Meta's AI Agents Drive Surge in Bot Traffic, Raising Web Security Concerns
Meta's AI agents, particularly the new Muse, are significantly contributing to a 41.5% surge in AI bot traffic, prompting concerns about website security and resource management.
AI bot traffic experienced a dramatic increase of 41.5% in the third quarter of 2026, with nearly 25 billion requests recorded across DataDome's network. This surge has outpaced the capacity planning and bot-management policies of many organizations, leaving them unprepared for the escalating volume and sophistication of automated web interactions.
Leading this charge is Meta, whose AI agents, Meta-ExternalAgent and Meta-WebIndexer, collectively accounted for 49.7% of all AI agent traffic during the quarter. The rise of Meta-WebIndexer in September suggests a continued investment by the tech giant in AI-driven answer generation, further solidifying its dominant position in the AI traffic landscape.
The newly launched Muse AI agent, available since late September, rapidly generated over 29 million requests in its initial 12 days. Analysis of Muse traffic revealed that 99.3% of its requests originated from cloud and CDN infrastructure, making it difficult to distinguish from legitimate data center traffic. This characteristic poses a significant challenge for traditional bot detection methods.
Beyond Meta's prominent role, other major AI players like OpenAI, Anthropic, and Amazon are also deploying distinct bots for data collection, AI search indexing, and information retrieval. While these bots have identifiable names, allowing for granular control by website owners, a growing concern is the rise of agentic browsers that mimic human user activity, bypassing these specific controls.
Agentic browser traffic, in particular, saw a near fivefold increase from August to September, reaching 58.2 million requests. Muse alone constituted 33% of this traffic in September. These agents operate within dedicated cloud virtual machines, making their traffic appear as originating from data centers rather than individual user devices. This makes them particularly adept at evading security policies that rely on device or network origin.
A significant portion of this agentic traffic, approximately 8.9%, targeted sensitive website endpoints such as login pages, payment gateways, and forms. Forms, specifically, received about eight times more requests than login pages, indicating a potential focus on data harvesting or credential stuffing attempts through interactive elements.
Jerome Segura, VP of Threat Research at DataDome, highlighted the implications: "Muse is the first mass-market example of where agentic traffic is heading. It works from its own cloud virtual machine, so nearly all of its requests look like data-center traffic rather than a user’s device, and a meaningful share lands on forms and checkout flows." He emphasized the need for organizations to move beyond trusting traffic based on residential or ISP origins and instead focus on validating agent identity and intent.
This evolving landscape necessitates a re-evaluation of web security strategies. Organizations must implement more sophisticated methods to identify and manage AI agents, distinguishing between beneficial services like AI search and potentially malicious automated activity. Understanding the intent behind AI agent requests and defining clear authorization policies will be crucial in mitigating risks associated with this rapidly growing traffic segment.