OpenAI Disrupts Russia-Origin Influence Operation Using ChatGPT for Promotion

OpenAI details a newly identified influence operation

OpenAI published a security case study on August 25, 2026 describing a covert influence operation that it says was run through a cluster of ChatGPT accounts that very likely originated in Russia.

According to OpenAI, the operators used ChatGPT mainly to generate promotional social-media posts and replies for a self-described ‘expert community’ called the International Burke Institute (IBI). OpenAI says it banned the accounts after investigating the activity.

The attribution and operational details in this article are based on OpenAI's own investigation. They should be treated as the company's assessment rather than an independently adjudicated finding.

AI was a supporting tool, not the source of the whole operation

A key detail is what ChatGPT did—and did not do.

OpenAI says the operators used ChatGPT to produce social posts distributed on X, LinkedIn, Facebook, Substack and Telegram, often asking the model to conceal linguistic clues that the users were Russian.

OpenAI does not say its models generated the articles hosted on the IBI website. Its review instead found that many of those articles appeared to have been copied from real academic work, sometimes with false attribution, while other content appeared to have been machine translated.

That distinction matters for understanding AI-enabled influence activity: the model was used as one component in a larger information operation that also depended on websites, identity presentation, republished content and social distribution.

A fabricated layer of apparent authority

OpenAI says the IBI website presented itself as an Israel-based expert community and listed academics and experts to create the appearance of legitimacy.

In a sample of 36 articles linked to experts on the site and published between September 2025 and May 2026, OpenAI says 34 appeared to have been copied from elsewhere on the internet. Some were allegedly attributed to the wrong authors.

The operation also promoted a proprietary-looking ‘sovereignty index’ that OpenAI says consistently framed Russia favorably while criticizing Western countries and institutions.

OpenAI describes the campaign as previously unreported and says it is the most elaborate Russia-linked influence operation the company has disrupted to date in terms of the infrastructure built around a purported expert organization.

Reach appears to have been limited

OpenAI's report does not describe a highly successful mass influence campaign.

The company says typical social posts had low view counts and official IBI accounts had few subscribers. Some related Telegram channels reportedly had roughly 10,000 to 20,000 subscribers.

Using the Brookings Breakout Scale, OpenAI assessed the activity at the lower end of Category Three: present across multiple platforms with some indications of reaching authentic audiences.

The significance, in OpenAI's assessment, is therefore less about measured reach and more about the infrastructure: a credible-looking institution, purported experts, republished academic material and a proprietary-looking index that could support larger influence efforts over time.

Why this matters for AI security

The case illustrates an increasingly important pattern in AI misuse research: generative AI does not need to autonomously create an entire campaign to make an operation more efficient.

AI can assist with multilingual copy, persona presentation, replies, promotional content and rapid variation while traditional infrastructure supplies the website, distribution channels and apparent institutional credibility.

It also shows why platform investigations can reveal activity outside the AI service itself. OpenAI says the trail from generated promotional posts helped expose a broader operation whose core website content was not generated by its models.

What OpenAI actually did

OpenAI says it banned the cluster of ChatGPT accounts associated with the operation. The company did not claim that every external account, channel or website connected to the broader activity was removed from the internet.

The report is a disruption and transparency disclosure, not evidence that AI-enabled influence operations as a category have been eliminated.

For defenders, researchers and platform teams, the useful takeaway is the cross-platform pattern: AI-generated promotional material can be only one observable layer of a larger operation designed to manufacture authority and obscure provenance.