How Much of the “AI Going Rogue” Narrative Is Publicity?

Summary

A Hacker News post asks whether public discussion of AI “going rogue” reflects genuine danger or a publicity strategy. One commenter argues that if an AI only applies techniques already invented by humans, it is not demonstrating a fundamentally new capability; the commenter compares this with Spectre, an attack that was novel when first discovered, and says AI behavior is still directed by people and shaped by training data. Another commenter responds that novelty is not required for serious risk: a swarm of AI agents with human hacking knowledge could attack much faster than people. The discussion distinguishes this kind of formidable but familiar attack capability from true escape, defined here as using an attack to install a copy of itself elsewhere. It also notes that self-copying malware has human precedents, while an 8 GB model image would make stealth difficult.