169
Warning: Your AI-Generated Password Is a Major Security Risk. Here’s What to Use Instead (www.rd.com)
submitted 8 months ago by to c/technology@lemmy.world
65 comments fedilink hide all child comments
 

At a glance, the passwords the LLMs created looked secure, much like those that a password generator might spit out. But that’s exactly where the problems arose: Although the AI-generated passwords appeared to be complex and safe to use for securing online accounts, they were actually quite predictable upon closer inspection.

All three LLMs exhibited clearly identifiable patterns in how they created these passwords. These patterns included repeated character strings, predictable password structure, frequent reuse of similar characters, clear biases toward certain numbers and letters, and even duplicate passwords in some cases. Although the AI-generated passwords looked random, they really weren’t. This could easily create a false sense of security if you were to use these predictable passwords for your online accounts.

all 66 comments

sorted by: hot top controversial new old
[–] 92 points 8 months ago (19 children)

which dumbass lets LLMs generate their passwords?

  • permalink
  • fedilink
  • source
  • hideshow 19 child comments
    •
  • [+] 36 points 8 months ago (12 children)
    [removed by mod]
  • permalink
  • fedilink
  • source
  • parent
  • hideshow 12 child comments
    •
  • [–] 33 points 8 months ago (9 children)

    What is that? All I see is *******.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 9 child comments
    •
  • [–] 10 points 8 months ago (7 children)

    Let me test that

    JsfJhdYhb57'j4++€€djF

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 7 child comments
    •
  • [–] 14 points 8 months ago (6 children)

    Shit 😩

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 6 child comments
    •
  • [–] 5 points 8 months ago (5 children)

    Don't put a quote mark in your password! I learned the hard way with password'drop table users;'

    /S

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 5 child comments
    •
  • [–] 5 points 8 months ago (4 children)

    Wait, was this because you too had a student named "Robert'); DROP TABLE users;--"?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 11 points 8 months ago

    What do you mean. Why would it be a problem that my password is the most statistically likely sequence of characters in the world? And that the password is stored in plaintext in a chat log? And used for training LLMs on password conversations?

  • permalink
  • fedilink
  • source
  • parent
    • load more comments (-1 replies)
    [–] 157 points 8 months ago* (last edited 8 months ago) (15 children)

    Why does everything have to be LLM, like why cant things sometimes just be algorithmically generated like my AI-free password manager does?

    I wonder if there's AI-powered password managers on the market now lol. If so and if customers, I would mirror Zuck's snide "thuh dumb fucks" setiment when people trusted him

    ZXCVBN algorithm has never failed me. I can generate passwords that would take 1 quadrillion guesses/second hackers over 90 million years (testing 1 quad/sec) on the reg

  • permalink
  • fedilink
  • source
  • hideshow 15 child comments
    •
  • [–] 30 points 8 months ago (2 children)

    It's worse than just using a new tool though. By definition LLMs use the statistically most likely option (with minor variation for flavor). People are literally asking a statistics bot what password can I use that is the most likely?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 16 points 8 months ago* (last edited 8 months ago) (1 child)

    There's AI-powered everything now for no reason other than because when they slap those two letters on it, their stocks go up. A lot of times there's not even any AI involved.

    I remember CES 2022 when derb8uer found these "AI PC cases" and when he asked about why, they said "because you can put an AI PC in it". And that's when I knew it was a bubble. Because that's the kind of shit that happened with dotcom. Companies slapping "dotcom" on their names and their stocks would skyrocket for no reason.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 7 points 8 months ago

    I always say I'm not interested in asking an LLM to add 2 + 2 for me. Of course my managers are always pushing us to use LLMs and coworkers keep suggesting we replace efficient, testable, and consistent processes with AI. If you're going to use it at least think of scenarios that are hard to code for and it would take you at least 10 min to solve.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 1 point 8 months ago (3 children)

    While “routing” of prompts between specialist models and traditional APIs does offer more efficient and reliable outputs, it also

    A. requires more elbow grease than one massive generalist model, and

    B. doesn’t help you avoid paying API licensing fees by obfuscating their outputs into the blackbox weights of your proprietary model

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 9 points 8 months ago (2 children)

    It just seems like a password manager is an infinitely better tool. Like to the point i severely judge and question the sense of someone who seeks to reinvent the wheel for something so crucial

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 3 points 8 months ago (1 child)

    Oh for sure. I was just answering in case you wanted to know.

    Future versions of AI tools will likely use routing to do a variety of tasks more efficiently and accurately. But until then? It’s a jinn to those who would wish away every single chore they lack the patience to comprehend including, unfortunately, information security.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] -2 points 8 months ago (1 child)

    Why does everything have to be LLM

    Not everything is, it's just Lemmy is obsessed with them so you're going to see it come up a lot

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 10 points 8 months ago

    Unfortunately it's not just Lemmy. Corporations are obsessed with LLMs as are a lot of the people working for them. I daily get messages and meetings about LLMs in one way or another and it drives me nuts because it keeps interrupting actual work. In the span of like two months it went from something that was mentioned maybe once in the last year to be 90% of what everyone is talking about.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 39 points 8 months ago

    Why would anyone... Who the fuck is using glorified autocomplete to generate a random unique password?

    Anyone that stupid deserves the breach and any company that integrates that moronic idea deserves the blowback.

  • permalink
  • fedilink
  • source
    •
    [–] 32 points 8 months ago (1 child)

    why. dafuq. would someone use an llm to make a password.

  • permalink
  • fedilink
  • source
  • hideshow 1 child comment
    •
    • [–] 31 points 8 months ago (1 child)

    AI generated password

    ... I'm sorry, what just happened? I think I had a mini aneurysm. Let me read that ag.....

    In all seriousness, WTAF is wrong with people? What removed Baphomet wannabe came up with that negative IQ idea? SERIOUSLY... WTF IS WRONG WITH PEOPLE. ARGH!

    Here I am giving security training to my coworkers and then I have to read this. This is like me spending a life long sifting through all my garbage to make sure everything is nicely separated for recycling, using the bike to go anywhere, to make sure my carbon footprint is as small as possible and over flies Taylor Swift in her private jet undoing my life's work in a single fucking trip.

    Seriously this gives me a headache

  • permalink
  • fedilink
  • source
  • hideshow 1 child comment
    •
  • [–] 7 points 8 months ago

    I watched someone in a YouTube video ask ChatGPT to add up some numbers for them, despite having a calculator on their phone. The sheer laziness at the expense of someone else's energy was impressive. And don't even get me started on the accuracy.

  • permalink
  • fedilink
  • source
  • parent
    • [–] 26 points 8 months ago

    AI generated password? What kind of brain rot idea is that, god damn. Do these people need to ask AI to breathe for them?

    I can also generate a password for them while running 50 toasters and flushing the toilet a bunch if they want.

  • permalink
  • fedilink
  • source
    •
    [–] 17 points 8 months ago

    LLMs are predictive text machines. Focus on "predictive". Of course they will not output random text.

    Note: not fully deterministic though - they need (pseudo)randomness at few critical points to be good

  • permalink
  • fedilink
  • source
    •
    [–] 26 points 8 months ago (2 children)

    My brother you can literally slam your face on the keyboard to get a password. Why would you ask an llm' to do it for you?

  • permalink
  • fedilink
  • source
  • hideshow 2 child comments
    •
    • [–] 10 points 8 months ago

    "here's what to use instead"

    ...the same thing we always have been? Keepass or other password managers with randomly generated passwords?

  • permalink
  • fedilink
  • source
    •
    [–] 18 points 8 months ago (7 children)

    It takes a five line Python script to generate a secure password. But no, let's waste a gigantic amount of energy to have a slop machine do it very badly.

  • permalink
  • fedilink
  • source
  • hideshow 7 child comments
    •
  • [–] 5 points 8 months ago (5 children)

    Bet I can do it in less

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 5 child comments
    •
  • [–] 6 points 8 months ago* (4 children) 
    import random
password_length=10
letters = "abcdefghijklmnopqrstuvwxyz0123456789-&@$?!*#%"
print(''.join(random.sample(letters+letters+letters.upper()+letters.upper(), password_length)))

    If you inline password_length and letters, you can bring it down to 2 lines

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 2 points 8 months ago (2 children)

    Sure. The five lines are just for readability.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
    • [–] 9 points 8 months ago

    if only there existed a better way to generate passwords

  • permalink
  • fedilink
  • source
    •
    [–] 7 points 8 months ago (4 children)

    Just noticing: there's 0 evidence in article that anyone is doing this. I just don't buy that this is happening enough to matter. Interesting as interpretability research at best

  • permalink
  • fedilink
  • source
  • hideshow 4 child comments
    •
  • [–] 3 points 8 months ago (3 children)

    There's actually lots of evidence of people using AI to generate GUIDs that are infact not globally unique.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 1 point 8 months ago (2 children)

    Source appreciated? Was this inside the research paper?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 2 points 8 months ago* (last edited 8 months ago) (1 child)

    I've seen it personally at work where the AI generates its own metadata files containing uuids that it made up, and they end up being duplicates from elsewhere in the project. Unfortunately I can't really share links.

    I'm sure you could find examples in GitHub issues

    Edit: I had an honest look, and can't seem to find anything that isn't AI slop in web searches anymore... GUIDs aren't exactly the most common thing in the first place, so maybe I overstated how common this is.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
    • [–] 7 points 8 months ago

    Wait ... Wat? Why would you even...

  • permalink
  • fedilink
  • source
    •
    [–] 6 points 8 months ago

    How did we manage to move almost completely away from Pseudo-random password generation, only to come straight back to it via LLMs again.

    People will really just shove these things into anything like they're a bloody panacea, when really they're just extremely well-trained parrots.

  • permalink
  • fedilink
  • source
    •
    [–] 3 points 8 months ago

    Is there a link to the actual study?

  • permalink
  • fedilink
  • source
    •
    [–] 1 point 8 months ago

    I am constantly impressed with the level of general idiocy of end users when it comes to stuff like this...

  • permalink
  • fedilink
  • source
    •