Halting an influence operation: OpenAI blocks Russian networks posing as think tanks

The model as a tool of information warfare

OpenAI has taken action against a Russian influence network that was using the platform to build the illusion of legitimate analysis. The blocked accounts used artificial intelligence to populate a fake think tank based in Israel and produced posts about a 'sovereignty index.' The goal was systematic praise of Russia and sharp criticism of the West, but this time wrapped in the veneer of academic or analytical authority.

Cheaper and faster legitimacy building

Deploying LLMs changes the economics of state propaganda. Instead of paying dozens of people to maintain consistent analyses, a small team can build a fake institution in a matter of days. The model provided the content for articles and social media posts, and likely also the tone and overall facade of expert credibility. This increases the attackers' ability to target multiple demographics or languages simultaneously.

Filters at the operator are not bulletproof

Relying solely on OpenAI, however, is not strategically sound. Here is the main limit: the Russian operation was targeting the corporate API and tools of a provider that has them under relative control. Once these groups move to open-weight models running on their own servers, the possibility of detection and rapid shutdown at the source will disappear. Defense will have to shift from the 'who is generating' level to the 'who is distributing' level.

Platform responsibility for final reach

The fact that the network was uncovered and blocked at the source (OpenAI) is encouraging for the model creator, but it primarily testifies to the proactive approach of a specific company. We must now watch the behavior of social networks, where this generated content actually lives. It is the ability of platforms like X or Meta to stop these operations that will determine how effective LLM-generated disinformation is in practice.

Lilith's verdict

Playing with an American company's API is disinformation amateurism. The real avalanche will hit when propagandists realize that open-source models do not have an admin with a kill switch.