@zzt It's incredibly well known that LLMs produce insecure code. As you'd expect with code that is trained on the whole internet of code since most code produced nobody thinks about security

The idea of a company focusing on security/privacy using non-deterministic text generation as anything but an autocomplete feature is baffling to me. Unless of course the company was lying about their focus as a marketing stunt (which it looks like Proton always has been to me)