New research has found that AI can be successfully trained to pass CAPTCHA tests flawlessly. They say previously it could only beat it 68 to 71 percent of the time.
If you use the internet in any sort of capacity, you’re probably familiar with CAPTCHAs. They appear on the login pages of many sites, asking users to verify that they are “not a robot”.
The most well-known version is Google’s reCAPTCHAv2. Users are asked to verify their humanity by selecting the images that match the prompt, like clicking all the bicycles.
Google also launched reCAPTCHAv3 in 2018. This version does not require a specific task like v2 but determines humanity based on past interactions on the device. Nowadays v2 is mainly used as a backup.
How did they train the AI to beat the CAPTCHAs?
The academics in Switzerland published their findings earlier this month. They used the existing You Only Look Once (YOLO) object recognition model to identify images.
Then they trained the AI on 14,000 images of objects, just like the ones that appear on reCAPTCHAv2. As the types of objects are limited and all centered around roads, the researchers say this made it easy to train the AI.
It was tested in a variety of settings, including if it moved the mouse in a human way and if there was browser history on the device. Google should also be taking this in account when determining if someone is a bot.
However, the researchers said: “Our findings suggest that there is no significant difference in the number of challenges humans and bots must solve to pass the captchas in reCAPTCHAv2.”
The YOLO model managed to get past the CAPTCHAs every single time in the tests.
A Google Cloud spokesperson told New Scientist: “We have a very large focus on helping our customers protect their users without showing visual challenges, which is why we launched reCAPTCHA v3 in 2018.
“Today, the majority of reCAPTCHA’s protections across 7 [million] sites globally are now completely invisible. The potential issues created by image recognition technology are not new, and we are continuously enhancing reCAPTCHA to deter abuse without creating friction for legitimate users.”
Feature image credit: Karen Grigorean on Unsplash
The post AI can pass old-school CAPTCHAs 100% of the time, say researchers appeared first on ReadWrite.