Can AI really break modern encryption and make bank accounts unsafe?

[SeekerOfTruth007]

I asked ChatGPT whether it could theoretically break encryption based on an algorithm I invented several years ago, and it said if the password was too simple or if the text was too predictable, it could break it, but it said if you used a rotating password with close to true randomness, then it cannot break it.

Problem: NObody uses a real random password and nobody uses a real random PIN number, and 4 to 6 digits for PIN numbers on debit cards and IPhones is ridiculously easy to break compared to the cypher I invented.

I use base 4 numerals to represent 64 characters, and then I triple substitute the numbers into a NxN grid and transpose those numbers, then I convert the transposed text to a daughter alphabet which is transmitted to the recipient. This is similar to an old german cypher, except there's an extra step and I'm using triple substitution instead of double substitution. The AI admitted that if I rotating good passwords, it could not break it by brute force, but there are other techniques it claimed, such as guessed text and obvious formatting of the text, which might help it guess the text over several messages.

I call BS. I'm smart enough not to format the text in a guessable manner, but your bank statement is guessable, so I would be concerned especially as we go forward with more and more powerful computers and better AI systems too.

One way to generate "random" passwords which are nevertheless easy to remember is to use an Irrational Number, such as "e" or "Pi", and Divide by another one-digit number, and use the first N digits of the decimal as each password... when you need to "Rotate" use the next N digits, and so forth. Guessing this is pretty much impossible if you use a long enough password.

So you might use the first 16 digits of Pi on message one, then the next 16 digits of Pi on message two, and so on.

[jshm2]

Lol, you didn't ask any of that...

But most passwords these days are from dumps, not directly compromised.

So a data breech, will usually steal user passwords in the dump along other things. Most users tend to keep the same or similar passwords across platforms.

[NathanDavis]

way easier to find and break code... which is what AI does most of the time

[Anonymous]

Nope

[SeekerOfTruth007]

I can do a base 5 triple substitution which closely maps to ASCII code, or I can do a base 10 triple substitution which would represent a 1000 character alphabet, which would be a primitive code book, where entire basic words represent a single character in the cypher text. It's good for disguising common connecting words, such as "And, or, but, of, for, the, etc". These are the most common words in the English language, and so compacting them into a single character in the cypher text makes the context of the message much harder to guess, especially if you also use random rotating passwords and Transposition.