# Large-Language Models as a Cognitive Virus

**Authors:** Ricard Solé, Giulio Ruffini, Francesca Castaldo, Marco Tuccio, Luis F. Seoane, Manlio de Domenico, Santiago F. Elena, David C. Krakauer, Michael Levin  
**arXiv:** [2609.03344](https://arxiv.org/abs/2609.03344)  
**Date:** 2026-09-03  
**Summary page:** https://www.summarizepaper.com/en/paper/2609.03344v1/

## Key points

- Large-language models (LLMs) are rapidly integrating into human culture, fundamentally altering information generation, sharing, and utilization.
- LLM adoption follows a viral spread pattern, permeating populations and influencing cognitive and cultural behaviors.
- The study analyzes transitions among different types of LLM users: uncoupled, coupled, and persistently dependent individuals.
- Social transmission, recovery mechanisms, and collective reinforcement can lead to tipping points and technological lock-in.
- Runaway dynamics may occur when a critical mass of LLM adoption is reached, leading to swift shifts towards persistent dependence and potential declines in cognitive abilities.
- Strategies for cognitive immunization against excessive reliance on LLMs include reducing transmission rates and promoting reversibility.
- The research emphasizes the nonlinear nature of collective transitions associated with LLM adoption and highlights significant implications for cognitive autonomy.

## Summary

In their paper titled "Large-Language Models as a Cognitive Virus," Ricard Solé and his team delve into the rapid integration of large-language models (LLMs) into human culture. These advanced language technologies are fundamentally altering how information is generated, shared, and utilized. Drawing on the analogy of a viral spread, the authors suggest that LLM adoption follows a similar pattern to permeate through populations and become ingrained in cognitive and cultural behaviors. The researchers develop a model to analyze the transitions among different types of LLM users - uncoupled, coupled, and persistently dependent individuals. Through this analysis, they demonstrate how social transmission, recovery mechanisms, and collective reinforcement can lead to tipping points and technological lock-in. A key finding is the potential for runaway dynamics: once a critical mass of LLM adoption is reached, even small increases in usage can trigger swift shifts towards persistent dependence. This can result in sudden declines in cognitive abilities. However, the study also identifies strategies for cognitive immunization against excessive reliance on LLMs by reducing transmission rates and promoting reversibility. The results underscore the nonlinear nature of collective transitions associated with LLM adoption and emphasize the significant implications for cognitive autonomy. Overall,<Organization> Solé et al. 's </Organization> research sheds light on the evolving landscape of language models within society. It highlights the importance of understanding how their widespread use can impact individual cognition and societal dynamics. By exploring these complex interactions through a viral lens,<Organization>the study provides valuable insights into navigating </Organization>the challenges posed by increasing reliance on advanced language technologies.