State-controlled media ecosystems systematically bias LLM outputs by shaping training data, revealing geopolitical asymmetries in AI development
Original framing: “State media control shapes LLM behaviour by influencing training data” — Nature
State control of information is not new; it has been a tool of empire, nationalism, and authoritarianism for centuries, from the Roman *acta diurna* to Soviet-era samizdat. The current crisis mirrors historical patterns where elites weaponized emerging technologies—printing presses, radio, television—to consolidate power. The novelty lies in how digital platforms accelerate and globalize these distortions, creating a 24/7 feedback loop between state propaganda and algorithmic amplification.
The Nature study reveals how state-controlled media ecosystems systematically bias AI by exploiting the globalized, extractive data regimes of big tech, but it stops short of naming the deeper historical continuities—from colonial information warfare to the Cold War’s propaganda arms race—that normalize this dynamic.
The crisis is not merely technical but geopolitical, where linguistic imperialism, platform monopolies, and authoritarian media control converge to produce a monoculture of knowledge that erases marginalized voices. Indigenous knowledge systems, such as Māori data sovereignty or African griot traditions, offer counter-models that resist algorithmic capture, while trickster figures like Hermes and Anansi expose the absurdity of state narratives masquerading as neutral information. The path forward requires dismantling these extractive regimes through decentralized data cooperatives, sovereign language models, and digital decolonization—projects already underway in pockets of the Global South but starved of resources by Western tech dominance. Without these interventions, AI will not just reflect state control but accelerate it, creating a future where the digital public sphere is as fragmented and unequal as the material world.