Retweeting through the Great Firewall
A persistent and undeterred threat actor
This report analyses a persistent, large-scale influence campaign linked to Chinese state actors on Twitter and Facebook.
This activity largely targeted Chinese-speaking audiences outside of the Chinese mainland (where Twitter is blocked) with the intention of influencing perceptions on key issues, including the Hong Kong protests, exiled Chinese billionaire Guo Wengui and, to a lesser extent Covid-19 and Taiwan.
Extrapolating from the takedown dataset, to which we had advanced access, given to us by Twitter, we have identified that this operation continues and has pivoted to try to weaponise the US Government’s response to current domestic protests and create the perception of a moral equivalence with the suppression of protests in Hong Kong.
Figure 1: Normalised topic distribution over time in the Twitter dataset
Our analysis includes a dataset of 23,750 Twitter accounts and 348,608 tweets that occurred from January 2018 to 17 April 2020 (Figure 1). Twitter has attributed this dataset to Chinese state-linked actors and has recently taken the accounts contained within it offline.
In addition to the Twitter dataset, we’ve also found dozens of Facebook accounts that we have high confidence form part of the same state-linked information operation. We’ve also independently discovered—and verified through Twitter—additional Twitter accounts that also form a part of this operation. This activity appears to be a continuation of the campaign targeting the Hong Kong protests, which ASPI’s International Cyber Policy Centre covered in the September 2019 report Tweeting through the Great Firewall and which had begun targeting critics of the Chinese regime in April 2017.
Analysing the dataset as a whole, we found that the posting patterns of tweets mapped cleanly to working hours at Beijing time (despite the fact that Twitter is blocked in mainland China). Posts spiked through 8 a.m.–5 p.m. working hours Monday to Friday and dropped off at weekends. Such a regimented posting pattern clearly suggests coordination and inauthenticity.
The main vector of dissemination was through images, many of which contained embedded Chinese-language text. The linguistic traits within the dataset suggest that audiences in Hong Kong were a primary target for this campaign, with the broader Chinese diaspora as a secondary audience.
There is little effort to cultivate rich, detailed personas that might be used to influence targeted networks; in fact, 78.5% of the accounts in Twitter’s takedown dataset have no followers at all.
There’s evidence that aged accounts—potentially purchased, hacked or stolen—are also a feature of the campaign. Here again, there’s little effort to disguise the incongruous nature of accounts (from Bangladesh, for example) posting propaganda inspired by the Chinese Communist Party (CCP). While the takedown dataset contains many new and low-follower accounts, the operation targeted the aged accounts as the mechanism by which the campaign might gain traction in high-follower networks.
The operation has shown remarkable persistence to stay online in various forms since 2017, and its tenacity has allowed for shifts in tactics and the narrative focus as emerging events—including the Covid-19 pandemic and US protests in May and June 2020—have been incorporated into pro-Chinese government narratives.
Based on the data in the takedown dataset, while these efforts are sufficiently technically sophisticated to persist, they currently lack the linguistic and cultural refinement to drive engagement on Twitter through high-follower networks, and thus far have had relatively low impact on the platform. The operation’s targeting of higher value aged accounts as vehicles for amplifying reach, potentially through the influence-for-hire marketplace, is likely to have been a strategy to obfuscate the campaign’s state-sponsorship. This suggests that the operators lacked the confidence, capability and credibility to develop high-value personas on the platform. This mode of operation highlights the emerging nexus between state-linked propaganda and the internet’s public relations shadow economy, which offers state actors opportunities for outsourcing their disinformation propagation.
Similar studies support our report’s findings. In addition to our own previous work Tweeting through the Great Firewall, Graphika has undertaken two studies of a persistent campaign targeting the Hong Kong protests, Guo Wengui and other critics of the Chinese Government. Bellingcat has also previously reported on networks targeting Guo Wengui and the Hong Kong protest movement.
Google’s Threat Analysis Group noted that it had removed more than a thousand YouTube channels that were behaving in a coordinated manner and sharing content that aligned with Graphika’s findings.
This large-scale pivot to Western platforms is relatively new, and we should expect continued evolution and improvement, given the enormous resourcing the Chinese party-state can bring to bear in aligning state messaging across its diplomacy, state media and covert influence operations. The coordination of diplomatic and state media messaging, the use of Western social media platforms to seed disinformation into international media coverage, the immediate mirroring and rebuttal of Western media coverage by Chinese state media, the co-option of fringe conspiracy media to target networks vulnerable to manipulation and the use of coordinated inauthentic networks and undeclared political ads to actively manipulate social media audiences have all been tactics deployed by the Chinese Government to attempt to shape the information environment to its advantage.
The disruption caused by Covid-19 has created a permissive environment for the CCP to experiment with overt manipulation of global social media audiences on Western platforms. There’s much to suggest that the CCP’s propaganda apparatus has been watching the tactics and impact of Russian disinformation.
The party-state’s online experiments will allow its propaganda apparatus to recalibrate efforts to influence audiences on Western platforms with growing precision. When combined with data acquisition, investments in artificial intelligence and alternative social media platforms, there is potential for the normalisation of a very different information environment from the open internet favoured by democratic societies.
This report is broken into three sections, which follow on from this brief explanation of the dataset, the context of Chinese party-state influence campaigns and the methodology. The first major section investigates the tactics, techniques and operational traits of the campaign. The second section analyses the narratives and nuances included in the campaign messaging. The third section is the appendix, which will allow interested readers to do a deep dive into the data.
ASPI’s International Cyber Policy Centre received the dataset from Twitter on 2 June and produced this report in 10 days.
The Chinese party-state and influence campaigns
The Chinese party-state has demonstrated its willingness to deploy disinformation and influence operations to achieve strategic goals. For example, the CCP has mobilised a long-running campaign of political warfare against Taiwan, incorporating the seeding of disinformation on digital platforms. And our September 2019 report—Tweeting through the Great Firewall—investigated state-linked information campaigns on Western social media platforms targeting the Hong Kong protests, Chinese dissidents and critics of the CCP regime.
Since Tweeting through the Great Firewall, we have observed a significant evolution in the CCP’s efforts to shape the information environment to its advantage, particularly through the manipulation of social media. Through 2018 and 2019 we observed spikes in the creation of Twitter accounts by Chinese Ministry of Foreign Affairs spokespeople, diplomats, embassies and state media.
To deflect attention from its early mishandling of a health and economic crisis that has now gone global, the CCP has unashamedly launched waves of disinformation and influence operations intermingled with diplomatic messaging. There are prominent and consistent themes across the messaging of People’s Republic of China (PRC) diplomats and state media: that the CCP’s model of social governance is one that can successfully manage crises, that the PRC’s economy is rapidly recovering from the period of lockdown, and that the PRC is a generous global citizen that can rapidly mobilise medical support and guide the world through the pandemic.
The trends in the PRC’s coordinated diplomatic and state-media messaging are articulated as a coherent strategy by the Chinese Academy of Social Sciences, which is a prominent PRC-based think tank. The academy has recommended a range of responses to Western, particularly US-based, media criticism of the CCP’s handling of the pandemic, which it suggests is designed to contain the PRC’s global relationships. The think tank has offered several strategies that are being operationalised by diplomats and state media:
- the coordination of externally facing communication, including 24 x 7 foreign media monitoring and rapid response
- the promotion of diverse sources, noting that international audiences are inclined to accept independent media
- support for Chinese social media platforms such as Weibo, WeChat and Douyin
- enhanced forms of communication targeted to specific audiences
- the cultivation of foreign talent.
The party-state appears to be allowing for experimentation across the apparatus of government in how to promote the CCP’s view of its place in the world. This study suggests that covert influence operations on Western social media platforms are likely to be an ongoing element of that project.
This analysis used a mixed-methods approach combining quantitative analysis of bulk Twitter data with qualitative analysis of tweet content. This was combined with independently identified Facebook accounts, pages and activity including identical or highly similar content to that on Twitter. We assess that this Facebook activity, while not definitively attributed by Facebook itself, is highly likely to be a part of the same operation.
The dataset for quantitative analysis was the tweets from a subset of accounts identified by Twitter as being interlinked and associated through a combination of technical signals to which Twitter has access. Accounts that appeared to be repurposed from originally legitimate users are not included in this dataset, which may potentially skew some analysis.
This dataset consisted of:
- account information for 23,750 accounts that Twitter suspended from its service
- 348,608 tweets from January 2018 to 17 April 2020
- 60,486 pieces of associated media, consisting of 55,750 images and 4,736 videos.
Many of the tweets contained images with Chinese text. They were processed by ASPI’s technology partner in the application of artificial intelligence and cloud computing to cyber policy challenges, Addaxis, using a combination of internal machine-learning capabilities and Google APIs before further analysis in R. The R statistics package was used for quantitative analysis, which informed social network analysis and qualitative content analysis.
Research limitations: ASPI does not have access to the relevant data to independently verify that these accounts are linked to the Chinese Government. Twitter has access to a variety of signals that are not available to outside researchers, and this research proceeded on the assumption that Twitter’s attribution is correct. It is also important to note that Twitter hasn’t released the methodology by which this dataset was selected, and the dataset doesn’t represent a complete picture of Chinese state-linked information operations on Twitter.
Download full report
Readers are warmly encouraged to download the full report (PDF, 62 pages) to access the full and detailed analysis, notes and references.
12 Jun 2020