The scraped data of 2.6 million DuoLingo users was leaked on a hacking forum, allowing threat actors to conduct targeted phishing attacks using the exposed information.
Oh no. Now they know the aliased email address, unique password, and that I didn’t try very hard to learn spanish.
(please note: this is a joke, I don’t see anything about them getting passwords)
Something to note here - with AI, if you’re using any sort of heuristic for your password, it’s pretty simple to work out a pretty good set of possibilities which makes brute force even easier and puts you at risk across the board.
Always come up with random passwords that are as random as possible. If there’s a path you took to get to a password, in theory it can be worked backward.
For example I know some people who only change a single letter when changing their passwords which is ultimately trivial to guess if the old password was compromised (hence the need to change the password or the need to proactively work against this possibility)
I wish more websites allowed random words as passwords instead of forcing numbers and special characters (but not THAT special character, you have to use one of the ones on this list).
People change their passwords by one letter or digit because they’re tied to these restrictive formats. If 5-6 random words was the norm, people would update more than just one character when needing to change passwords.
“poison navy series ruler handshake papaya” is a fantastic password.
“Ilovemygrandkids!123” is a horrible password.
Just use a password manager and a unique, long, random generated password for every site. There’s no need or reason to know the password to anything other than your password manager and your primary email.
You know somebody has to link this.
Agreed! I also think that the next steps would be getting rid of the need for users to even know their own password and instead replace with other securities like biometrics (with sufficient permutations possible to match or exceed passwords) and a physical device or something else entirely that removes the need to let the user in on what the exact password is
Tools like Bitwarden will let you fairly customize the randomly generated password it makes. You can tailor it to not use certain characters for those sites that don’t allow it. And each vault object can be customized like that independently so you don’t compromise all your passwords by not allowing _ or (, you can also have it do pass phrases like you gave an example of
I use a heuristic to update my main passwords. It’s not a character but easily guessable if you see it in plaintext and now you’ve made me facepalm my actions.
I only use that for certain things because I use Google Oauth or Bitwarden for most things and you’ve just woken me up about what could be exposed.
The goal should usually be as random as possible, if it’s got a series of steps to create, they can be traced backward
Now the trick I’m not telling you is that randomness is hard to get because you need a sufficient amount of entropy (basically just means randomness, chaos, formally it’s how much uncertainty there is in the system) to ensure that it’s strong enough which can be challenging sometimes. For example, if your password is only 3 characters long and has 10 possibilities for each spot in the string, you’re only looking at 10^3 possibilities to guess accurately which is nothing to pcs and people with time on their hands haha
That’s why I let Bitwarden generate a random 64 character password with special characters and numbers
I also take advantage of Bitwarden’s ‘passphrase’ generation as I understand that pass phrases can be even more secure.
If the password requirements allow longer passwords I typicallyuse a passphrase generated by bitwarden, shorter ones I use generated passwords.
something I did before letting bitwarden take over my passwords, was using a phrase consisting of 2-3 words + a series of numbers and special characters. Safer than anyone I knew at the time’s passwords. Admittedly it was not the most secure, as i only changed the beginning part of the 2-3 word phrase, and left the last word, numbers and symbols the same. So if one of those passwords were breached, it wouldn’t be too difficult for AI to brute force the missing pieces. So yeah I don’t do that anymore.
That’s why correcthorsebatterystaple is the best way to do passwords imo, just 4 random words with a random special character dividing them and a random number tacked onto the end. Good luck brute forcing that or using AI to guess 4 randomly generated words in the correct order.
we were talking about password changes, not creation though
Guessing someone’s password with no prior history vs with an “averaged” prior history of the world/some large dataset are two different sized sets.
If you’ve got a feel for how the majority of people are changing their passwords, guessing those passwords is significantly easier when compared to traditional brute force
Edit:
Passphrases are fairly good too but I want to see some real word examples of AI trained on some password dumps to see how much better it performs in comparison to traditional brute force and through targeted info gathering. I’m curious to see if there’s any user friendly techniques that’d work against AI specifically and it’s ability to find patterns most people wouldn’t pick up on
I could explain in detaiI if that’s cool? I have a degree in computer Science and beginner level experience with making my own models from datasets (not my own but from various repos) and with other applications (LLM, image gen, and then some other miscellaneous stuff that’s not really important)
Otherwise I can try to find somethin for ya sure
Next email from duo: give me your credit card details
Do the people that release these get paid somehow? Or do they just do it for hacker cred and say fuck these 2.6M people?
In January 2023, someone was selling the scraped data of 2.6 million DuoLingo users on the now-shutdown Breached hacking forum for $1,500.
…
As first spotted by VX-Underground, the scraped 2.6 million user dataset was released yesterday on a new version of the Breached hacking forum for 8 site credits, worth only $2.13.
“Today I have uploaded the Duolingo Scrape for you to download, thanks for reading and enjoy!,” reads a post on the hacking forum.
They’ll send fake emails where the green owl comes to collect “late fees” for your 216-day streak of missed Spanish lessons.
Oh no, not my German and Japanese scores!!!
I guess the email could become a spam target?? Gmail does a good job sorting that for me.
They know your email, your name, and that you’ve taken German anf Japanese. Next they use that information to craft a phishing email that only the very stupid would fall for, which fools an alarming number of people. Something like “Hi, this is Duolingo suppert, and your billing information may have been comprimised. Log into this portal with your credit card credentials to confirm that you were not affected.”
Damn, they’ll know I didn’t finish that Spanish lesson the bird bothered me about!
