Subscribe for premier reporting on free speech, privacy, Big Tech, media gatekeepers, and individual liberty online.

Google’s “I am not a robot” reCAPTCHA can easily be beaten by a robot

If you’re tired of censorship and surveillance, subscribe to Reclaim The Net.

Google’s reCAPTCHA system has had its share of controversies before, but now reports suggest it may not even be very good at what it does.

That’s at least – when it comes to the technology’s stated purpose of verifying legitimate human users and separating them from bots and other unwanted visitors to a website or service.

Namely, reCAPTCHA is Google’s iteration of CAPTCHA – “Completely Automated Public Turing test to tell Computers and Humans Apart” – both also known to many users as some of the most annoying things on the internet, giving even intrusive ads and trackers a run for their money.

However, critics say that reCAPTCHA, which is image-based, is also a way for Google to crowdsource its users’ time and effort, harvest their data in large amounts, hamper their ability to access websites, and even train the giant’s AI-powered machine learning algorithms. This especially concerns those algorithms meant to serve a future filled with driverless cars, preferably Google’s.

Hence the prevalence of imagery associated with cars, traffic lights, and zebra crossings in these puzzles, observers have noticed.

Now, apparently fighting machine learning with machine learning, the University of Louisiana at Lafayette team has announced at a bot they created, ImageBreaker, defeats reCAPTCHA’s version 2 pretty much every time – more precisely, solving 92.4 percent of tests online, and 94 percent of those it attempted to crack offline.

It doesn’t take a lot of time, either – 14.86 seconds in the first, and 5.27 in the second scenario. This is also the first time that a tool purported by researchers to be able to defeat reCAPTCHA effectively has been used online and on-the-fly.

Click here to display content from

The Louisiana researchers structured their bot in three modules; the second module being powered by machine learning to identify objects in images presented in reCAPTCHA puzzles. But the apparent success of the experiment, that has been shared with Google, has a lot to do with an “essential flaw in the design” of the system.

Internet users concerned about the fate of reCAPTCHA should take solace in the fact Google has version 3, that might or might not fix the problem.

If you’re tired of censorship and surveillance, subscribe to Reclaim The Net.

Read more

Join the pushback against online censorship, cancel culture, and surveillance.

Already a member? Login.