A hacking challenge with a Cognitive Captcha to break GenAI this weekend.

Today Saturday I am going to leave you one of the two challenges of Cognitive Captchas the ones I’m playing with with my friends. They don’t seem very difficult, but they have their crumbs. What you need to get is an automatic script that, using a LLM with computer vision and Reasoning be able to break this protection effectively.

Figure 1: A hacking challenge with a Cognitive Captcha

to break GenAI this weekend.

To do this, we must be able to resolve most of the captchas to the first validation request made. When you play with this for a while, I’ll post the solution we found later. And if it has entertained you, I do the same with the other one, which is much more complex but at the same time much more entertaining and fun.

Challenge 1: Visual Captchas

The first is a very simple visual captcha. Then I will tell you which platform it is used on, but it is a challenge that seems very simple, and yet it is not so simple. I leave you three samples that GPT4-Vision and GPT4o They don’t solve it the first time.

Figure 2: Test Captcha 2


Figure 3: Test Captcha 2


Figure 4: Test Captcha 3

And if we see some examples of how asking GPT4-Vision already GPT4o We are not able to solve them the first time.


Figure 5: Example of how GPT4-Vision fails


Figure 6: Example of how GPT4o

If you manage to automate it, as in the examples that I have already told you, leave me the explanation in the comments, or send me an article with the solution directly to my MyPublicInbox mailbox and I will publish it for you. Here are some examples of what we are looking for.

Evil Greetings!

For Latest Updates Follow us on Google News


PREV Best Prime Day Roku deals: amazing savings on streaming devices and TVs
NEXT The new Lenovo Tab Plus incorporates 8 JBL speakers with Dolby Atmos to improve the immersive experience