Chinese operatives used OpenAI models for three documented influence operations targeting US infrastructure and Japanese politics

{"The Atlantic reported OpenAI detected users tied to a Chinese provincial contractor who prompted ChatGPT to produce English-language comics framing AI energy demands as excessive, then distribute them on X. Separate queries sought help planning fake-resident email complaints and social amplification against Takaichi's immigration stance. OpenAI refused the Japan operation but the data-center content reached limited circulation before accounts were banned.","This fits a pattern where Chinese state actors rapidly adopt foreign commercial AI for narrative control, as they did with mobile phones and WeChat. Kenton Thibaut of the Atlantic Council noted AI now assists both campaign planning and execution at scale. Primary evidence comes from OpenAI's misuse logs, corroborated by embassy denials and Republican claims linking domestic data-center opposition to foreign efforts.","The institutional driver is the mismatch between US AI firms' commercial incentives to maximize API usage and Beijing's centralized information-warfare apparatus. Open-source models accelerate this by removing even the partial guardrails present in Western APIs. The Atlantic piece understates how logged refusals themselves become training signals for future evasion.","Forward, expect Chinese campaigns to shift toward fine-tuned domestic models once open-source weights proliferate, reducing detectable Western fingerprints. Detection thresholds will rise as volume increases."}