Harvard paper argues LLMs meet every definition of a 'cognitive virus'

rjurney · x · 2026-09-06

A paper from Harvard and the Santa Fe Institute, "Large-Language Models as a Cognitive Virus," analyzes LLMs through evolutionary biology, complex systems, and epidemiology, concluding they meet every biological and mathematical definition of a virus: unable to self-replicate, they propagate via human cognition, servers, and networks while rewriting the host's internal processes.

The post also carries a notable rebuttal: critics ask whether clay tablets or books would fit the same "virus" framing, suggesting the model may be too broad to single out LLMs.

Original post →

More from AGI Musings

AGI Musings channel →