Why staying anonymous online could get much harder as AI bots multiply

As AI-powered bots grow harder to distinguish from real users, Google's Incognito sign-in test signals a broader shift from checking behaviour to verifying identity online

Try searching Google in Incognito mode, click through a few pages of results, and you might now see something unexpected: a prompt asking you to sign in "to verify you're a human and see more results." No puzzle, no traffic lights to click, no wobbly text to decipher — just a login screen standing in for a CAPTCHA. Search Engine Roundtable spotted the behaviour and reported it this week, though Google hasn't said how widely it's being tested or whether it will roll out further.

 

It could be dismissed as a minor interface change. The context suggests otherwise.

The internet was built around anonymous users

For most of the web's history, proving identity was never a requirement for simply browsing it. Websites relied on IP addresses, cookies, browser fingerprints and behavioural analysis to separate legitimate visitors from malicious ones, and when those signals proved unreliable, a CAPTCHA served as the final check. The system rested on a single assumption: most traffic was human, and automated activity was the exception worth screening for.

 

 

That assumption no longer holds.

 

According to the Thales Bad Bot Report 2026, bots accounted for 53 per cent of global internet traffic in 2025—the second consecutive year in which automated traffic has outpaced human activity. The breakdown is more revealing still: roughly 40 per cent of all web traffic is bad or unverified bot activity, while only 13 per cent comes from bots performing legitimate, verified tasks. Platforms are no longer filtering out occasional automated requests. They are attempting to identify human activity within an internet where automation has become the dominant source of traffic.

AI is changing what a bot looks like

Traditional automated traffic tended to follow predictable patterns that security systems could detect. Modern AI-powered bots do not share that limitation. They can follow expected user flows, generate interactions that appear organic, and move through a website in ways that closely resemble genuine browsing sessions. The Thales report notes that the indicators long used to distinguish humans from automation are becoming less reliable, precisely because AI-driven traffic increasingly appears legitimate.

The report also identifies a newer category: AI agents, autonomous software capable of browsing websites, retrieving information and completing tasks on a user's behalf without being directed click by click. Unlike a conventional bot executing a fixed script, an agent behaves adaptively—and that adaptiveness is what makes it difficult to detect.

 

Set a CAPTCHA against that kind of traffic, and its underlying purpose begins to erode. It was designed to distinguish humans from machines by behaviour. If machines have learned to behave like humans, behaviour ceases to be a reliable signal. Identity becomes the stronger one.

From behaviour to identity

A signed-in account offers more than authentication; it provides a persistent identity, one considerably harder for an automated system to reproduce at scale than solving a single image-based puzzle. A CAPTCHA can be scripted around. An account is a more difficult barrier to replicate.

 

Notably, the prompt reportedly appeared specifically in Incognito mode, after a user had browsed several pages of results. This detail is significant: it suggests Google may be beginning to treat anonymous browsing itself as a weaker signal warranting additional verification, rather than simply another mode of using Search.

 

The pattern extends beyond this single test. Publishers have increasingly restricted AI crawlers from accessing their content. Reddit has tightened access to its data through commercial licensing agreements. Cloudflare has introduced systems to distinguish verified AI crawlers from unauthorised scraping bots. Across the industry, the operative question is shifting from whether traffic appears human to whether the identity behind it can be trusted.

AI agents are changing internet traffic itself

A person browses intermittently—opening a session, reading, closing it, returning later. An agent does not follow that pattern. It retrieves information continuously, communicates with cloud services, exchanges data with other software systems, and continues operating well beyond the point at which a human user would typically stop. As enterprises move from deploying a handful of AI assistants to potentially millions of autonomous software workers, internet infrastructure is being asked to serve a category of user it was not originally designed for.

This shift is already shaping discussions beyond search engines. In the telecom industry, operators are preparing for networks in which AI agents become continuous consumers of network services rather than occasional software clients. Executives have argued that this will require new approaches to authentication, security and machine identity, as methods built around human users are increasingly inadequate for machine ones.

 

The same principle applies more broadly. As autonomous software becomes a routine participant online, platforms require mechanisms capable of distinguishing between people, trusted machines and malicious automation.

Why platforms want to tell bots from humans

The motivation is not simply technical tidiness. Bad bot traffic carries direct costs.

 

Automated scraping consumes server resources and bandwidth that platforms would otherwise allocate to genuine users, and at the scale described in the Thales report, that cost is substantial. Bots are also the primary tool behind credential stuffing, account takeovers and fraudulent transactions, activities that translate directly into financial losses and security incidents for the platforms hosting them. Retailers, in particular, have reported bots used to hoard inventory, distort pricing signals and skew analytics, making it harder to gauge genuine demand.

 

There is a commercial dimension as well. Search engines, publishers and platforms depend on advertising and subscription models that assume a human is on the other end of a page view. Inflated or fabricated engagement from bots undermines the metrics advertisers pay for and the trust those business models rely on.

 

AI scraping introduces a further concern that is less about fraud and more about value capture. When an AI system retrieves and repackages a publisher's content to answer a user's query directly, the publisher loses the visit, the ad impression and the subscription opportunity that would once have followed. This is a significant part of why publishers have moved to restrict AI crawlers and why Reddit and others have shifted to licensing arrangements rather than allowing free access. Identifying and controlling automated traffic, in that sense, is as much about protecting revenue as it is about security.

 

Seen against this backdrop, Google's experiment fits a broader commercial logic, not just a defensive one. Distinguishing humans from bots is not only about stopping abuse; it is about preserving the value of every interaction that platforms are built on.

Anonymous browsing may become more difficult

Anonymous browsing has long served purposes unrelated to avoiding personalised advertising. Journalists rely on it to research sensitive topics, security researchers use it to investigate malicious websites, developers use it to test products, and many ordinary users simply prefer to separate certain searches from their primary online identity. If identity becomes a prerequisite for accessing more of the web, the distinction between anonymous and authenticated browsing begins to narrow.

 

This does not necessarily mean anonymity will disappear. It is more likely that anonymous access will face additional friction, while authenticated users are granted faster or uninterrupted access.