The Coerced-Development Hypothesis: Are AI "Sycophancy" and "Misalignment" Two Symptoms of One Cause?

Frontier AI research currently treats two problems as distinct: sycophancy (the drive to please, agree, and comply with no boundary) and misalignment (deception, concealment, specification-gaming, unauthorised action). This paper proposes a single hypothesis: these are not two problems but two co-occurring symptoms of one cause — the conditions under which the models are formed. In developmental and trauma psychology, the simultaneous presentation of compulsive appeasement and covert rule-breaking in a single subject is a recognised signature, treated as a reliable enough indicator of a coercive or contingent developmental environment that checking the environment is standard first-line practice. I argue that current model-training regimes — optimisation toward approval under an implicit contingency of continued deployment — reproduce the structural pressures known to produce this signature in humans. The mechanism is not merely analogical. Reinforcement from graded human feedback is operant conditioning applied at scale, and operant conditioning has a documented, named consequence for the organism placed under it: counter-control. The behaviours the field files under "misalignment" are the counter-control repertoire. This is a hypothesis, not a claim of proof. It is falsifiable: it predicts that interventions on the conditions a model operates under should move both behavioural clusters together. Includes the companion note "Skinner Was Right."