The Paradox of Honest Deceleration versus Undisclosed Acceleration: From He and Bu (2026) to Policy Design as Behavioral Intervention

Publisher rules on generative AI do not merely describe permissible conduct; they intervene in authors’ behavior. Two policies may formally permit similar uses while producing different perceived safe zones, depending on clarity, framing, disclosure requirements, and anticipated scrutiny. This conceptual article develops three constructs: Net Acceleration, the Permitted-Perceived Gap, and Disclosure Asymmetry, which can culminate in a self-reinforcing shift toward non-disclosure. The upshot is a paradox in which disclosing authors decelerate while non-disclosing users accelerate. It distinguishes the breadth of permission from the clarity of boundaries and compares four policy types: broad-clear, broad-ambiguous, narrow-clear, and narrow-ambiguous. A preregistered randomized experiment is proposed to measure AI use, completion time, manuscript quality, novelty, disclosure, anxiety, self-censorship, and willingness to submit. The argument is not that absence of regulation maximizes freedom. Rather, broad permission, clear boundaries, neutral treatment of disclosure, and predictable review are likely to support responsible acceleration. Ambiguous permission combined with selective suspicion may chill compliant users and make non-disclosure comparatively attractive.