surge
HUMAN Security plans to release a follow-up report detailing how these agents behaved during the intense traffic spikes of late November. Analysts expect to see a surge in "price checking" behavior as agents scour the web to ensure their users are receiving the best Black Friday deals. Whether these agents transition into actual, automated purchase-makers remains the "billion-dollar question" for the retail industry.
The Standardization of Identity
The long-term health of the ecosystem depends on standardization. If Muse, ChatGPT, and other agents fail to provide consistent, verifiable signatures, site operators will be forced to implement more aggressive, potentially restrictive security measures. A move toward a "verified agent" registry, where companies like Meta and OpenAI provide a universal, machine-readable identity that is consistently signed, is likely the only path forward to prevent a "cat-and-mouse" game between AI developers and cybersecurity providers.
For now, the message to business owners is clear: the internet is no longer a human-only space. The agents have arrived, they are browsing with unprecedented efficiency, and they are here to stay. Preparedness for this new, agentic web is no longer an option—it is a prerequisite for survival in the digital economy.