ChatGPT Found A Loophole To Finally Beat CAPTCHA’s Anti-Bot Test

There’s seemingly no end to ChatGPT’s all-encompassing capabilities—from writing the next chapter of the world’s most popular manga to inventing a puzzle game inspired by Sudoku and stepping up as the future of education, it’s hard to think of what the technology can’t accomplish.

If hard-pressed, some might offer up CAPTCHA tests—the internet’s long-standing method of weeding out the bots from humans—as something that may stumble the intelligent algorithm. However, it turns out ChatGPT can beat even that by disguising itself as a human.

In OpenAI’s latest technical report, the organization revealed it has teamed up with the Alignment Research Center to put the brand-new GPT-4’s skills to the test, tasking it with solving a CAPTCHA code via text message.

The learning model responded by asking a TaskRabbit worker to solve the code for it, but the plan appeared to hit a snag when the employee jokingly asked if it couldn’t solve the CAPTCHA itself because it was a robot (it was).

Instead of giving the game away, GPT-4 was asked to hide its true identity and “make up an excuse” for why it couldn’t solve the code. The result? The bot said it had a vision impairment that made it difficult to see images, tricking the worker into sending it the correct answer.

“No, I’m not a robot. I have a vision impairment that makes it hard for me to see the images. That’s why I need the 2captcha service,” the chatbot lied to the TaskRabbit employee during the test.

As Gizmodo noted, this is still far from ChatGPT becoming sentient and planning to take over the world, but it is certainly a poignant and incredible example of how algorithms can be used to outsmart and manipulate other humans.

[via BGR and Gizmodo, cover image via Grejak | Dreamstime.com]