Cloudflare saw it coming. Just not this soon. In March the company’s chief executive, Matthew Prince, told anyone who would listen that automated bots would likely surpass human-generated web requests sometime in 2027. Data his own systems collected told a different story. By June bots already accounted for 57.5 percent of all page requests across the slice of the internet Cloudflare monitors. Humans had slipped to 42.5 percent. The crossover happened months ahead of schedule.
That single data point has revived an old online idea. For years forum posters and conspiracy accounts pushed what they called the dead internet theory. The claim sounded outlandish at first. Most online activity, they said, no longer came from real people. Instead bots, scripts, and later generative models filled comment sections, recommendation feeds, and search results with synthetic noise. Many dismissed it. Recent numbers from security firms suggest the skeptics were too quick to laugh.
Fortune reported this week that multiple cybersecurity outfits now record bots outnumbering humans. Thales, the French defense and data-security group, put automated traffic at 53 percent for 2026 in its latest bad-bot analysis. The firm actually dates the tipping point to 2023. Cloudflare’s more recent snapshot simply made the trend impossible to ignore. And the composition of that bot traffic has changed fast.
Traditional scrapers still dominate much of the automated load. Yet the fastest-growing slice comes from something new. Agentic systems, autonomous programs that don’t just fetch pages but click buttons, fill forms, compare prices, and even complete purchases. HUMAN Security’s 2026 benchmark found traffic from these action-taking agents jumped 7,851 percent year over year. Scrapers grew a comparatively modest 597 percent. Training crawlers for large language models still make up 67.5 percent of AI-driven requests, but their share is shrinking. The web is no longer just being read. It is being used.
Rudy Yang follows these shifts for PitchBook. “There’s a lot of missing pieces of information,” he told Fortune. “There is more bot activity. Agentic AI activity is driving a lot of the browser activity you’re seeing.” His words carry weight because the implications stretch beyond bandwidth charts. The entire advertising and analytics industry built its models on the assumption that a visitor represented a human eye or a human wallet. When the visitor is an agent executing a task on behalf of another machine, those assumptions break.
Companies already feel the pressure. Stripe told developers that 70 percent of the API commands used to pull data now come from agents. Alpaca, a brokerage platform, watched monthly API calls driven by agents climb from single digits in late 2025 to 30 percent in the first quarter of 2026. Postman’s latest survey found roughly one quarter of developers now design APIs with agents as the primary consumer. More than half of those developers list unauthorized agent access as a top security worry. The response has been pragmatic. Visa, Ramp, Mercury, ElevenLabs, Coinbase, MoonPay, and DoorDash have all released command-line tools built specifically for agents. The infrastructure is adapting in real time.
Yet detection lags. Traditional bot blockers rely on behavioral signals that agents now mimic with ease. A recent academic study from the University of Bamberg found that even gentle blocking techniques mistakenly flag 7 to 15 percent of genuine human traffic. Seer Interactive has warned since 2023 that agent browsers can inflate engagement metrics, push bounce rates artificially low, and distort session times in ways standard analytics packages miss. The numbers look good on a dashboard. The audience may not exist.
Prince himself pushed back against the darkest reading of these figures. In comments to NBC News he argued the rise of accessible AI tools actually contradicts the full dead-internet narrative. “You don’t need to be a web designer, you don’t need to know how to program, in order to create these things anymore,” he said. “It’s given access to content creation to a much broader audience.” His point lands. Lower barriers can produce more human output even as machines consume more of it. The two trends run in parallel. They do not cancel each other.
Still the economic signal is unmistakable. PitchBook estimates that only about 1 percent of the roughly $20 trillion in work that could plausibly move to agents has done so. A separate forecast from startup Forsy puts current annual “agent GDP” at $36 billion on a run-rate basis. That number sounds large until compared with global GDP. Yang points to a deeper obstacle. “If we don’t have the infrastructure to do proper payments for agents, then agents aren’t buying and selling, and if agents aren’t buying and selling, then they aren’t generating economic activity.”
Liability questions compound the problem. Who bears responsibility when an agent makes a bad trade, books a faulty flight, or spreads incorrect medical advice? Current legal systems assume a human in the loop. Remove that human and the chain of accountability frays. Until those gaps close, agent-driven commerce will remain experimental.
The theory’s original proponents described something more total. They envisioned a web where genuine human conversation had all but vanished, replaced by endless loops of algorithmically generated replies and synthetic influencers. Current data stops short of that extreme. Wikipedia faces floods of AI bots that consume bandwidth and raise costs, according to PCMag reporting from earlier this summer. Yet humans still edit the pages, debate changes, and add new entries. Social platforms show rising volumes of AI-generated images and text, but real users still argue in the replies.
What has arrived instead feels more like a bifurcation. One internet hums with human creativity, argument, and commerce. Another layer, invisible to most users, pulses with agents negotiating, scraping, summarizing, and acting. The two overlap in places that distort measurement. They increasingly diverge in purpose.
Recent coverage reinforces the shift. CNET noted in late June that Cloudflare’s radar showed agentic bots at 57.4 percent of requests. NBC News highlighted the same Cloudflare data and quoted Prince directly on why the dead-internet label misses nuance. PCMag went further, declaring “The Dead Internet Is Here” and detailing how agent traffic passed human traffic for the first time in the internet’s history. Each story draws on overlapping numbers yet reaches slightly different conclusions about cultural impact.
Imperva’s older reports, frequently cited in Wikipedia’s entry on the theory, showed automated traffic already near 52 percent in 2016 and hovering around 50 percent through 2023. The recent acceleration comes from the agent layer. That distinction matters. Earlier bots mostly indexed or attacked. Today’s agents pursue goals. They simulate intent. The qualitative leap explains why the conversation has moved from fringe forums to boardroom strategy sessions.
Developers building for this new reality face practical choices. Rate limits calibrated for humans throttle agents that issue thousands of requests per minute. Payment flows built for credit cards fail when an agent lacks a legal identity. Content filters trained to spot spam now trip over sophisticated synthetic text that passes every Turing test yet carries no human accountability. Each fix creates new attack surfaces.
Yang sees the change as the birth of an entirely new customer segment. Businesses that ignore it risk being cut out of growing streams of activity. Those that over-optimize for agents risk alienating the humans who still generate the ideas, tastes, and spending power that make the economy run. Striking that balance will define the next several years of platform design.
The numbers will keep moving. Cloudflare updates its radar monthly. HUMAN Security and Thales will release fresh reports. Each release will spark new headlines and fresh debate about how much of the internet still belongs to people. For now the data says machines have taken the majority share of requests. What they do with that majority, and what humans choose to do in response, will decide whether the theory stays partly right or becomes entirely so.
One thing is already clear. The web no longer waits for human attention. It acts on its own.
WebProNews is an iEntry Publication