Artificial intelligence is increasingly becoming part of the toolkit used in online influence operations, not necessarily because it can invent entirely new propaganda strategies, but because it can make existing campaigns faster, more polished and harder to identify at first glance.
OpenAI has identified a Russia-linked influence operation in which operators reportedly used ChatGPT to help make their activity appear less Russian and more natural to the audiences they were trying to reach. The case highlights a growing challenge for technology companies: generative AI can be used not only to create content, but also to remove linguistic and cultural clues that might otherwise reveal where coordinated messaging originates.
AI as a Tool for Localization
One of the most significant aspects of the operation was the apparent use of ChatGPT to improve the presentation of material intended for foreign audiences. Rather than relying on awkward translations or language patterns that could expose a Russian origin, operators could ask an AI system to rewrite, translate or adapt text so that it sounded more natural in another linguistic environment.
This kind of use illustrates why generative AI matters to influence operations even when the underlying political narratives are not created by the model. A campaign can begin with a predetermined message and then use AI as an editing and localization layer, improving grammar, tone and cultural fluency while reducing obvious signs of foreign authorship.
Influence Operations Are Becoming More Efficient
Generative AI can dramatically lower the cost of producing variations of the same message. Operators can adapt posts for different platforms, audiences and languages without maintaining large teams of translators or copywriters. They can also rapidly test different tones, from conversational social-media posts to more formal commentary.
That does not automatically make an influence campaign successful. Distribution, credibility, audience engagement and the ability to build authentic-looking communities remain difficult problems. But AI can remove some of the friction involved in content production, giving coordinated networks the ability to operate at greater speed and scale.
OpenAI’s Disruption Efforts
The discovery also demonstrates how AI providers are increasingly monitoring their platforms for coordinated abuse. OpenAI has previously described efforts to identify and disrupt actors using its models in covert influence activity, including attempts to generate articles, social-media posts, comments, translations and other material designed for online distribution.
The broader security challenge is that many of the individual tasks involved are ordinary uses of a language model. Translation, rewriting and editing are legitimate activities used by millions of people. Detecting abuse therefore requires more than simply examining a single prompt. Providers must look for behavioral patterns, account relationships and other signals that can indicate coordinated activity.
A New Layer in the Disinformation Problem
The Russia-linked case points to a shift in the information environment. In earlier online influence campaigns, grammatical mistakes, unnatural phrasing or inconsistent cultural references could provide useful clues for researchers and users. Modern language models can weaken those signals by producing fluent text tailored to a target audience.
This means identifying coordinated influence will increasingly depend on factors beyond writing style. Researchers and platforms may need to focus more heavily on account behavior, synchronized posting, amplification networks, provenance and the relationships between seemingly independent sources.
AI-generated or AI-edited content should also not automatically be treated as disinformation. The technology is widely used for legitimate communication, translation and productivity. The relevant issue is the intent and coordination behind its use, particularly when operators conceal their identity or affiliations while attempting to manipulate public debate.
The Larger Lesson for AI Platforms
The incident underscores the dual-use nature of generative AI. The same capabilities that help businesses communicate internationally or help individuals write more clearly can also help covert operators disguise linguistic origins and produce more convincing material.
For AI companies, the challenge will be to disrupt malicious networks without broadly restricting legitimate users who rely on the same writing and translation features. For governments, researchers and social platforms, the task is similarly complex: detecting manipulation will require stronger coordination and better behavioral analysis rather than relying only on whether a piece of text appears to have been generated by AI.
As generative models become more capable, influence operations are likely to continue experimenting with them. The key question is no longer whether AI will be used in such campaigns, but how effectively platforms can identify coordinated abuse before improved language and localization translate into meaningful influence.