Scott Aaronson four days ago on the possibility that AI models will soon break widely used cryptographic systems, and on what responsible disclosure looks like when the parties who need to be notified are effectively the entire world.