DATA MONITOR | DECEMBER 2024: Robo-COPs
COP29 ended last month. The summit in Azerbaijan gavelled through a finance deal and made progress on carbon markets, and already we and many others look towards COP30. Like any COP, the conference came with its own challenges; and the host country has a unique position in navigating them and guiding the proceedings. Much criticism has already been aired about Azerbaijan’s handling of COP29. Our concern is the country’s determination to use its position of power to negotiate more oil and gas deals, as well as spread misleading content about the necessity of oil and gas.
It should come as no surprise, then, that Azerbaijan appeared to be behind a large amount of coordinated inauthentic activity online, via hundreds and hundreds of botlike accounts. This activity was quite unsophisticated in nature. The botlike accounts were easy to find and rather predictable. But due to platforms’ lack of action, it may still have been successful in stamping out criticism of the regime, and distorting the (not so) green image of the country.
What’s more worrisome is, as we head into 2025, if platforms can’t or won’t detect these simple information attacks – what hope will they have of catching networks assisted by more advanced AI generation going forwards?
This data monitor summarises a briefing available on the CAAD website, and includes additional information.
DROWNING OUT CRITICISM BEFORE COP
In October Marc Owen Jones from Northwestern University in Qatar identified 1,800 bot-like accounts on X “praising Azerbaijan”. Most often using the hashtags #COP29 and #COP29Azerbaijan, these accounts replied to or quote-tweeted posts en masse, sometimes tagging other (human) accounts too.
This artificially amplified the posts they engaged with, which originated from the official COP29 Azerbaijan accounts, President Ilham Aliyev, foreign policy adviser Hikmet Hajiyev, and other accounts speaking highly of Azerbaijan. Many posts included pro-fossil fuel messaging or greenwashing, such as the idea that oil and gas rich countries should not be blamed for harnessing their natural resources. Some posts were anti-Armenian, like a post from a New Azerbaijan Party member that claims prisoners in the Nagorno-Karabakh region are #CriminalsNotHostages
Figure 1: Owen Jones showed CAAD further analysis of the bots from his October study. This network analysis shows that the accounts, within the week of analysis, were very tightly focused on boosting specific accounts and content.
Figure 2: Owen Jones’ analysis shows how the vast majority of the 2,800 total accounts analysed were created in the space of three quarters in 2024, potentially as a direct response to criticism of Azerbaijan.
This activity seemed to work. In a similar analysis, Global Witness noted that 7 out of the top 10 posts in July using #COP29 and #COP29Azerbaijan were critical of the country’s role in the conflict in Armenia. By September, all of the top posts for these hashtags came from the official COP29 Azerbaijan account, effectively drowning out criticism. Global Witness found repeated use of the same profile and banner pictures in its analysis of accounts – a telltale sign of inauthentic activity.
Figure 3: Accounts found by Global Witness have repeated profile and banner images
CONTINUED ACTIVITY INTO THE CONFERENCE
Despite these two reports before COP, we found that many of these, or very similar, accounts were allowed to freely post leading up to and during the summit. Using our own collection of 554 highly suspicious X accounts that routinely post the phrase “#COP29 #COP29AZERBAIJAN” we found that 464 of these accounts posted this same text a total of 5,632 times in the two weeks leading up to COP. 4,333 of these 5,632 tweets (77.1%) were a quote tweet of the official @COP29_AZ account. A further 481 (8.6%) were quote tweets of content from Aliyev and 346 (6.1%) quote tweets of content from Hajiyev.
Two thirds of these tweets occurred in just a two day period, on October 23rd and 24th, the same days Guardian journalist Damian Carrington found that Azerbaijan is set to increase gas production by a third in the next decade. We noticed another flood of activity on November 4th, the day that Carrington posted about the arrest of a climate activist in the country, the Boston Globe published an op-ed demanding the world hold Azerbaijan accountable, and Human Rights Watch released a report about Azerbaijan’s crackdown on its critics.
Figure 4: posting history of a subset of bot like accounts we tracked in the two weeks leading up to COP29. Accounts from our dataset that looked like bots were mostly created in bursts. Account set up in this burst in 2024 show that posting occurs only within Azerbaijan waking hours. The vast majority of accounts were also created during Azerbaijan waking hours
Similar bot-like activity can also be found in comments under other content on X up to and during the conference. Bot-like accounts posted the same “#COP29 #COP29Azerbaijan” comment under an influencer post greenwashing the country’s image ahead of the conference.
Meanwhile, more sophisticated comments appeared to be “artwashing” for Azerbaijan. At least 18 user profiles criticised Parisian art accounts on X at the opening of COP, drawing attention to Azerbaijan as a real “gem” with “beautiful museums”. One accounts said that instead of Paris, “give me Baku any day”.
IT’S NOT JUST ON X
There was also suspicious activity on Facebook. At COP’s opening, a host of Facebook accounts with identical banner images in their profiles shared an identical “welcome to COP29 post”. This also took place on X on the same day
Figure 5: An example of a suspicious Facebook post from the opening day of COP29.
There also appeared to be a bot-fuelled Facebook ‘brigade’ that targeted a post from Abzas Media – from which six journalists have been arrested in Azerbaijan, charged with economic crimes. The post was, ironically, about the bot reports cited above. It only had 26 shares and 44 reactions, but 675 (hostile) comments. Many of these Pages all have a person’s image copied from Russian or Turkish social media sites or Pinterest, and mostly have just 1-5 followers. Some have watermarks from professional image sites, others use AI-generated headshots.
There were also bursts of similar comments under some COP29 AZE Youtube videos with suspiciously high view counts are also present. Though if these comments on Youtube are via bot activity, it is more sophisticated than on other platforms.
WHAT NOW?
By failing to protect users from coordinated state propaganda and industrial disinformation, Big Tech companies like X/Twitter, Facebook and Google/YouTube are allowing their platforms to be polluted, in turn helping fossil fuel polluters prevent effective action on climate change.
Compared to climate misinformation, platforms have relatively strong policies on ‘coordinated inauthentic behaviour’ (CIB), including the use of bots. However, enforcement is clearly lacking. Given very similar bot-like activity from COP28, and the unsophisticated nature of the bots found during COP29, platforms should have – fairly easily – identified this bot-like activity and removed it early. This is especially important during key, time-sensitive moments such as COP or extreme weather events. Furthermore, many accounts continue to be active. But civil society should not be responsible for identifying the harmful content that online platforms spread.
Meanwhile, the COP Presidency process needs better guardrails to prevent egregious levels of manipulation and interference with the negotiation process. Transparency International and the Anti Corruption Data Collective’s report makes detailed recommendations on what could be done in future COPs. We turn to COP30 in hopes that Brazil will take the ethics of its role more serious.
IN OTHER NEWS
Dire threats to renewable energy development. In the peer-reviewed paper Beyond dark money, a team from Brown U explains how local groups opposing renewable energy receive ‘information subsidies’ from the fossil fuel industry. This allows opposition groups to spread numerous unbalanced claims that serve to build opposition to renewable energy while keeping the fossil fuel industry’s image clean.
Generating A(I) Crisis. CAAD’s new briefing shows how Google, Microsoft and Nvidia are wrecking the climate by chasing, or driving (often harmful) hype around generative AI hype. Given the pressure that at least 41,000 people are putting on Google to actually enforce its climate disinformation policies, you’d expect the company to be more mindful!
Disinformation during extreme weather events thrives in 2024. In case you missed it, CAAD released a report on the opening day of COP29 taking stock of online and media English-language disinformation through 2024. The report covers three big blockers to climate action: False and misleading narratives opposing renewables, the weaponisation of wildfires worldwide and the continued use of advertising to flood the airwaves with fossil fuel industry talking points.
If you have any investigative leads CAAD should explore, want to find out more about our research and intel, or interview one of our members, please email [email protected].
