Maybe I *am* a robot!

Double Agents

The captchas have turned!

According to a new report from Vice, hCaptcha, a version of the widely-used gatekeeping integrations designed to keep unwanted bots out of digital spaces, is suddenly starting to keep humans out of those spaces, too. Why? Because they keep asking people to identify increasingly strange, AI-generated images of increasingly bizarre objects — some of which are even entirely nonexistent.

Take, for example, the "Yoko," a bizarre little object that looks to be some sort of a yo-yo-meets-compass-meets-cyborgian-eyeball.

"This @hCaptcha is too difficult," tweeted one unlucky captcha attemptee, "unless we're expected to assume Yoko" — seemingly a reference to the acclaimed artist and musician Yoko Ono — "is hiding behind the yo-yos."

It's a comically weird example of unexpected AI-generated chaos, especially considering that hCaptcha markets itself as a "privacy-focused" replacement for the standard reCAPTCHA. And indeed, hCaptcha might just do the trick at keeping bots out of your website — if at the cost of keeping the actual humans at bay as well.

Uncanny Valley

Per Vice, the phenomenon seems to be happening a lot on the messaging platform Discord.

"This @hCaptcha being used by @discord is pretty bad," tweeted one impacted netizen, who was asked to identify to identify a "puzzle cube" — a Rubix Cube, seemingly — but was met with a series of decidedly avant-garde versions of the colorful cubes. "I keep failing because I'm not sure if I'm supposed to select the images that look like they come from the uncanny valley or not."

In a particularly goofy case, the robotic gatekeeping tech asked a user to select images with robots. But like the other cases, the AI-generated imagery was questionable, and the user couldn't get past the door.

"I have been trying to sign in for 10 minutes," the Redditor captioned the post, prompting another user to respond with one interesting theory: "AI couldn't solve captcha, so they infiltrated our ranks to design their own, AI-solvable, captcha."

Big Shrug

For its part, hCaptcha says that its tech is just fine, and this is just a bad look from a few isolated instances.

"While most hCaptcha interactions do not result in a visual challenge, many variants are used at any given time," a spokesperson for hCaptcha told Vice. "This particular question was a brief test seen by a small number of people, but the sheer scale of hCaptcha (hundreds of millions of users) means that when even a few folks are surprised by a challenge this often produces some tweets."

More on AI and creative captchas: Uh Oh, OpenAI's GPT-4 Just Fooled a Human into Solving a Captcha


Share This Article