Cracking Google’s reCAPTCHA with Machine Learning

  • Published 8th Mar 2017
    Last edited 10th Jun 2017

I’ve had this ‘master plan’ of taking on Google’s reCAPTCHA with some of the web automation tricks I’ve learnt, and I had been sitting on the idea for quite a while now – months.

So I figured that I actually wasn’t going to carry out the master plan. Instead I’d already created a working proof of concept leveraging Artificial Intelligence / Machine Learning (gotta drop those terms for the search engines, yup) via a model that’s already well-trained – the Clarifai API.

The master plan would be training my own model either on Clarifai or Tensorflow (using Google to crack Google = awesome).

Now that they’re coming out with this invisible reCAPTCHA thing – which I really don’t think is worth hyping about. I’d be interested to see once it’s fully released whether I can get my script to pick it up and solve it, I’m sure I could.

Anyway, here is what I’ve been able to do with just under 400 lines of JavaScript (Selenium + WebdriverIO + Node.js + the Clarifai API) – need to post this content before it gets too stale.

Featured Content

I’m setting up wallets and sending private tokens to my team (Ethereum / ERC20)

Cracking Google’s reCAPTCHA with Machine Learning (TensorFlow)

I Don’t Need No Stinking API – Web Scraping in 2016 and Beyond

Comments

  • DaveBowman

    March 12, 2017 at 5:23 am

    Very impressive! But I wonder if it’s economically viable to use, considering not cheap API rates. And after image recognition – do you use any other AI to compare list of object’s terms to the reCaptcha definition or is it compared via a premade vocabulary? (which is a fine thing for a PoC)

  • Francis Kim

    March 12, 2017 at 4:17 pm

    Hi Dave,

    Thanks! Clarifai is actually really cheap at $0.0012 per call. I use a synonyms table and negative keywords to help with accuracy. I’m currently in the process of retraining Inception via Python & TensorFlow.

  • Yorman Alfonzo

    March 14, 2017 at 11:00 am

    Can you help me with this, how do i do this i need to do?

  • Francis Kim

    March 14, 2017 at 11:54 am

    Hi Yorman,

    Sorry this was only intended for my personal research and education so there are no products, code or tutorials that I can point you to. I hope you understand.

  • Cracking Google’s reCAPTCHA with Machine Learning * ChatBots

    June 14, 2017 at 11:12 pm

    […] found “only” one post on Google that caught my […]

  • Checkium Folf

    July 21, 2018 at 1:07 am

    I’ve also got something working with clarifai to solve recaptchas, sadly clarifai isn’t free and it would actually be more expensive than a service especially for recpatcha solving, do you know any tensorflow trained models that would work well with this?

  • Francis Kim

    July 21, 2018 at 9:27 am

    TensorFlow will work well here but I don’t think you’ll find pretrained models just for this. Training a model isn’t too hard – I’d suggest giving it a shot!

Write a comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.