³Ô¹ÏºÚÁÏÍø

On 30 January 2024, for a brief period, the internet went down in Russia. All major Russia-based servers went out of service at about 8pm Moscow time and it took approximately two hours to solve the problem, with the outage declared over by 10pm.

This was no doubt inconvenient for ordinary Russians. It may also have been a problem for Russian bot farms, which under Kremlin instruction are tasked with sowing disinformation among Russia’s enemies via social media. Bot farms are likely to use Russian local infrastructure to coordinate their actions, receive content and execute instructions. This infrastructure was not available during the outage period.

The outage was therefore an ideal opportunity to obtain a snapshot of the activities of bot farms by examining what they were unable to do during the evening of 30 January. We assessed the impact on what we call ‘goblins’, as distinct from trolls. While trolls set out to derail online discussions, the algorithmic nature of modern social media – in which algorithms serve content to users based on analysis of their interests – creates huge opportunities for a new type of propaganda agent.

Goblins seldom post anything. Instead, they specialise in boosting into people’s feeds human-curated propaganda, disinformation or politically useful accounts. They ‘play’ on algorithms to circulate content to the largest numbers of users. Goblin armies can seriously boost or de-boost content, swarming users’ timelines. With the advent of more advanced chatbots like ChatGPT, automated goblin accounts have gained basic interactive capabilities, further increasing their usefulness for engagement.

So what happened during the internet outage? We tested two groups of X (formerly Twitter) accounts: a treatment group and a control group. The former comprised accounts likely to be reposted by Russian goblin bots: people and organisations that, for example, deny Russian war crimes in Ukraine, portray parts of the Ukrainian government and army as Nazis or Nazi-sympathizers, portray NATO as the immediate cause of the war in Ukraine, or lobby for a reduction in Western help for Ukraine. We fixed on a group of 221 accounts including influencers Iain Miles Cheong and Jackson Hinkle, Hungarian prime minister Viktor Orbán and an account that reposts Donald Trump’s Truth Social posts.

Our control group meanwhile contained accounts highly unlikely to be boosted by Russian bots, including accounts associated with pro-Ukrainian social-media movement NAFO (North Atlantic Fella Organization), political commentators such as Igor Sushko, and politicians such as Republican congressman Dan Crenshaw and Finnish President Alexander Stubb.

Comparing the posting patterns of the two groups, and also comparing the period of the outage on 30 January with a control period on 29 January, we found a notable drop-off during the outage in reposts of treatment-group tweets. Specifically, during the peak of the outage, from 9pm to 10pm Moscow time, the average effect amounted to 49 fewer reposts for a treatment-group account compared to the counterfactual.

This might not sound like a huge number, but the drop-off must be considered in the context of the short duration of the outage, the time zone (it was a Tuesday morning in the US) and the relatively limited number of accounts we analysed. On average, in ‘normal’ times, posts from accounts in our dataset are reposted six times in the two hours after posting.

During the two-hour outage, however, more than 70% of posts received no reposts at all. Hence, the drop of 49 reposts per treatment-group account during the peak hour is a notable change.

These results are still preliminary, and more research will be needed to assess the true magnitude of these effects and better understand the numbers. Nevertheless, our admittedly limited experiment indicates that Russian goblins may be playing a substantial role in ensuring that pro-Russia messages get seen much more than they might do otherwise.

Overall, the outage provides new evidence of the Russian state’s use of bot networks to influence public discourse on Western social media platforms. This use of goblins may have played roles in several elections and referendums, including Moldova’s very narrowly won vote on 20 October on changing its constitution to favour European Union membership. And of course, with the US presidential election coming up, goblins are likely to be ramping up their biased boosting activities even more.

This First Glance was also published in in Italian.