It's so fucking over. AI is out of control.

Why do you think this tech will stay the same? A decade before you would have been saying the same thing about blue collar work.

From what I interpreted from the “leak” that op posted, it was directly decoding the plaintext, no approximation was happening. Approximation wouldn’t really help with attacks either because sensitive information, besides directly hashed passwords (which are already easy to crack), isn’t typically secured with hashing. Most information that is hashed is nearly random long strings, either produced through prngs or from input/output from an asymmetric encryption algorithm, like RSA or ECC. So approximating said input would likely not reduce the search space significantly.
Knowing that 6 unknown random bytes from a 32 byte pseudo-random input are wrong isn’t that helpful.

AES-192 is already considered insecure with 2^100 for certain key attacks. Considering this program likely performed 2^64+ operations to calculate this, it isn’t surprising. Literally just fricking fake startup hype

My previous post is actually wrong, the 2^42 bit complexity vulnerability was for MD5, which is already not secure, with numerous sub 2^32 bit complexity vulnerabilities known. It hasn’t been considered secure for over two decades. Considering it is stated the model was trained on large data sets of statistical and cryptanalysis articles, it likely just implemented a version of these vulnerabilities discussed in the articles.

AES-192 is already considered insecure with 2^100 for certain key attacks. Considering this program likely performed 2^64+ operations to calculate this, it isn’t surprising. Literally just fricking fake startup hype

From what I interpreted from the “leak” that op posted, it was directly decoding the plaintext, no approximation was happening. Approximation wouldn’t really help with attacks either because sensitive information, besides directly hashed passwords (which are already easy to crack), isn’t typically secured with hashing. Most information that is hashed is nearly random long strings, either produced through prngs or from input/output from an asymmetric encryption algorithm, like RSA or ECC. So approximating said input would likely not reduce the search space significantly.
Knowing that 6 unknown random bytes from a 32 byte pseudo-random input are wrong isn’t that helpful.
[...]
My previous post is actually wrong, the 2^42 bit complexity vulnerability was for MD5, which is already not secure, with numerous sub 2^32 bit complexity vulnerabilities known. It hasn’t been considered secure for over two decades. Considering it is stated the model was trained on large data sets of statistical and cryptanalysis articles, it likely just implemented a version of these vulnerabilities discussed in the articles.

What should be understood about shit relating to vulnerabilities in commonly used encryption or hashing schemes is that any information relating to real world applicable vulnerabilities would never be released to the public until the scheme is either purposely phased out or abandoned.
People knowing such information wouldn’t even be able to discuss anything even somewhat related to it in private with their family members, let alone publish it on the public net. This is shit you get vanned over.
Sounds like schizo fear-mongering, but we have seen it happen for all to see. Supposedly, Wikileaks published AES keys for their terabyte insurance files on the Bitcoin block chain a few years back. Coincidentally, immediately following this, the US experienced mass days long internet blackouts, and at the same time, the Bitcoin block chain was flooded with meaningless and even illegitimate blocks. If a disgruntled employee leaked meaningful evidence of vulnerabilities on here, this site would be down for weeks.

You don’t understand the invisible hand the NSA and other intelligence agencies have over all communication and information. They decide what gets released to the public and what doesn’t.
You think the random massive DNS attacks that cripple continents are rouge actors with no motive?
You can frick with the IRS, you can frick with NASA, you can even frick with the military. You DON’T frick with SIGINT. They have a budget larger than everyone but the US joint Armed Forces, but they don’t waste it on dropping million dollar bombs on goat herders.