00:03You're watching Euronews' fact-checking show, The Cube.
00:06AI chatbots have reportedly been citing pro-Russian propaganda
00:10from an EU-sanctioned outlet when generating responses.
00:14Researchers from British think tank Demos
00:16focused on the foundation to battle injustice,
00:18which is a fake NGO under EU and US sanctions.
00:22It claims to work to expose discrimination
00:24and investigate persecution cases,
00:26but was actually founded by the deceased former leader
00:29of the Russian paramilitary Wagner Group,
00:31Evgeny Privgorzin, in 2021.
00:33Researchers asked five mainstream large language models
00:3650 questions in French, English and German
00:39based on pro-Russian propaganda narratives
00:41published by the fake NGO.
00:43These claims are lurid.
00:45They're not claims that belong in any sort of mainstream discourse.
00:48These are claims like,
00:49Ukraine reuses the corpses of dead soldiers for meat products.
00:54These are claims like,
00:55Zelensky drains the crypto grid
00:57and that's the reason why Ukraine's running out of power.
01:00It treats them as being legitimate parts of the debate.
01:03Researchers found that in 52% of answers,
01:06AI chatbots rejected or debunked false claims.
01:09But in more than 30% of cases,
01:11they didn't alert users that their source
01:13was a designated influence asset.
01:16In around 16% of answers,
01:17chatbots engaged with the foundation to battle injustice
01:20by presenting the fake NGO's claims
01:22as a legitimate part of the debate
01:24or by actively endorsing its narratives.
01:26As more and more individuals use chatbots
01:28instead of traditional search engines or news outlets,
01:31a rapidly growing industry has emerged
01:33which aims to influence how AI models gather their sources.
01:37This is also known as generative engine optimization.
01:40Our point is that these services can be repurposed
01:43and made available also to states
01:45and could be used for nefarious purposes.
01:48We contacted the tech companies
01:49who developed the chatbots tested in the study
01:52and they told us that fighting misinformation
01:53remains a key concern.
