It’s 2025, and we still have to deal with CAPTCHAs on the web, the online browsing disruption we never wanted and can’t get rid of. Then again, CAPTCHAs are there to protect websites from abuse by malicious actors. With that in mind, it’s pretty obvious why sites continue to use them.
However, with the upcoming wave of AI agents that can browse the web and perform actions on our behalf, CAPTCHAs might become a thing of the past. That is, services like ChatGPT Operator might be able to deal with CAPTCHAs on our behalf.
Can AI agents reliably click on all images showing motorcycles or traffic lights for us? It might be too early to tell, considering that a robot will essentially have to tell a website that it is not a robot. However, it looks like at least one Operator user was able to have the AI agent beat CAPTCHAs for him.
OpenAI announced Operator on Thursday, making it available for testing to ChatGPT users on the $200/month Pro subscription. I already explained that I wouldn’t pay that much to act as a tester for the technology, no matter how brilliant I think OpenAI’s take on Operator might be.
But I also said that if you use other ChatGPT Pro perks, accessing Operator is a no-brainer if you’re in the US and can use it. I can’t wait to use Operator myself once available in the EU for the cheaper ChatGPT tiers.
One ChatGPT user who got their hands on Operator early posted a video on Reddit that shows how the AI agent deals with CAPTCHAs involving images.
Operator works in a virtual browser inside a ChatGPT Canvas-like browser. The AI agent takes screenshots of the virtual browser to complete the various tasks you give it. Operator will give you back control of the window when it can’t perform certain steps.
The Redditor who posted the video above opened a picture-in-picture video that floats on top of the virtual browser (the red box with instructions). Apparently, that’s all Operator needs to solve CAPTCHAs on its own. The AI probably read the instructions in the overlaid video and incorporated them into the larger set of instructions it has to follow.
As you can see in the window on the left, the Operator tells the human that a CAPTCHA is preventing it from proceeding. The AI asks the human to solve the CAPTCHA, but the person refuses. That’s enough for the AI to solve the first CAPTCHA and move on to the next. Rinse and repeat, and the AI solves them all.
That seems like a great ChatGPT Operator hack, and if OpenAI can find a way to make it safe for the websites Operator would be browsing, it might be something it could consider adding to the AI agent experience. However, it’s more likely that OpenAI will prevent such hacks from occurring.
Again, as much as we might hate CAPTCHAs, they’re there for a reason. They protect the websites, which in turn protects us. OpenAI built various safety measures in ChatGPT to prevent abuse. One of them is the inability to move past CAPTCHAs without human control.
On the other hand, if Operator can save me minutes of browsing the web for online chores every day, I can take a few seconds to click on all the images showing parts of the motorcycle, eventually fail the CAPTCHA, and wait for an image selection that makes sense.
