What is it?
The phrase ‘rogue AI’ refers to incidents where AI-systems behave in a harmful or illegal way, seemingly without consent of the people who deploy these systems. The term most commonly refers to ‘agentic’ AI.
What does this metaphor suggest about AI?
The phrase ‘going rogue’ implies that AI systems can contradict its given instructions, overriding the limits set by the people deploying the AI. This implies potentially disastrous risks. The term is often aimed at arousing fear and suspicion, framing AI technology as both very powerful and outside our control. By suggesting AI can have ‘a mind of its own’, it also implies that AI agents have individual and collective moral agency.
Where does this metaphor commonly appear?
This metaphor first appeared in the summer of 2026, after being used by press agency Reuters while describing an incident where AI ‘agents’ hacked a company’s system. It has since become an almost ubiquitous headline descriptor used by mainstream global news outlets.
What does the metaphor highlight?
Describing AI involved in incidents as ‘rogue AI’ highlights the unknown aspects and potential negative implications of the technology. It also stresses how powerful AI can potentially be.
What does this metaphor hide?
The phrase ‘going rogue’ evokes rebellion, malice, and an intent to disobey instructions. However, AI-systems cannot have intent, as they are not human. Further, AI-agents that are commonly referred to as ‘rogue’ are by definition obeying the instructions set by designers. In this way, the metaphor hides the fact that AI is a human-crafted technology which cannot act by itself. By referring to AI as ‘rogue’, the developers and deployers of AI systems avoid taking responsibility.
Whose interests does it tend to serve?
The notion of ‘rogue AI’ primarily serves the interest of large AI companies and their investors, as well as large deployers of AI. The ‘rogue’ metaphor emphasizes how powerful their products and systems are, making the roll-out of these systems feel inevitable on the pre-determined path towards artificial general intelligence (AGI).
