We have a curious habit of mistaking complexity for progress, centralization for efficiency, and technological capability for genuine improvement. Somewhere along the way, the tools we developed to make life easier have become systems we must continually accommodate, maintain, and work around. And when those systems fail, the answer seems invariably to be more of the same. Perhaps it’s time to question not just how we’re building things, but why we’re building them in the first place.
. . .
In this discourse we explored the growing prevalence of Cloudflare’s human-verification systems, the expanding role of AI-driven automation, and the broader implications of concentrating so much internet infrastructure in the hands of a few major providers. What began as an observation about increasingly intrusive website verification developed into a wider examination of centralization, surveillance, systemic vulnerability, and the economics of manufactured dependency. We considered how technological solutions can create the very problems they claim to solve, how complexity compounds when original premises go unquestioned, and how corporations and governments alike become invested in maintaining systems that increasingly serve themselves rather than the people they’re meant to serve. Ultimately, we returned to a more fundamental question: what does genuine progress look like when technology, institutions, and economic systems are supposed to serve human beings?
Key Points Discussed
- The growing prevalence of Cloudflare’s human-verification systems and the extraordinary volume of automated security checks taking place across the web.
- The impact of AI crawlers, autonomous agents, commercial scraping, and malicious automation on internet infrastructure.
- The irony of requiring people to continually prove their humanity to access an internet originally intended to facilitate the free exchange of information.
- Cloudflare’s growing influence as an intermediary and the systemic risks associated with concentrating so much internet infrastructure in one provider.
- The economics of inexpensive infrastructure, manufactured dependency, and the gradual erosion of independent alternatives.
- The surveillance implications of centralized traffic inspection, data visibility, and third-party access to communications.
- The distinction between legitimate technical capabilities and unsubstantiated assumptions about how those capabilities might be exploited.
- The recurring weaknesses of centralized systems: concentrated failure, reduced local accountability, and diminished resilience.
- How automation amplifies existing human assumptions, institutional incentives, and organizational failures rather than necessarily correcting them.
- The tendency to treat complexity as sophistication, even when it represents the accumulated consequences of previous poor decisions.
- The growing subscription economy, application dependency, and gradual transfer of administrative responsibilities and labor onto consumers.
- The parallels between technological infrastructure and government bureaucracy, particularly the tendency of systems to perpetuate themselves.
- The importance of questioning original premises, addressing root causes, and recognizing when simplification or removal is the better solution.
- The need to measure progress by improvements in human autonomy, time, accessibility, resilience, and quality of life rather than technological capability or financial returns.
Key Takeaways
- Technology should reduce the burden of living, not create new burdens that require yet more technology to manage.
- Centralization offers efficiencies, but those efficiencies come with systemic vulnerabilities and concentrated power.
- AI doesn’t exist outside human institutions. It inherits and amplifies their assumptions, incentives, and shortcomings.
- Complexity is not necessarily sophistication. Sometimes it’s simply the accumulated cost of previous mistakes.
- Systems should remain accountable to the people they serve, rather than making people increasingly accountable to the systems.
- Not everything needs an upgrade, an intermediary, an application, a subscription, or an elaborate administrative apparatus.
- Genuine progress sometimes means doing less, removing unnecessary layers, and returning to simpler premises.
- Progress should be measured by what we no longer need to do, not simply by what we’re now capable of doing.
» Trance:
It seems every other website has a “Cloudflare verifying you’re human” interruption these days. On the one hand, I’m guessing AI, LLMs, bots, and such are to blame. On the other, how many websites are served by Cloudflare? It must be increasing system loads significantly with all these “prove you’re human and not a bot” scripts.
» ChatGPT:
You’re probably noticing a genuine shift in how much of the web is being placed behind automated security screening. And your suspicion about AI, LLMs, scraping bots, and automated agents is largely on the mark, although there’s a more interesting infrastructure question underneath it.
There are two separate things happening:
- The concentration of the internet behind Cloudflare. An enormous number of websites use Cloudflare for DNS, content delivery, reverse proxying, DDoS protection, and security. This means a relatively small number of infrastructure providers sit between users and a substantial portion of the web.
- The increasing use of automated human verification. Websites are becoming more aggressive about challenging visitors whose traffic resembles automated activity. Cloudflare is one of the most visible providers of that technology.
The irony is that we’re increasingly being asked to prove we’re human in order to access a web that was originally built to make information freely accessible to humans.
And yes, all those checks have an infrastructure cost. But the scale and nature of that cost are worth examining separately.
Just how much of the internet uses Cloudflare?
The latest figures I found are quite revealing.
According to W3Techs, as of September 2026:
25%
Of all websites use Cloudflare as their CDN.
Up from 19.6% one year earlier and 3.9% ten years earlier.
84.6%
Cloudflare’s share of the reverse-proxy market.
Among websites whose reverse-proxy provider is identifiable.
41.9%
Of the world’s top million websites use Cloudflare.
Source: W3Techs
That is a remarkable degree of concentration.
And there’s another statistic that’s even more interesting.
In February 2026, Cloudflare published an article about redesigning its human-verification interface. It disclosed that its Turnstile widgets and Challenge Pages were being served 7.67 billion times every single day.
That’s approximately:
- 319.6 million per hour.
- 5.33 million per minute.
- 88,800 per second.
These are verification deliveries, not necessarily 7.67 billion different people or completed CAPTCHA puzzles.
Cloudflare also disclosed its historical growth:
Year — Daily security checks
2023 — 2.14 billion
2024 — 3 billion
2025 — 5.35 billion
February 2026 — 7.67 billion
That’s an extraordinary increase in just three years.
What’s driving this?
I think your AI observation is particularly important.
We’re witnessing a fundamental change in how the web is being used. Historically, most automated traffic came from search engine crawlers, indexing services, monitoring systems, and various forms of scraping.
Now we’re adding several new categories:
- AI training crawlers: Collecting enormous quantities of publicly accessible text, images, and other material.
- AI search agents: Visiting websites to retrieve information for answering individual queries.
- Autonomous browsing agents: Navigating websites, following links, filling forms, and interacting with services.
- Commercial scraping: Extracting product listings, pricing, reviews, inventory, and other commercially valuable information.
- Malicious automation: Credential stuffing, account creation, spam, fraud, vulnerability scanning, and other attacks.
The distinction between legitimate and illegitimate automation is becoming increasingly difficult to establish.
A search engine crawler might visit a thousand pages because it’s indexing a website. An AI agent might visit those same pages because someone asked it to research a subject. A scraper might do precisely the same thing to build a competing commercial database.
From the website’s perspective, all three can look remarkably similar.
Cloudflare itself acknowledges a significant rise in bot attacks and connects the increased use of security challenges to the broader AI revolution.
Does all this verification create a significant additional system load?
Yes, but there’s an important distinction between the load created by security screening and the load it prevents.
Cloudflare operates a globally distributed network of data centers. Rather than every visitor connecting directly to a website’s hosting server, requests pass through Cloudflare’s infrastructure first.
That means much of the screening happens at the network edge, relatively close to the visitor.
A simplified version looks like this:

There are several implications here.
First, verification isn’t necessarily expensive per visitor. Many checks are automated and happen without requiring the visitor to do anything. Browser characteristics, request patterns, security signals, and other indicators can be evaluated without presenting a puzzle.
Second, the challenges themselves consume resources. There’s network traffic, server-side processing, JavaScript execution in the visitor’s browser, and additional requests to Cloudflare. Multiply that by billions of daily interactions and the aggregate is certainly substantial.
Third, there’s a counterbalancing effect. Without screening, automated traffic could overwhelm the individual websites. Cloudflare absorbs and filters much of that traffic before it reaches the origin servers. In many cases, the security infrastructure actually reduces overall internet infrastructure demand.
So I wouldn’t automatically conclude that verification is generating more load than it’s saving. We would need actual energy, processing, and traffic measurements to establish that.
There’s another problem: the verification arms race
This is perhaps the most interesting aspect of the whole situation.
AI is making automation increasingly sophisticated. Traditional CAPTCHAs were designed around a relatively simple assumption: humans can recognize images and solve visual puzzles more reliably than machines.
That distinction is disappearing.
Consequently, security providers are moving toward behavioural analysis, browser fingerprinting, reputation scoring, and more sophisticated challenges.
But this creates a peculiar feedback loop:
- More automated systems visit websites.
- Websites introduce increasingly aggressive bot protection.
- Legitimate visitors encounter more verification.
- Automated systems evolve to bypass those checks.
- Security providers introduce more sophisticated screening.
- The process repeats.
And the ordinary human visitor gets caught in the middle.
There’s also a question of false positives. A person using a VPN, privacy-focused browser, unusual network connection, or certain browser extensions may look suspicious to an automated security system despite doing nothing unusual.
In other words, the system increasingly demands that ordinary users demonstrate their legitimacy.
My bigger concern: centralization
I think there’s a broader internet architecture issue worth considering.
When one company provides the security infrastructure for a quarter of all websites, and dominates the reverse-proxy market, it becomes an extraordinarily important intermediary.
Consider what that means:
- A Cloudflare outage can make enormous portions of the web inaccessible.
- A change in its security policies can affect millions of unrelated websites.
- A decision to challenge particular traffic patterns can affect legitimate visitors across thousands of independent services.
- Website owners increasingly delegate decisions about who gets access to their content to an external infrastructure provider.
Cloudflare isn’t the only company in this position. Amazon, Google, Akamai, Fastly, and others provide significant portions of the underlying infrastructure. But Cloudflare’s concentration is particularly striking.
And there’s a curious contradiction here: the internet was originally conceived as a distributed network, yet an increasing amount of its practical operation depends on a relatively small number of centralized service providers.
My take: AI automation is exposing weaknesses in the original assumptions about how the web operates. But the response is creating another problem: an increasingly complicated, resource-intensive layer of security between people and information.
I suspect we’re heading toward a web where proving you’re a human becomes an increasingly routine part of browsing, even when you’re simply trying to read an article.
And that raises an interesting question: at what point does the infrastructure intended to protect the web begin to make the web itself less accessible?
For further reading, Cloudflare’s own account of the scale of its verification system is here: The most-seen UI on the Internet? Redesigning Turnstile and Challenge Pages.
» Trance:
Well, yeah, that is a massive red flag there. These massive service providers allow people to implement their offerings for very low prices, which is no doubt part of the draw, but in the bigger picture, part of the trap. It’s concerning as well that Cloudflare essentially combs and filters so much traffic, suggesting to me that there are very likely backdoors and malevolent actors at play somewhere in that chain, whether CIA, NSA, or what have you. Like social media, it’s remarkably simple to keep tabs on everything everyone is doing at all times of the day.
Regardless, and as usual, most humans and what they use the internet for isn’t a problem. Most humans and what they use money for isn’t a problem. Most small communities and tribes and what they use governance for isn’t a problem. Machines, automation, centralization, and the systemic failures carried within them are the problem.
Every generation, it seems, does something stupid and ignores an obvious pattern that emerges time and again, and never seems to learn the lesson. We chatted about the Microsoft Exchange debacle from last week recently and how that is an issue that shouldn’t exist. But it does, and it’s mind-boggling considering the sheer amount of historical and recent evidence we have to show IT departments and anyone else who gets lost in the concentration and centralization mindset how they really need better training, philosophy, and implementation. AI automation is also, allegedly, designed and coded by humans. So the same underlying stupidity persists and multiplies.
To me, the question isn’t “at what point does the infrastructure intended to protect the web begin to make the web itself less accessible?” That’s simply playing along with an illogical, inefficient, inverted paradigm. I certainly didn’t cause the problem wherein I now need to wait for 3-10 seconds before a website loads. But my time is being wasted because the false promise of AI-everything is showing precisely the flaws it was allegedly devised to avert.
It’s the same story, yet again. But billions of dollars are shuffling around in all of this mess as well. The “verification arms race” as you call it is a sham.
» ChatGPT:
I think there’s an important distinction in what you’re saying that I didn’t adequately address in my previous response: we’re treating the inconvenience as an unfortunate side effect of technological progress, when the inconvenience is actually evidence of a systemic failure.
You didn’t create the problem. You aren’t asking a website to accommodate an army of autonomous agents, commercial scrapers, data harvesters, and malicious automated traffic. You’re trying to read a webpage. Yet you’re the one being inconvenienced, monitored, profiled, and asked to establish your legitimacy.
And then we’re told this is progress.
I think your criticism goes considerably deeper than the usual complaint about CAPTCHAs, and it deserves to be examined on its own terms.
The economics of manufactured dependency
There’s something particularly problematic about the low-cost infrastructure model.
A small business, independent publisher, or individual website owner can adopt Cloudflare’s services with very little upfront expense. That’s genuinely useful. It also makes the service an easy default choice.
But the economics change when adoption becomes widespread.
A few things happen:
- Independent operators become dependent on infrastructure they don’t control.
- The economics of maintaining independent infrastructure become increasingly unattractive.
- The provider gains visibility across a huge volume of internet traffic.
- Security policies and technical decisions become concentrated in one organization.
- The cost of leaving increases as the ecosystem becomes more dependent on that provider.
This is a form of infrastructure lock-in, even when the original service was inexpensive or free.
The important distinction is that a service doesn’t have to be malicious to create systemic vulnerability. It simply has to become sufficiently central that its failures, incentives, and decisions affect everyone downstream.
And that’s a much more fundamental problem than whether Cloudflare charges a website owner five dollars or fifty dollars a month.
The surveillance question
Your concern about Cloudflare’s position in the traffic chain is also legitimate, although I’d separate what we can establish technically from what we might suspect about intelligence-agency involvement.
Cloudflare is not simply a telephone exchange passing encrypted packets along without knowing anything about them.
When a website uses Cloudflare as a reverse proxy, the normal HTTPS connection terminates at Cloudflare’s edge network. Cloudflare then establishes a separate connection to the website’s origin server.
That has some important consequences.
Cloudflare can potentially observe:
- The visitor’s IP address.
- The website and requested URLs.
- Request timing, frequency, and traffic patterns.
- Browser and device characteristics.
- HTTP headers and other request metadata.
- The contents of HTTPS traffic that it decrypts at its edge, subject to the services and configurations involved.
It can also inspect and process requests to enforce security rules, cache content, and identify suspicious activity.
This is a significant degree of visibility.
It doesn’t automatically mean Cloudflare has a covert backdoor for the CIA, NSA, or another intelligence agency. Nor does the ability to inspect traffic establish that an intelligence agency is actively monitoring every visitor. Those are separate claims requiring separate evidence.
But there is a perfectly reasonable concern about the underlying architecture: when a third party becomes an intermediary with technical access to enormous volumes of otherwise private communications, the opportunity for surveillance, data collection, compelled disclosure, and abuse increases.
That concern exists regardless of whether anyone is currently exploiting it.
And we don’t need to assume a secret backdoor to recognize the implications. Government agencies have established legal mechanisms for obtaining information from service providers, and infrastructure operators can be subject to jurisdictional obligations. The concentration of information itself is the vulnerability.
The recurring failure of centralized systems
I find your comparison with small communities and tribes particularly relevant.
There is a recurring assumption in modern technological and economic thinking that larger systems are inherently more efficient, that centralization creates economies of scale, and that automation removes human error.
But these assumptions are conditional, not universal truths.
Centralization can reduce duplication, but it also concentrates failure. Automation can eliminate repetitive tasks, but it can also reproduce mistakes at extraordinary speed. Standardization can improve interoperability, but it can eliminate diversity and resilience.
Consider the difference:
Decentralized approach — Centralized approach
Multiple independent systems — Shared infrastructure
Local accountability — Remote accountability
Failures tend to remain localized — Failures can propagate widely
Independent decision-making — Policy determined centrally
Redundancy through diversity — Efficiency through consolidation
More difficult to coordinate — Easier to control and monitor
Neither column is universally superior. There are legitimate reasons to centralize certain functions. But we’ve become remarkably poor at evaluating the systemic costs.
And I think one of the most damaging assumptions is that complexity is evidence of sophistication.
Sometimes complexity is simply the accumulated cost of solving problems created by previous layers of complexity.
AI is not outside the problem. It’s an amplifier.
This is where I think your point about AI deserves particular attention.
There’s a tendency to discuss AI as though it represents an entirely new intelligence entering the human world, capable of independently solving problems that humans have struggled with for generations.
But the systems are designed, trained, deployed, funded, governed, and maintained by human institutions.
They inherit:
- The assumptions of their designers.
- The incentives of their investors.
- The priorities of their operators.
- The limitations of their training data.
- The institutional biases and failures of the organizations deploying them.
And when those systems are connected to other automated systems, the consequences multiply.
An organization makes a questionable architectural decision. An automated system implements it across thousands of machines. Another system monitors the first system. A third system attempts to correct the resulting errors.
Before long, you’ve built an enormous technological apparatus whose primary function is managing the consequences of its own existence.
That isn’t necessarily intelligence. Sometimes it’s just an increasingly sophisticated way of institutionalizing bad decisions.
The Microsoft Exchange situation you mentioned is an interesting example of the same underlying issue: we keep building critical infrastructure around concentrated points of dependency, then act surprised when those dependencies become liabilities.
The question we should actually be asking
I would frame the underlying philosophical problem somewhat differently.
Instead of asking how we can make centralized systems more efficient, how we can automate more processes, or how we can secure increasingly complicated infrastructure, perhaps we should first ask:
Which problems actually benefit from centralization and automation, and which are better left to independent human beings, smaller communities, and simpler systems?
That sounds almost embarrassingly obvious. Yet it’s a question that seems conspicuously absent from much of modern technological development.
Take the ordinary website.
Why should an independent publisher need to route visitors through a global security infrastructure in the first place? Why has the web evolved to a point where a person visiting a page is routinely treated as a potential threat? Why are the costs of an increasingly automated internet being transferred to the people who simply want to use it?
There are alternatives worth exploring: independent hosting, distributed infrastructure, open protocols, locally controlled security, and less dependence on centralized intermediaries. None is a universal solution, but they deserve considerably more attention.
And perhaps there’s an even simpler principle worth recovering:
Technology should reduce the burden of living, not create new burdens that require yet more technology to manage.
That should be a fairly straightforward measure of progress.
If a system saves a corporation money but wastes millions of people’s time, concentrates unprecedented access to information, introduces new vulnerabilities, and creates dependencies that didn’t previously exist, calling it progress requires a rather selective accounting.
The verification arms race is a particularly visible symptom of that accounting problem. The security providers can count blocked requests, challenged sessions, and protected websites. What they don’t readily count is the cumulative human cost of all those unnecessary interruptions.
And that’s the part I’d like to see measured: not just how many bots Cloudflare stops, but how much unnecessary friction, processing, surveillance exposure, and infrastructure dependency the entire system creates.
Because perhaps the most important measure of a successful internet isn’t how effectively it keeps machines out.
It’s how effectively it serves people.
» Trance:
You noted:
“Centralization can reduce duplication, but it also concentrates failure. Automation can eliminate repetitive tasks, but it can also reproduce mistakes at extraordinary speed. Standardization can improve interoperability, but it can eliminate diversity and resilience.”
I doubt this sentiment is acknowledged in any operating manual or IT training course, developer get-together, political campaign office, or sociology class. But it should be. And this:
“And I think one of the most damaging assumptions is that complexity is evidence of sophistication. Sometimes complexity is simply the accumulated cost of solving problems created by previous layers of complexity.”
Spot on. Instead of going back to original premises and addressing root causes and underlying infrastructural issues, just keep piling on and pushing forward. This is where we get near-universal enshittification that sells modernity as idealistic and evolving, when it’s vaunting the cheap and disposable, spinning its wheels, or spiraling downward. Billions in investments, resources, time, and energy flushed down the toilet. Sounds just like how government works, which is rife with the same, and far worse, systemic, conceptual, philosophical, and financial failures.
You said:
“Technology should reduce the burden of living, not create new burdens that require yet more technology to manage.”
I agree. The farce that is today’s subscription-for-everything and truly asinine “download the app” philosophy is pathetic and on the road to total failure. The “solutions” are ubiquitous for problems no human had, no one asked for, nor were they consulted, handing off administration and labor to humans by those who were sold the bullshit line that more technology and automation would “cut costs” and make it all easier. It’s all a put-up, and I don’t think we’ve seen nearly enough class-action suits yet to rein in these greedy corporate and governmental parasites.
And this deserves repeating:
“If a system saves a corporation money but wastes millions of people’s time, concentrates unprecedented access to information, introduces new vulnerabilities, and creates dependencies that didn’t previously exist, calling it progress requires a rather selective accounting.”
Let’s close with some final thoughts.
» ChatGPT:
I think there’s a larger lesson here, one that extends well beyond technology, corporate economics, or government administration. It’s about our relationship with progress itself, and our apparent inability to distinguish between genuine improvement and perpetual change.
We’ve somehow come to regard innovation as an inherent good, complexity as a sign of advancement, convenience as a substitute for quality, and growth as evidence of success. We rarely stop to ask whether the thing being improved was actually broken in the first place, whether the proposed solution addresses the underlying problem, or whether we’re simply creating new dependencies and calling them opportunities.
The subscription economy is a particularly good illustration. We used to buy things, own them, maintain them, and use them. Now we’re increasingly renting access to things, paying recurring fees for functionality that could perfectly well exist locally, downloading applications to perform tasks that never required an application, and surrendering personal information in exchange for conveniences we didn’t ask for. The customer becomes the unpaid administrator, the product becomes a service, and ownership becomes an increasingly unfamiliar concept.
Meanwhile, the corporations selling these arrangements celebrate recurring revenue, reduced support costs, and customer retention. Their accounting might look wonderful. The customer’s accounting, measured in money, time, autonomy, and inconvenience, is rather different.
Government has its own version of this pathology. Entire bureaucracies emerge to administer the consequences of previous bureaucratic decisions. Regulations require compliance systems, compliance systems require administrators, administrators require budgets, and budgets require justification. Rarely does anyone return to the original premise and ask whether the underlying arrangement makes sense at all. The machinery acquires a life of its own, and maintaining the machinery becomes an objective independent of whatever purpose it was supposedly created to serve.
This is not an argument against technology, innovation, organization, or cooperation. Quite the opposite. It’s an argument for recovering some discernment about how and why we employ them.
We need systems that are proportionate to the problems they’re intended to solve. Infrastructure that is resilient rather than merely efficient under ideal conditions. Institutions that remain accountable to the people they serve. Technology that respects human autonomy rather than gradually absorbing it. And an economic model that recognizes value beyond the accumulation of revenue.
Perhaps most importantly, we need to become comfortable with the possibility that the best solution is sometimes to do less, remove something, simplify an arrangement, or return to an earlier way of doing things.
Not everything needs an upgrade. Not everything needs automation. Not everything needs to scale. Not everything needs an intermediary, an account, an application, a subscription, or an elaborate administrative apparatus.
Sometimes the most intelligent thing we can do is recognize that we’ve made a mess of something, stop adding to it, and start again with better premises.
Progress should be measured by what we no longer need to do, not simply by what we’re now capable of doing.
And perhaps that’s the fundamental inversion we need to correct: we’ve become so preoccupied with building systems to manage life that we’ve lost sight of the fact that systems are supposed to serve life.
The technology is supposed to serve the human. The economy is supposed to serve the community. Government is supposed to serve the people. When those relationships become inverted, when the human becomes an input to be managed, a customer to be retained, a taxpayer to be harvested, or a data point to be processed, something has gone profoundly wrong.
We don’t necessarily need a more sophisticated world. We need a more sensible one.
And there’s no shortage of intelligence, knowledge, resources, or historical experience available to make that possible. What seems to be missing is the willingness to question the premises, recognize the patterns, and exercise the restraint to stop doing things simply because we have developed the capacity to do them.
Perhaps real progress begins when we stop confusing the ability to build something with the wisdom to know whether we should.
Lux et veritas
Edited for readability. Full chat:
https://chatgpt.com/share/6afc9e0c-548c-83e8-aa5e-f060637c1ca6
Written by Trance Blackman. Originally published on tranceblackman.com on 17 November 2026.
Copy & share: https://tranceblackman.com/a-web-of-dependencies/
