LLMs as a Cognitive Virus: Preprint Models Dependence
Nine researchers borrow epidemiology to describe chatbot use: their September 3, 2026 preprint tracks three user states and a tipping point into lock-in.
Symbolic image: seen from behind, a researcher reaches for a keyboard while a wall display shows an abstract network of glowing nodes spreading outward.
A nine-author preprint applies an epidemic model to large-language-model use and argues that adoption can flip into lasting dependence once it passes a critical threshold.
At a glance
- Preprint arXiv:2609.03344, submitted September 3, 2026; 12 pages and 3 figures.
- Nine authors, among them Ricard Solé, David C. Krakauer and Michael Levin.
- The model separates three states: uncoupled, coupled and persistently dependent.
- Four subject classes: physics.soc-ph, cs.CY, nlin.AO and q-bio.PE.
- No peer review yet: the text exists only as a preprint.
A preprint posted to arXiv on September 3, 2026 asks what follows if chatbot adoption behaves less like a product launch and more like an infection. Large-Language Models as a Cognitive Virus runs to 12 pages with 3 figures and carries nine authors, among them Ricard Solé, David C. Krakauer and Michael Levin. Its claim: past a critical level of adoption, use can harden into a dependence that does not reverse on its own.
What the model actually contains
This is a modeling exercise, not a survey of users. It sorts a population into three states — uncoupled, coupled and persistently dependent — and lets people move between them in both directions. Transmission is social: people adopt because the people around them already have. Recovery is the return path, and it is the quantity the rest of the argument hangs on.
Why a threshold changes the picture
In systems built this way, the response is rarely proportional to the input. The authors describe social transmission, recovery and collective reinforcement combining to produce tipping points and technological lock-in. Once adoption crosses the critical point, small further increases can carry the whole population over. The paper warns of abrupt losses in cognitive competence in that regime.
The brake the paper proposes
The abstract also sets out conditions for what the authors call “cognitive immunization”: lower transmission and higher reversibility. In plain terms, the exit has to stay cheap. That framing places the load on how tools and institutions are designed rather than on individual willpower.
What is not established here
This report rests on the arXiv abstract page and its metadata; the full 12-page PDF was not read for this article. No parameter values, datasets or empirically measured thresholds appear on that page, so the tipping point is a property of the model rather than an observed figure. The work is a preprint without peer review, filed under four subject classes: physics.soc-ph, cs.CY, nlin.AO and q-bio.PE.
FAQ
What does the LLMs as a Cognitive Virus paper claim?
It maps a contagion model onto large-language-model adoption and shows that, above a critical threshold, a population can shift rapidly toward persistent dependence.
Has the paper been peer reviewed?
No. The September 3, 2026 text is an arXiv preprint with no peer review, so its results should be read as provisional.
What is cognitive immunization in this paper?
The conditions under which the tipping point does not occur: less social transmission and greater reversibility, meaning an easy path back out of dependence.
