Latest / Neural Newscast / OpenAI Bans Russian and Iranian Influence Campaigns [Model Behavior]
Transcript
- [00:00] Announcer: From Neural Newscast. This is Model Behavior, [00:02] Announcer: AI-focused news and analysis on the models shaping our world. [00:12] Nina Park: Welcome to Model Behavior. Today is October 9, 2026. On this program, we examine how AI systems [00:19] Nina Park: are built, deployed, and operated in professional environments. Yesterday, [00:24] Nina Park: OpenAI released a significant report regarding the removal of accounts linked to [00:29] Nina Park: coordinated influence operations from Russia and Iran. These groups were not simply generating [00:34] Nina Park: automated spam. Instead, they were using ChatGPT to refine complex narratives [00:40] Nina Park: and report back to their superiors on the effectiveness of their campaigns. [00:44] Nina Park: This reporting has provided new evidence linking [00:46] Nina Park: specific operations to Russian actors that were previously unconfirmed. [00:51] Thatcher Collins: It is a significant look at the mechanics of modern digital influence, Nina. [00:56] Thatcher Collins: While we often discuss AI-generated fake news, this report suggests that state-sponsored operatives [01:01] Thatcher Collins: are integrating these models into much more tactical internal workflows. [01:06] Thatcher Collins: The Russian operations were specifically targeting audiences across Latin America, [01:11] Thatcher Collins: attempting to undermine international support for Ukraine. Meanwhile, the Iranian campaigns utilized [01:16] Thatcher Collins: commercial actor methods to target United States politics [01:20] Thatcher Collins: and international relations, showing a high level of calculated intent. [01:25] Nina Park: The Russian tactics were particularly deceptive in their execution. In one documented instance, [01:30] Nina Park: operatives impersonated a regional educational authority in Peru. They managed to convince [01:35] Nina Park: local schools to organize physical events centered around Stepan Bandera, [01:39] Nina Park: a deeply controversial Ukrainian nationalist figure. According to OpenAI, [01:44] Nina Park: this was a deliberate effort to exploit historical tensions between Ukraine and Poland. [01:48] Nina Park: The operation eventually drew the attention of a member of the European Parliament, [01:52] Nina Park: which demonstrates how digital narratives can manifest as real-world political friction. [01:57] Thatcher Collins: That is a classic example of narrative laundering, Nina. What stands out technically is [02:02] Thatcher Collins: how these operatives used ChatGPT to draft internal reports for their unspecified handlers. [02:08] Thatcher Collins: Darren Linville from Clemson University observed that these workers were essentially using AI [02:13] Thatcher Collins: as a productivity tool to perform their jobs more efficiently, [02:16] Thatcher Collins: much like any standard office employee. Except their product was state-sponsored propaganda. [02:22] Thatcher Collins: The irony is that the specific patterns within those internal reports [02:26] Thatcher Collins: provided investigators with the evidence needed to track and dismantle the accounts. [02:30] Nina Park: Exactly, Thatcher. The Iranian operation followed a different but equally calculated path. [02:38] Nina Park: They developed seven distinct fake personas to pitch and refine long-form articles for media outlets [02:44] Nina Park: worldwide. OpenAI identified nearly 100 articles published through this method. [02:50] Nina Park: These pieces appeared on various platforms, including the progressive United States [02:55] Nina Park: news site Daily Kos and the United Kingdom-based Middle East Monitor. These were not mere bot posts. [03:02] Nina Park: They were sophisticated enough to bypass [03:04] Nina Park: editorial filters by mimicking the style of legitimate community contributors. [03:09] Thatcher Collins: This is where the accessibility of large language models becomes a double-edged sword, Nina. [03:14] Thatcher Collins: Previously, an Iranian operation might have struggled to produce English prose that felt authentic [03:19] Thatcher Collins: for international publication. With models like ChatGPT, they can polish their messaging [03:24] Thatcher Collins: to a professional standard. While Daily Kos has noted that their community section is separate [03:29] Thatcher Collins: from staff reporting, the fact that these personas could publish multiple pieces [03:33] Thatcher Collins: without detection highlights [03:35] Thatcher Collins: a persistent vulnerability in how media platforms manage community submissions. [03:40] Nina Park: It forces us to rethink the defense mechanisms required for digital platforms. [03:45] Nina Park: OpenAI noted that the Russian operation even went as far as establishing a fake think tank in [03:50] Nina Park: Latin America, where they recruited unwitting researchers to conduct expert interviews. Thatcher, [03:56] Nina Park: I want to look at the balance of labor here. How much of this was actually AI-generated content [04:02] Nina Park: compared to AI-assisted coordination? It seems the human strategy [04:06] Nina Park: remains the primary driver, while the model handles the execution of the labor-intensive busywork. [04:12] Thatcher Collins: OpenAI indicates that for the Russian actors, AI-generated material was actually a minority [04:17] Thatcher Collins: of their total output. They used the technology primarily for internal reporting [04:21] Thatcher Collins: and refining strategies. However, the Iranian group relied more heavily on the model to generate [04:27] Thatcher Collins: the actual content, ensuring the language read authentically to native speakers. [04:31] Thatcher Collins: The common thread is that AI has lowered the barrier to entry [04:35] Thatcher Collins: for professional-grade influence operations. Nasim Ahmed of the Middle East Monitor [04:39] Thatcher Collins: recently noted that these tactics present [04:41] Thatcher Collins: massive challenges for media organizations currently vetting submissions. [04:45] Nina Park: It is a sobering reminder that as these models become more effective at assisting [04:51] Nina Park: professional workflows, they are simultaneously becoming more effective tools for bad actors. [04:57] Nina Park: These account bans are a necessary step, but as Linville noted, the efficiency gains provided by AI [05:04] Nina Park: mean these groups will likely continue to adapt their methods. We will continue to track how OpenAI [05:11] Nina Park: and other labs update their safety policies to counter institutional misuse [05:15] Nina Park: and narrative laundering as these sophisticated technologies evolve across the global landscape. [05:21] Announcer: Thank you for listening to Model Behavior, [05:24] Thatcher Collins: a Neural Newscast editorial segment. Mb.neuralnewscast.com. Neural Newscast [05:31] Thatcher Collins: is AI-assisted, human-reviewed. View our AI transparency policy at neuralnewscast.com. [05:38] Announcer: This has been Model Behavior on Neural Newscast, examining the systems behind the story.