There are currently rumors floating around about OpenAIs Q* or GPT-5 having broken AES-cryptography. There are screenshots from a paper floating around. I've seen it many times today but here is the shortest version of this rumor/myth.
As someone with basic understanding of cryptography I can immediately tell it's fake. In a sense that I know that the paper wasn't written by someone smart enough to work at OpenAI.
  1. The usage of "we thought only quantum computing could achieve" is an immediate tell. AES is a symmetric enc scheme. QC has nothing to do with symmetric schemes. Populist pop-science like to portray Quantum Computers as just "fast computers". They're not that.
  2. No mention of block cipher mode. This is an immediate tell. AES alone is uninteresting. No cryptographer would ever say they're breaking symmetric crypto without saying what cipher mode they're trying to break.
Ofc I cannot tell you if the panic the last week was because of GPT-5 being scary smart are true. Or if it suggested how to improve itself. Idk about any of that. But I'm sure this paper wasn't written by someone from OpenAI.
Thanks for sharing. People really need to work on their skepticism. These stories about OpenAI have a high potential to be click bait. People have been so spun up with fear about AI. Most of it isn't rational. When I talk with one of these people I ask a lot of questions and rarely get rational questions. Usually it gets into the SciFi real quick.
reply