Lukasz Olejnik
Academic publishing really needs to catch up on AI use policy! IACR/CRYPTO/EUROCRYPT appear to be enforcing disclosure rules based on an obsolete ACM policy. ACM itself has already updated its approach. The old-style requirement can mean documenting prompts used to generate content. Which raises the obvious question: all prompts? Including “okay, let’s continue”? ;-) How about: Prompt #47: “yes”. Prompt #48: “okay, let’s continue”. Prompt #49: “make it shorter”. Prompt #50: “no, the previous version was better”? ;-) This is what happens when policies designed around “one prompt → one output” meet modern multi-turn AI workflows, including agentic and loop engineering. I contrast it with the more mature, current ACM guidelines.
