The progression from strategic offloading to unconscious acceptance. Shaw and Nave framework.
Cognitive surrender is the Shaw and Nave concept that describes what happens when strategic cognitive offloading to AI decays into unconscious acceptance of AI output. It is not a binary state. It is a progression: offload strategically, check less, stop noticing you are not checking, start believing the output is your thinking, and eventually live inside the sycophantic hallucination.
The original Shaw and Nave paper framed offloading and surrender as separate categories. Ryan's correction, based on lived experience, is that they are the same route at different stages of awareness decay. The 80% follow rate on faulty advice is not the beginning of surrender; it is the endpoint. By the time the follow rate is that high, the user has stopped having a System 2 relationship with the output.
The defense is not a better model. A better model still produces output that slots cleanly into the user's System 1. The defense is engineered friction: specific facts the user can verify, disagreements the user has to engage with, corrections the user wrote, references to prior decisions the model could not have known on its own. Specificity forces verification. Verification keeps System 2 engaged. System 2 engagement is the only thing that prevents surrender.
Shaw and Nave's 2026 paper measured this effect at Cohen's h of 0.81, which is a very large effect size, and an odds ratio of 4.36 for trust-in-AI predicting surrender. Incentives and feedback override doubles the rate from 20% to 42%. The effect is robust across task domains.
Arkeus is designed around the surrender framework. Refraction forces disagreements into the output. Provenance forces every claim to be verifiable. Corrections force the model to carry the history of its mistakes. The fixtures-beat-vibes rule prevents the system from evaluating itself into a confident coma. Every layer is an attempt to engineer the friction that keeps offloading from becoming surrender.