Captchas have lost the battle to AI, robots are bypassing protection
Captchas are supposed to differentiate you from robots with a visual test that takes a few seconds. Except that AI manages to bypass this protection, like Google Vision or Clip (OpenAI), which outperform humans in terms of object detection. And the advent of these models has real repercussions.
AI is already causing problems for reservations
For example, reservation systems are regularly overwhelmed by these bots which easily bypass Captchas. And there are concrete examples between driving tests in the United Kingdom with several months of waiting, tickets for football matches, online sales… AI monopolizes slots or products to resell them at a higher price. Real users are then deprived.
However, when Captchas were created in the early 2000s by Carnegie Mellon University, it was an innovative solution. Faced with bots that create fake accounts or spread scams, these texts based on text distortion made it possible to protect sites.
Then, the system evolved with ReCaptcha in 2007 by adding the digitization of old books. Then there was version v2 from Google in 2014 with image recognition.
But what solutions can be found to authenticate humans and distinguish them from robots? Developers are exploring new avenues. For example, Google's ReCaptcha v3 analyzes behavior: mouse movements, typing rhythm, etc. Subtleties that robots are incapable of reproducing. For now ?
Other solutions are emerging such as biometrics (fingerprints, voice or facial recognition) but questions remain in terms of confidentiality and accessibility.
The democratization of AI makes the situation even more complex. Companies like OpenAI and Claude are preparing autonomous agents capable of taking control of our PC. So in the future, when these robots act on our behalf on a site, how will the platforms distinguish between “good” bots “malicious” ? Perhaps via digital authentication certificates, an avenue being studied.
Online verification must therefore be completely redesigned with a new standard between ease of use for humans and effectiveness against bots. The challenge promises to be enormous in the face of increasingly advanced AI.
