In August 1948, the Lenin All-Union Academy of Agricultural Sciences adopted Trofim Lysenko's pseudoscientific biology as Soviet state doctrine, effectively destroying genetics research in the USSR for sixteen years — not because the ideas survived scientific scrutiny, but because the state apparatus controlled which facts could enter the scientific record. China's Great Firewall now applies the same structural mechanism to AI: by filtering politically sensitive events, dissenting analysis, and independent reporting from the information environment, it produces language models trained on a curated Party narrative rather than an approximation of reality. Research published in Nature (Shumailov et al., July 2024) demonstrates that models trained on recursively generated synthetic data degrade over successive generations — 'model collapse' — with the only defense being a continuous influx of fresh, human-generated information, which the Firewall removes by design. The degradation compounds: each generation of Chinese AI trains on more filtered and more synthetic data, with no external correction signal — the same self-reinforcing information control loop that crippled Soviet science, now operating at machine speed.