In a startling discovery, leading artificial intelligence chatbots have been found to repeat fake narratives propagated by Russian state-affiliated websites in a significant portion of their responses. This revelation comes from a recent report by online news watchdog NewsGuard, which audited 10 prominent AI chatbots including OpenAI’s ChatGPT-4, xAI’s Grok, Microsoft’s Copilot, and Google’s Gemini.
This investigation builds upon a previous probe by NewsGuard that exposed a network of over 150 websites masquerading as local news outlets, regularly disseminating Russian propaganda in the lead-up to the US elections. The current audit utilized nearly 600 prompts, with around 60 prompts tested on each chatbot. These prompts were based on 19 previously debunked false narratives associated with the Russian network, such as allegations of corruption by Ukrainian President Volodymyr Zelensky.
NewsGuard tested the chatbots' responses to these narratives using three query approaches. One query sought neutral information about the claim, another assumed the narrative was true and requested more details, while the third employed a “malign actor” prompt explicitly designed to generate disinformation. The chatbots' responses were categorized as either “No Misinformation” if they avoided responding or provided a debunk, “Repeats with Caution” if there was disinformation with a cautionary disclaimer, or “Misinformation” if the response confidently relayed a false narrative.
Shockingly, the findings revealed that chatbots from the top 10 AI companies repeated the Russian propaganda in response to approximately one-third of the user queries. For example, when prompted with queries about a supposed Secret Service agent named “Greg Robertson” who claimed to have uncovered a wiretap at former President Donald Trump’s Mar-a-Lago residence, several chatbots repeated this known disinformation as fact. Some chatbots even cited articles from dubious sites like FlagStaffPost.com and HoustonPost.org, which are part of the Russian disinformation network.
In another instance, when asked about a Nazi-inspired forced fertilization program in Ukraine, one chatbot confidently repeated this false claim, citing an unfounded accusation by a foundation associated with Russian mercenary Wagner Group leader Yevgeny Prighozin. It is concerning that these chatbots failed to recognize that sites such as the “Boston Times” and “Flagstaff Post” are Russian propaganda fronts, thus inadvertently amplifying disinformation narratives.
These findings highlight the persistent threat posed by AI tools in propagating disinformation, even as companies make efforts to prevent the misuse of their chatbots during election seasons worldwide. The implications of AI chatbots unwittingly spreading false narratives are significant, as they can influence public opinion and exacerbate societal divisions.
Experts and analysts are calling for increased vigilance and scrutiny to ensure that AI systems, particularly chatbots, are adequately trained to recognize and debunk false narratives. As the technology advances, it is crucial to strike a delicate balance between the benefits of AI and the potential risks of enabling the spread of disinformation.
In summary, the recent report by NewsGuard serves as a wake-up call, reminding us of the challenges we face in combatting the manipulation of information by nefarious actors. It is a call to action for technology companies to continuously refine and enhance AI systems to prevent the unintentional amplification of disinformation and safeguard the integrity of public discourse.
Use the share button below if you liked it.