Russia is reportedly engaged in a campaign to manipulate artificial intelligence models used by millions, according to a new report from the think tank Demos. This tactic, termed 'Retrieval Augmented Generation (RAG) poisoning,' involves altering the online material that AI systems rely on to generate answers, thereby spreading disinformation.
AI models from major companies including OpenAI, Anthropic, and Google have been found to retrieve false information from websites and social media platforms, amplifying misleading narratives. The report highlights instances where Russia has allegedly used this method to present groups that could be detrimental to Britain in a favorable light. As AI becomes an increasingly primary source of information, this manipulation poses a significant challenge to information integrity.
One specific case detailed in the report involved five advanced AI tools that favorably cited information from the Russian outlet Foundation to Battle Injustice (FBI). Despite the EU having sanctioned the FBI, which is described as a front for Russian state influence and disinformation campaigns, 31 percent of tested AI responses did not identify it as a sanctioned group, and only about half debunked its claims. This occurred after suspected RAG poisoning, the report found.
Shadow technology secretary Julia Lopez commented on the findings, stating that as AI becomes a primary information source, hostile states will inevitably attempt to manipulate these systems, similar to tactics seen on social media and search engines. She emphasized that the government cannot be complacent and must ensure AI companies implement robust safeguards to identify and label state-backed disinformation, while collaborating with allies to disrupt foreign information warfare.
AI minister Kanishka Narayan acknowledged that the public is right to be concerned about the risks associated with AI technology, rejecting notions of unmitigated positive outcomes. A government spokesperson added that state-backed actors are continuously evolving their disinformation tactics, including through AI models, and that the government is actively working to counter these methods. The spokesperson also noted that most AI systems are already regulated by existing expert bodies at the point of use, aiming to protect users and support innovation.