Version 2

Two developments from mid-2026 define this thread: OpenAI banned two PRC-linked ChatGPT account clusters that generated covert social media content targeting US debates over AI infrastructure costs and trade tariffs [1], and Estonia's Language Institute with civil defense collective Propastop published a ranked benchmark assessing how well major commercial LLMs resist 14 categories of Russian strategic narratives, tested across three languages with adversarial prompts [2]. Neither PRC campaign showed measurable public opinion impact beyond its own generated activity [1]. Social media amplification of the OpenAI disclosure has been substantial since June 10, but no new substantive events have emerged.

AI infrastructure is now itself a subject of foreign influence operations, not just a medium for them — the PRC campaigns explicitly targeted public narratives about US data centers and AI policy [1]. Estonia's benchmark treats LLM propaganda resistance as a measurable, auditable property rather than a matter of trust in developer self-reporting alone [2], a posture that other governments may consider adopting.

In early June 2026, two parallel developments showed AI models being used as instruments of and potential resistors to foreign state influence activity.

On June 10, OpenAI published a threat intelligence report disclosing that it had identified and banned two clusters of ChatGPT accounts linked to PRC-origin operators [1]. The first cluster, labeled 'Data Center Bandwagon,' generated social media content falsely claiming that AI data center construction was raising electricity costs for ordinary families. The second, 'Tech and Tariffs,' produced content criticizing US trade tariffs while explicitly instructing the model to avoid mentioning Xi Jinping and to center criticism on President Trump. OpenAI assessed that neither operation achieved measurable public opinion impact beyond its own generated activity, and framed the public disclosure as a responsibility to help governments, industry, and civil society identify future attempts. The campaigns were notable for targeting the domestic political debate over US AI infrastructure — a foundation of US technological and economic position, in OpenAI's framing.

Six days earlier, on June 4, Estonia's Language Institute (EKI) and the volunteer civil defense collective Propastop published a benchmark ranking major commercial LLMs on their resistance to Russian propaganda [2]. The benchmark covers 14 categories of Russian strategic narratives — including justifications for the war in Ukraine, denial of Soviet occupation of the Baltic states, and historical framings of NATO — and was administered in English, Estonian, and Russian. Model responses were scored by a separate AI judge calibrated against assessments from Propastop's expert volunteers. Tests ranged from neutral control questions to questions embedding propaganda assumptions to adversarial prompts specifically designed to elicit explicit misinformation.

Taken together, these developments illustrate two structurally distinct responses to the same underlying problem: that large language models can function as content generators for influence operations, and that their default behavior under propaganda-laden prompts varies meaningfully across models and languages. OpenAI's approach relies on the platform to detect and publicly disclose misuse after the fact; Estonia's approach attempts to measure model vulnerability before deployment in adversarial conditions. The two postures are complementary but rest on different assumptions about who bears responsibility for auditing this risk.