When logging on to HBO Max at the end of May, people noticed something strange.  Usually when someone logs into the site, HBO asks them to verify that they are human by solving a captcha — you know, the little "I am not a robot" checkbox or the "select all squares with stoplights" image grids that prove to the website that you are, in fact, a human.

But this time, when users logged on they were asked to solve a complex series of puzzles instead. The bizarre tasks ranged from adding up the dots on images of dice to listening to short audio clips and selecting the clip that contained a repeating sound pattern. These odd new tasks, ostensibly to prove users were human, haven't been limited to HBO: Across platforms, users have been stumped by increasingly impossible puzzles like identifying objects — such as a horse made out of clouds — that do not exist.

The reason behind these new hoops? Improved AI. Since tech companies have trained their bots on the older captchas, these programs are now so capable that they can easily beat typical challenges. As a result, we humans have to put more effort into proving our humanness just to get online. But head-scratching captchas are just the tip of the iceberg when it comes to how AI is rewriting the mechanics of the internet.

Since the arrival of ChatGPT last year, tech companies have raced to incorporate the AI tech behind it. In many cases, companies have uprooted their long-standing core products to do so. The ease of producing seemingly authoritative text and visuals with a click of a button threatens to erode the internet's fragile institutions and make navigating the web a morass of confusion. As AI fever has taken hold of the web, researchers have unearthed how it can be weaponized to aggravate some of the internet's most pressing concerns — like misinformation and privacy — while also making the simple day-to-day experience of being online — from deleting spam to just logging into sites — more annoying than it already is.