Harvard-led paper models LLMs as a "cognitive virus" with a tipping point for human cognition

Researchers from Harvard, the Santa Fe Institute and other top institutions—including Ricard Solé, David Krakauer and Michael Levin, nine authors in total—published a 12-page arXiv paper, "Large-Language Models as a Cognitive Virus," analogizing LLMs to a cognitive virus spreading through human populations and warning of a tipping point beyond which human cognitive abilities could decline abruptly.

Confirmed

• The paper appeared on arXiv around September 6, is 12 pages long, and its authors come from Harvard, the Santa Fe Institute and other leading institutions.
• Analyzing LLMs from evolutionary biology, complex systems and epidemiology perspectives, it argues that language models satisfy the definition of a virus both biologically and mathematically: they cannot self-replicate and rely on human cognition to spread.
• The core model treats societal cognitive dependence on AI as a feedback problem: as more people outsource reading, writing, reasoning and verification to LLMs, the habits that keep those skills sharp degrade, forming a positive feedback loop. Users are grouped by dependence level, with "cognitive immunity"-style parameters capturing the dynamics.
• The paper argues for a tipping-point effect: even a small rise in LLM adoption, once past that point, leads to persistent dependence and a sharp drop in cognitive ability.

Why it matters

• This is the first cross-institutional study to systematically model "AI cognitive decline" with epidemic-dynamics frameworks, turning the scattered intuition that "AI makes people dumber" into an analyzable mathematical model.
• The tipping-point conclusion implies cognitive harm could be nonlinear and irreversible, offering an early warning for AI education and policy debates.
• The "mind virus" framing sparked viral meme-style discussion—@zainhas jokingly called LLMs a mind virus and quipped he was probably already infected—reflecting the claim's high spread and controversy in the AI community.

2026-09-06 ~ 2026-09-07 · 7 related posts

• Harvard paper argues LLMs meet every definition of a 'cognitive virus' — rjurney · 2026-09-06
• [source] Harvard-Led Paper Warns Small LLM Adoption Rises Could Trigger Cognitive Dependence Tipping Point — rohanpaul_ai · 2026-09-06
• [source] Harvard-led arXiv paper models LLM adoption as a 'cognitive virus' with dependence tipping points — rohanpaul_ai · 2026-09-06
• "LLMs as the Mind Virus": AI Devs Joke About Abrupt Cognitive Decline Claims — zainhas · 2026-09-06
• [source] 'LLMs as a Cognitive Virus': Solé, Krakauer, Levin paper warns of dependence tipping points — PawelHuryn · 2026-09-07
• LLMs as a Cognitive Virus: New Paper Models AI Dependence Tipping Points — serrjoa · 2026-09-07

1 near-duplicate retellings: mikeflache