Coordinating and doxing data: Hong Kong protesters’ and government supporters’ data strategies in the age of datafication

ABSTRACT

Scholars across various disciplines have discussed the role of ICTs and social media in contentious politics and social movements. Other studies have also highlighted the duality of data particularly in the context of contentious politics: data is simultaneously an object of political struggle and a weapon in repertoires of political struggle. Taking the protesters’ collective and networked program of sousveillance during the 2019 Hong Kong Anti-ELAB Movement as an example, this article examines protesters’ and pro-establishment’s engagement with data, the datafication of police as proxies of the state, and attitudes towards various forms of ‘veillance’. The doxing at the heart of this case study provides an example in which data as constituents of movement repertoire and data as a contentious issue in its own right are conflated. This article bridges the surveillance and social movement literature by examining data practice and doxing during the protest.

KEYWORDS:

• Datafication
• doxing
• anti-ELAB movement
• contentious politics of data
• data activism
• data ethics

Introduction

Hong Kong offers a unique case to consider the intersection of data and contentious politics. The city’s postcolonial history has positioned the jurisdiction as an ‘open’ exception with reference to a ‘closed’ China. However, in contrast to the narrative of an ‘open, smart city’, the legislative framework inherited from the British colonial government defining the role of data and guaranteeing rights to information remains weak (Hoyng, 2021; Meng et al., 2019). Access to data has been framed as a service, rather than a right by the Hong Kong government, and as a privilege by private sector corporations (Hoyng, 2021). The resulting information asymmetry has led to Hong-Kong-based data intermediaries to engage in ‘counter-data action’ by building their own datasets (Meng et al., 2019). In the context of increased tension over Hong Kong’s postcolonial political and cultural identity, cross-border data mobility under China’s Greater Bay Area plan¹ and the vague definition of ‘personal data’², activists and citizens have begun to exploit, contest and reimagine the norms around the openness of data. Most recently, this contestation over data in Hong Kong took place during the Anti-Extradition Law Amendment Bill (Anti-ELAB) movement of 2019.³

The networked and datafied contentious performances seen during the movement provide a unique context to re-examine the nexus between data-driven governance and individual data practice during times of contention. The Hong Kong government adopted a hybrid strategy of repression and policing that blended ‘hard’ and ‘soft’ methods (Wang et al., 2020). ‘Hard strategies’ included the forceful suppression of rallies with tear gas and water cannon. ‘Soft strategies’, on the other hand, included the collection of protesters’ data through facial recognition software (Mozur, 2019), transport usage records and social media posts. In response to datafication and surveillance carried out by the state (and pro-government supporters), protesters engaged in anti-surveillance practices such as wearing masks, destroying CCTV cameras, and disabling the facial recognition login on their phone to disrupt the government’s ability to collect biometric information. They also appropriated the techniques of the state through an organized program of doxing – the public exposure of private information without consent – that crowdsourced and publicized information about police and people suspected of being ‘snitches’. This weaponization of data influenced the movement’s dynamic and the contentious performances exchanged between the state and protesters. Police started wearing masks and stopped prominently displaying their badge numbers to conceal their identity from protesters (Mozur, 2019). Government supporters also collected and publicized images of individuals participating in illegal activities such as ‘unsanctioned’ protests and the destruction of property on platforms such as hkleaks.pk and reboot-thug-archive. The doxing campaigns seen in the Hong Kong Anti-ELAB protests of 2019 serve as a case study that neatly encapsulates the duality of data – that data is simultaneously an object of political struggle and a weapon in repertoires of political struggle (Beraldo & Milan, 2019; van der Graaf et al., 2016).

This article sets out from the assumption that datafication now also exists as a foundational structure within society that can create and remove opportunities for change, particularly during episodes of contention. We situate doxing as a practice that sits at the nexus of Beraldo and Milan’s (2019) data as weapon/data as stakes dichotomy. This dichotomy is particularly relevant in the Anti-ELAB movement case because we see an overlap in both means and objects of datafied contention between the two politically opposed groups. We explore the specific dynamics of doxing as an individual and a collective performance within Hong Kong’s contentious repertoire. Doxing, when done collectively in the context of contentious politics, is shown to shift power dynamics between claimants, as doxers take on the practices of (or even the role) of the state. Thus, in certain contexts, doxing can become an organized program of either ‘sousveillance’ (the surveilled subject is ‘gazing back at (state) power from below’) or ‘surveillance’ depending on the ideological positioning of the doxers vis-a-vis the state. This assertion frames doxing, typically seen as an unethical practice, as something that can be justified, even morally appropriate, in certain circumstances. Making a judgement on this ethical paradox, as well as qualitatively interrogating the way Hong Kongers perceive this paradox, are beyond the scope of this study and warrant further research.

We focus on the discourses about doxing police and their families on LIHKG⁴, as well as two Telegram channels (Dadfindboy and Youcangotojail) organized by protest sympathizers and government supporters respectively to examine the various strategies, target audiences, and moral justifications given for doxing. Linkages between these moral justifications and the persuasive functions of movements (Benford & Snow, 2000; Stewart et al., 2007) can be drawn – which further connects individual instances of doxing as a data practice with collective identity in the movement (Milan, 2015; Trottier, 2017). Our findings suggest that doxing in the Hong Kong example changed the power relations between police, protesters, and government supporters; it also became forms of sousveillance and surveillance that blended more traditional image-based monitoring techniques with contemporary techniques of datafication. We conclude by discussing the consequences of connecting individual data practices with the proactive mobilization of collective identity and the (re)definition of moral boundaries.

Literature review

Contentious data politics

Studies of contentious politics and data activism have shed light on the importance of agency, power, and practice in the digital age (Beraldo & Milan, 2019; Flesher Fominaya & Gillan, 2017; Pellegrino et al., 2019). Contentious politics refers to when ‘ordinary people – often in alliance with more influential citizens and with changes in public mood – join forces in confrontation with elites, authorities, and opponents’ (Tarrow, 2011: 7). It can occur over any issue and manifest itself through a variety of mediums, including data. Data are contentious in nature (Beraldo & Milan, 2019; Milan & van der Velden, 2016) and are often repurposed for various political purposes such as mobilization, surveillance and repression (Dencik et al., 2016). Beraldo and Milan’s (2019) claim a paradigm shift has occurred in datafication – from data politics to the contentious politics of data. They (2019: 2) define the latter as:

The multiplicity of bottom-up, transformative initiatives interfering with and/or hijacking dominant, top-down processes of datafication, by contesting existing power relations and narratives and/or by re-appropriating data practices and infrastructure for purposes distinct from the intended.

In this sense, data activism can serve as a ‘heuristic tool’ that can be used to explore how activism evolves in response to big data (Milan & van der Velden, 2016, p. 66). The term ‘data activism’ has emerged to describe the adoption of data to reach movement goals or to engage in contestation (Kazansky et al., 2019; Lehtiniemi & Haapoja, 2020). Data activism can take many forms and be ‘pro-active’ as well as ‘re-active’. The former refers to instances where data are used as part of civic engagement, advocacy, and campaigning for social change, whereas the latter focuses on how individuals and groups resist corporate intrusion into their private lives, datafication and state surveillance (Pellegrino et al., 2019; Schrock, 2016). Doxing can be both a pro-active and re-active form of data activism and has increasingly been seen in contentious events of recent years such as the Arab Spring, the Black Lives Matter movement and the Anti-ELAB movement of Hong Kong. In these contexts, doxing is a practice grounded in datafication and thus inextricably linked to surveillance.

Surveillance, anti-surveillance and sousveillance

To connect surveillance studies and social movement research, we need to first start with studies on policing protest (della Porta et al., 2006; Earl & Soule, 2006; Jackson et al., 2019). Surveillance – or ‘watching from above’—is typically understood as a form of disciplinary power wielded by the state or capital (Foucault, 1977). With specific reference to surveillance and social movements, research has focused on the repression-protest nexus (Jackson et al., 2019; Schlembach, 2018), in other words how policing and protester tactics mutually shape each other and political opportunity structures. Studies have explored the effects, real and perceived, that surveillance has on human behavior and mobilization. Many (e.g. Ullrich & Wollinger, 2011) have highlighted the connection between prominently displayed surveillance devices such as CCTV cameras and compliance with the law. Others (Lyon, 2018; Stoycheff et al., 2019) have considered how less visible forms of surveillance, such as network infiltration and the datafication of individuals (whether it be by states or big-tech), have prompted self-censorship and thus had a ‘chilling effect’ on the mobilization of dissent. Individuals are shown to be enmeshed within a ‘surveillance assemblage’ (Haggerty & Ericson, 2000; Ullrich & Knopp, 2018), comprising multiple layers of surveillance actors, including the government, police, their peers, opponents and the general public. Peer-to-peer surveillance has become increasingly easy with advancements in personal technological devices and is particularly important in the context of contention over national (or in Hong Kong’s case, local) identity, as it connects popular (people-driven) surveillance practices to state power (Bitso et al., 2012).

Efforts to resist surveillance and equalize embedded power asymmetries have also become popular topic of research. Some (Dencik et al., 2016; Hermida & Hernández-Santaolalla, 2018) have focused on anti-surveillance, whereby individuals adopt practices that disrupt or evade the government’s attempts to collect their information. Scholars (Maragkou, 2019; Wood & Thompson, 2018) have discussed the role of crowdsourced information which ‘facilitate[s] surveillance discovery, avoidance, and counter-surveillance.’ Others (Mann & Ferenbok, 2013; Mann, 2005, 2013) in the field of surveillance studies have highlighted the emergence of a new kind of ‘veillance’—sousveillance—whereby the subject is ‘gazing back at (state) power from below’ in order to document misconduct and foster accountability. To date, literature concerned with sousveillance has focused on wearable devices that capture, process, store, recall and transmit human sensory information and thus has predominantly considered the individualized nature of sousveillance as a contentious performance. This article provides an example of sousveillance that is focused on a practice not considered as such before and highlights how it can be a highly coordinated, networked program that transcends individuals, and further constitutes a movement repertoire in influencing collective identity. Furthermore, we extend discussion of how sousveillance (and surveillance) has been integrated into mobilization framing and persuasion (Benford & Snow, 2000; Stewart et al., 2007).

Doxing and data ethics

Doxing refers to the public exposure of private documents and information without consent (such as photos, phone numbers, and home addresses) online, with the intent to humiliate, harass, intimidate, punish and/or blackmail targets, or condemn certain actions and ideas (Carty & Reynoso Barron, 2019; Fish & Follis, 2016). In light of the damaging effects doxing can have on its targets, it is typically seen as unethical (McNealy, 2018). However, doxing has been increasingly connected with digital activism and framed as a technique sometimes necessary to reveal wrong-doing or hold wrong-doers to account (Douglas, 2016). The practice enables users to redefine what counts as acceptable conduct and contest the boundaries of what the state defines as acceptable intervention. From this perspective, Trottier (2017) describes certain kinds of doxing as ‘digital vigilantism’, a user-led policing that defies, distrusts, or supplements police and the state who are unable or unwilling to prosecute perceived offenses. In this framework, the goals of doxing are broadened to include responding to crimes, preventing offenses, providing information, and negotiating values.

To navigate between ‘justifiable’ and ‘unjustifiable’ instances of doxing, scholars have developed several typologies that categorize the intention behind doxing, the type of information leaked and the consequences the information release will have on the doxed subject. Douglas (2016), for example, distinguished between doxing that de-anonymizes, targets, and de-legitimizes its subject. In this framework, de-anonymizing doxing publicizes personal information that establishes the identity of a previously anonymous person; targeting doxing refers to releases of information about an individual’s circumstances that are not publicly shared; and lastly, de-legitimizing doxing discloses intimate information that harms the integrity of an individual. Trottier (2017) categorizes doxing along several axes including: doxer, target, and/or law enforcement. MacAllister (2017) conceptualizes the purpose of doxing from a legal standpoint. She suggests three functions: to ‘punch down’ a target either as a symbolic act, to enact revenge or punishment, or simply for the ‘fun’ of it; to increase the transparency of an issue or group to the public, heightening attention to newsworthy information; or to expose a perceived injustice.

Data and methods

Following Douglas (2016) and Trottier’s (2017) definitions that characterize doxing as the release of private information on a public platform, we chose two platforms – LIHKG and Telegram – commonly used by movement and government sympathizers to ‘name and shame’ perceived opponents. LIHKG was launched in Hong Kong in November, 2016. It is a Reddit-like, multi-category discussion forum that allows people registered with an institutional or ISP email (not gmail or hotmail) to sign up and anonymously post, upload, vote, and chat. On LIHKG, users can vote discussion threads up or down and thereby popularity drives which threads will be presented first. During the Anti-ELAB Movement, LIHKG became popular amongst protesters to exchange ideas, disseminate information, and make decisions (Lee, 2020). Telegram, by contrast, is a not-for-profit, encrypted, cloud-based messaging app. It distinguishes itself from other platforms such as Facebook and Google by emphasizing its commitment to privacy, and is therefore popular among protesters to share information and organize activities (Maragkou, 2019).

We examined posts on LIHKG specifically related to doxing to get a sense of discourse on doxing the police and their families. To gain insight into how protesters and government supporters actually doxed their perceived opponents, we chose to further investigate posts on two Telegram channels created with the stated purpose to dox perceived opponents: Dadfindboy⁵ and Youcangotojail channel⁶. Dadfindboy was established on 7 July 2019, with the stated purpose of publicly releasing the private information of police officers to de-anonymize them. Youcangotojail channel was established on 24 November 2019. The stated goal of Youcangotojail channel is to take collective action against ‘rioters’ because their practices are ‘illegal’ and ‘disturbing the social stability’.

Protest-related posts on LIHKG and Telegram are predominantly written in colloquial Cantonese, the language spoken in Hong Kong. This lowers the barrier of entry for Hong Kongers to participate in conversations on the protests while simultaneously limiting those who cannot read traditional Chinese characters from engagement. This infrastructure provides what Tufekci (2017) describes as ‘network internalities’, the benefits and collective capabilities obtained from the constant work of negotiation and interaction required to maintain the networks as functioning and durable social and political structures.

Methodologically, first, we registered an account with each platform to review the content available and understand any differences between publicly available information and the information available to members only. We concluded that the content available to members and non-members is similar in the channels this research focuses on. Based on the linguistic conventions adopted by users (e.g. ‘dog’ is a common usage referring to the police), we scraped LIHKG posts that included the following keywords in their titles: ‘dadfindboy’, ‘bad cop + doxing’, ‘cop’s wife’, ‘cop’s son/daughter’, ‘bad cop + punish’, ‘fifty cents army’⁷, ‘dog’s badge number’, ‘dog’s home’, ‘dog’s name’, among others. A total of 154,920 posts were scraped. We then used a data analytic program, KNIME⁸, to filter out duplicated posts, posts that only contained emojis or the words ‘push’ and ‘pin’, and removed posts that were made outside the period of study (before June 2019 when the protest started). After these processes, 37051 posts remained. We exported all posts from the two channels (10,965 from Dadfindboy and 1,062 from Youcangotojail) into Excel spreadsheets for manual coding. We developed a coding matrix with examples. Our research assistants first coded 50 posts which were exchanged amongst the team to test inter-coder reliability. This data is sensitive, even more so since the passage of the 2020 National Security Law (NSL). To enhance the privacy and security of the data as well as the researchers, the scraped data was stored outside of Hong Kong. Access to the data was protected by password and only granted to coders.

Coding matrix

Key foci of this research are to map the citizenry’s datafication of the state (and each other), the role doxing played in mobilization and to understand the individual and collective elements of doxing within the context of contentious politics. To achieve these aims, it is important to distill: 1) the type of information participants collected and released; 2) the real or intended consequences of the information release; 3) the sentiments used to frame the release – particularly those that reference a collective identity; 4) how the posts align with the persuasive functions of movement. Thus, we developed a coding matrix (see Table 1) drawn from Marx’s (1999) seven types of identity knowledge and Douglas (2016) and Trottier’s (2017) types and intent of doxing.

Table 1. The coding matrix for the two Telegram channels.

Identity knowledge type	Description
Legal name	The name under which someone is known for official and legal purposes
Locatability	Information that reveals where someone lives or where s/he can be contacted personally.
Pseudonyms (linked to name or location)	A name or code that represents a single individual such as a bank account number in a system that is related to their legal name.
Pattern knowledge	Someone who can be recognized by their regular public actions or habits, such as catching the same bus at the same time every day.
Social categorization	Information that can be used to place someone in social categories (or stereotypes) such as physical appearance, accent, style of dress, etc.
Symbols of eligibility/non-eligibility	Possessing artefacts or knowledge that identifies someone as being entitled to particular privileges and treatment, such as a uniform, a password etc.
Intent of doxing
Harass	Include remarks that incite irritating pranks
Punish	Include statements/words such as: creating troubles for XXX
Hold to account	Include statements/words such as: sending XXX to jail or for public judgement
Degrade	Include statements/words such as: dog, retarded, low-educated, etc.
Types of doxing
De-anonymizing	Disclosure of legal name and locatability of the subject, presented without any ‘pattern knowledge’ and without any narrative from the poster.
Targeting	Disclosure of legal name and locatability of the subject accompanied by details that establish ‘pattern knowledge’ and ‘social categorization’
De-legitimizing	Photos of the subject showing them transgressing social norms. Accounts detailing the subjects excessive use of force or abuse of their power as a police officer.
Sentiments
Patronization	Terms or expressions used to patronize and tease targets such as: wake up, daydream, face the reality, idealistic, wishful thinking, silly, poor thing, etc.
Sarcasm	Tones or expressions used to mock targets such as: what a brave soldier, what a handsome policeman, beautiful police wives, hard-working police, great communism, etc.
Anger	Expressions show one is upset and feel the current status quo is unfair, such as: pissed off, feel unfair, upset, fuck your mother, fuck you, fuck off, fuck bad dog, punishment to your family and children, etc.
Insults	Terms or expressions that degraded the subject in some way, such as: dog, dickface, (fucking) cunt, undercover dog, slut, rapist, cop cockroaches, blue dead bodies, low IQs, etc.
Cheers	Terms or expressions used to mobilize and cheer participants, such as liberate Hong Kong, revolution of our times, hang in there, brothers and sisters, thank you brothers and sisters, need people to back up, be water, etc.

The disclosure of police and protester information in the Dadfindboy and Youcangotojail Telegram channels fits the criteria to be both de-anonymizing and targeted types of doxing, by virtue of being posted in forums established with the specific purposes of revealing the perceived opponents’ identity to yield consequences (punishment). Thus, to interrogate differences in the types of doxing within this setting, we introduced an additional criterion to distinguish between posts that merely de-anonymized subjects and those that actively sought to target subjects. Posts that released information and contained some narrative calling for the harassment of or action against subjects were treated as targeting doxing.

For LIHKG data, to further explore the intent behind the doxing practices, we supplemented the three categories listed above with sentiment and content analysis in order to understand how doxing configured the persuasive functions of social movements. To increase the data accuracy and be more target-oriented in the sentiment analysis, we used a process of inductive coding whereby we reviewed 1,000 posts and identified regularly appearing terms and the dominant sentiments they convey. The dominant linguistic devices include: cheers, insults, curses, sarcasm and patronization.

Findings

Channel content and positioning

Perhaps unsurprisingly, users in both Dadfindboy and Youcangotojail channels posted content not expressly related to doxing. Dissemination of news, attitudes towards data collection and educative material appeared in similar proportions on both channels. Interesting differences in the discourse found on the two channels were also observed. As Table 2 suggests, although Dadfindboy has the reputation of being a channel dedicated to doxing, doxing only makes up 42% of its content. In contrast, the content of Youcangotojail is more concentrated on doxing (78%). The rest of the post includes sharing the news (9%), data tactics education (6%), and other physical world (non-data related) strategies such as making gas bombs or when to retreat.

Table 2. Post content of the two Telegram channels.

Post content	Youcangotojail (based on 10965 posts)	Dadfindboy (based on 1,062 posts)
Doxing	78%	42%
News	15%	9%
Purpose of channel	2%	1%
Attitudes towards data collection	5%	5%
Education on data tactics and resistance	1%	6%
Carrie Lam^9	0%	6%
Physical protest field strategies	0%	31%
Total	100%	100%

We examined the linguistic devices utilized by participants across LIHKG and the two Telegram channels (see Table 3). We found that although insults and sarcasm are prevalent overall, Youcangotojail appeared to have more sarcasm and condescension, whereas in Dadfindboy and on LIHKG insults were the dominant method of expression. As Table 4 suggests, on LIHKG the comments containing insults and curses mostly targeted the police, whereas a higher proportion of sarcastic comments were directed towards China supporters and the spouses of police officers.

Table 3. Linguistic device on LIHKG directed at target.

Linguistic device	Youcangotojail	Dadfindboy	LIHKG
Insults	30%	88%	58.3%
Curses	-	-	9.9%
Cheers	1%	1%	13.9%
Patronization	6%	0%	-
Patronization + insults	2%	0%	-
Patronization + sarcasm	2%	0%	-
Sarcasm	42%	11%	17.8%
Sarcasm + insults	14%	0%	-
Total	100%	100%	100%

Table 4. Linguistic device of the two Telegram channels and LIHKG.

Target	Insults	Cursing	Sarcasm	Cheers
Police	66.64%	82.7%	20.18%	0%
Police spouse	11.35%	0.11%	24.08%	0%
Police family members	16.83%	17.18%	11.16%	0%
Pro-government supporters	5.19%	0%	0.65%	100%
China supporters	0%	0%	24.12%	0%
Fifty cents/undercovers	0%	0%	19.80%	0%
% of total 37,051 posts	58.3%	9.9%	17.8%	13.9%

On Telegram, users in Youcangotojail used a more varied range of linguistic techniques than those in Dadfindboy. Interrogating the temporal dimension of the data highlighted shifts in the use of linguistic devices over time (see Figure 1). Users in Youcangotojail cycled through four linguistic devices – sarcasm, insults, sarcasm again, and finally condescension. The shift between these techniques coincides with key phases in the protest cycle. In contrast, up until the passage of the NSL in June 2020, users in Dadfindboy consistently employed insults to express themselves, giving the appearance of a protracted protest event between November 2019-June 2020. The prevalence (and peak) of insults and threats from late September to late-October 2019 coincides with the most violent period of the movement. It is likely this linguistic departure from the pro-democracy movement’s motto of being ‘peaceful, rational and non-violent’ (wo-lei-fei) is linked to the escalating episodes of violence between protesters and police. In May 2020, there is a drop in the use of insults towards police and sarcasm becomes more widely used. This change in linguistic device is probably the result of the increased legal cost of insulting police officers under the revised political opportunity structure created by the Chinese state through the NSL. The apparently coordinated use of linguistic devices by both channels also confirms the emergence of a contentious repertoire as Tilly (2006, 2008) defines it.

Figure 1. The line graph of sentiments on the two Telegram channels over time (BAOBAO here refers to Youcangotojail channel).

Protesters infiltrated Youcangotojail and posted images and videos of the attacks in Yuen Long carried out by masked men in white shirts against movement sympathizers. These posts asked users to widely share these footages to allow people around the world to know how riotous and violent these ‘protesters’ are. These users adopted satire to achieve the goal of balancing anti-protest discourse as well as mobilizing more people in support of protest.

Evolution of doxing as a contentious repertoire

Activity on Dadfindboy peaked between July and October of 2019, the time when street protests were at their peak, and also the time between the announcement of the Extradition Bill and its withdrawal in October 2019. The datafied practice of collecting and releasing police officers’ information as contentious practices matured quickly. In the beginning of Dadfindboy’s existence, participants simply posted photos of individual police officers and numbered them in chronological order of appearance. Others would respond with information and quote the reference number, thus giving the impression that the number of appearance was a ‘reference number’ in a cataloguing system for a ‘database’ of officer information. After July of 2019, photos that showed police officers with family members and their social media contact details also began appearing. Posts progressively became more detailed and by the end of July a template for posting information was established. The template encouraged participants to provide as much detailed information as possible about the police officer, their spouse and their children. The creation of a template emphasized collective ownership of and responsibility for the collation and use of the information in meaningful ways. The template was often used in isolation, that is to say that users did not add their own views or commentary to the post. This had the effect of framing the post as a pseudo neutral, factual information release or ‘database entry’. In October 2019, citing the malicious intent of the Telegram channels’ organizers, the Hong Kong High Court granted an interim injunction to ban the doxing of Hong Kong police officers on Telegram LIHKG (Chan, 2019). This decision symbolized the state’s denunciation of doxing.

In response to Dadfindboy, government supporters established Youcangotojail channel in November 2019. Activity on Youcangotojail channel was highest between November 2019 and January 2020, with a small spike in traffic again in May 2020 when protests against the introduction of the NSL occurred. ‘Anti-protester surveillance’, ‘anti-violence’, and ‘anti-riot’, were the major expressions that appeared within the channel’s discourse. The data collected in this channel typically included photographs of protesters engaging in ‘violent’ and ‘illegal activities’, and the location where the photo was taken, rather than providing information based on a template. Data collection and sharing practices in this channel did not evolve to the same degree as was seen in Dadfindboy. Participants posted photos of people attending rallies and captioned the photo with the date, time and place the photo was taken. These practices combined created the impression of an alliance between users and the police. Users framed themselves as releasing information about illegal activities that police could follow up and take action on. This ‘adopted’ proximity to state power may have influenced the connective logic among users on Youcangotojail channel. With an assumption that the state has ‘better’ surveillance capacity and technology, detailed templates for data input become unnecessary.

Supporting Tilly’s (2006, 2008) conceptualization of contentious repertoire, the performances exchanged between the claimant-object pairing evolved dialectically and ultimately to shifts in the political opportunity structure. The contentious repertoire based in the digital world influenced and shifted the course of the contentious repertoire manifest in the physical world. The exposure of police officer’s personal information led them to stop wearing identification badges and start wearing masks to cover their faces (Mozur, 2019).

Targets, types and intent of doxing

Three claimant-object pairings were observed across LIHKG and the two Telegram channels. In Youcangotojail, government supporters positioned themselves as claimants against movement supporters. In Dadfindboy and on LIHKG, the police and their families were targeted by movement supporters. After the passage of the NSL a third claimant-object pairing emerged on LIHKG – movement participants and individuals suspected of being members of the fifty cents army, who are thought to be surveilling and reporting the protesters. Users were alive to the possibility that their networks had been infiltrated by government supporters and police, and discourse participants engaged in surveillance of the posted content. Any posts that seemed suspicious were flagged and identifying information about the poster was collected and publicized. A google document titled ‘the user accounts of fifty cents who posted fake news on LIHKG’ was created. The information shared includes the user IDs of participants believed to be members of the fifty cents army. To substantiate the assertions, the online behaviors of the individual were documented (such as commenting in support of greater unification between Hong Kong and China, remarks against independence, and spreading fake news), and links to the individual’s social media profiles were included. Going beyond these obvious differences in doxing targets, an interesting variation was observed between the two channels. During the period of study, 65% of data releases in Dadfindboy targeted individual officers. The remaining 35% of releases related to the officers’ family members. In Youcangotojail channel, a different pattern is seen—67% of data releases relate to groups of people and only 33% focus on a specific individual. The difference in intended use of the data (to assist police identify protesters) is a likely explanation for this (see Table 5).

Table 5. Types of doxing on two TG channels.

Intent of doxing	Youcangotojail	Dadfindboy
Degrade	8%	9%
Harass	0%	1%
Justify doxing	47%	5%
Justify doxing and harass	42%	0%
Justify doxing and punish	1%	0%
Punish	0%	1%
Threaten	0%	83%
Total	100%	100%

Doxing that targeted its subjects was the dominant type of doxing seen on Youcangotojail and Dadfindboy (see Table 6). In Dadfindboy, the identity knowledge types most released included officer’s legal name, locatability and pseudonyms such as badge numbers and car number plates. However, posts in Youcangotojail channel rarely released identity knowledge of the types Marx’s (1999) set out. Instead, Youcangotojail channel’s users practice of posting photographs containing a few protesters’ faces suggests an assumption that in the age of advanced facial recognition software, an individual’s face can now be an identity knowledge type that can be used to find out an individual’s legal name and locatability.

Table 6. Intent of doxing in two TG channels.

Type of doxing	Youcangotojail	Dadfindboy
De-anonymizing	1%	1%
De-legitimizing	0%	3%
Targeting	82%	96%
Targeting and de-legitimizing	17%	0%
Total	100%	100%

The Telegram channels played the role of coordinating actions that go beyond discussion on social media. Participants came up with practices that used the private data of police as a weapon to instigate the harassment of these individuals. Examples included putting the addresses and phone numbers of police officers on to flyers advertising the sale or rent of apartments, school enrollment and restaurant deals. This practice thus drew in unsuspecting civilians as those interested in buying or renting an apartment were co-opted into the harassment of police officers. Other strategies focused on generating anxiety of the unknown, for example one post stated:

Extending punishment to their wives and children is not just about doxing per se, we want to create a scenario in which the police will be fearful of their children being bullied on campus, or having something ‘extra’ in their meals when eating in a restaurant …

By justifying the doxing of wives and children as parties linked to but distinct from the primary target of the doxing – police officers, this comment suggests an awareness that doxing in certain circumstances might be denounced, which means people might find it morally questionable and therefore an unacceptable tactic.

Justification for doxing

In Dadfindboy, the justifications given for doxing typically fall into one of two categories. That the police ‘deserve’ the treatment, or that the threat to Hong Kong’s democratic freedoms necessitates extreme measures like doxing. In the former line of reasoning, police are framed as agents of Chinese state power, excessively violent and/or as having questionable morals. Examples include: ‘#XXX, has a history of stealing and secretly filming ladies’ underwear, beware of this pervert everyone!’; ‘a dog has to be responsible for its misbehavior, it’s not innocent!’ and ‘so many people and their parents and friends are hit by the police, don’t they [the police] need to pay the price for the violence?’ In the latter, the power disparity between the people and the state is highlighted, so too is the importance of movement aims such as democracy. Comments in this category are in closer alignment with the commonly cited aims of the movement. Examples include: ‘We can’t fight against the police, they have powerful weapons, but we can create influence in their everyday lives … ’

In Youcangotojail government supporters justify doxing with reference to ‘maintaining social stability.’ Protesters are often framed as ‘rioters’ or ‘cockroaches’—a slur implying they are subhuman but resilient – which provides the rationale for harsh and cruel treatment against protesters. Photographs of protesters are often captioned with descriptions such as ‘sneaky cockroaches hanging around the streets, must be doing something bad’; ‘ … a group of rioters used hammers and sticks to attack the police and destroy shops. It’s nowhere close their claims of being peaceful, rational, and non-violent’; ‘there will and must be karma for the rioters who doxed the police; you followed the police car and then we follow you!’ Posts were also accompanied by remarks such as ‘remember you can go to jail’ or ‘you dare to protest in the street then I dare to take photos of you!’

Discussion

The 2019 Hong Kong protest has substantiated that data practices are of increasing importance in social movements’ strategies as well as a debated subject. Empirically, this article has documented the creative incorporations of data-based practice into Hong Kong’s contentious repertoire. Through data practice and perception of it, the action repertoire changed and new forms of mobilization/conflict also emerged (Flesher Fominaya & Gillan, 2017; van der Graaf et al., 2016). Unlike the #MeToo movement in China, India, and Turkey (Bitso et al., 2012), in the Hong Kong case we saw two ideologically opposed groups weaponizing data against each other in the same way (doxing), and the same data objects (legal name, locatability, and pattern knowledge) becoming the source of contention. However, the justification of these practices was very different. In the situation where opposing camps have weaponized data in the same way, and the same data objects are the subject of contention, the different moral justifications of doxing cause us to rethink the positionality of data as weapon vis-a-vis data as object. This extends Beraldo and Milan’s (2019) observation of the coinciding aspects of data by showing that the relationship between these components does not just exist in a constant state of contradiction as the means and the object of contention. Instead, the weaponization (means) and data objects (ends) can be aligned and realigned with each other in a form of symbiosis where means and ends are used to mutually justify the doxers’ ultimate end – the de-legitimization of their opponent.

Datafication and power

The state’s datafication of individuals through the collection of legal names, locateability and pattern knowledge was a technique refracted through participants and shone back onto the state. When used by protesters, the process of datafication enabled the various levels of identity of police officers to be brought together. Datafication in this instance positioned officers as members of a whole community, with a name, address and family members rather than as individuals identified by numbers. This has important consequences for power dynamics. De-anonymizing the police officers (e.g. disclosing their badge numbers) and casting them as individuals, as citizens, strips them of their powers as police and the protection of their government role. Instead, it places them on par with protesters. This removal of the role, in other places, has given police officers and other government officials the opportunity to take action as individuals (as we saw in Black Lives Matter protests where policemen supported protesters). However, in this case, removing the distinction between the officers’ private and public lives has resulted in significant damage at the personal level.

By the same token, the datafication of the protesters by government supporters had the effect of changing their social status. Collecting footage of other citizens positioned government supporters as proxies of the state, sort of ‘watchmen’. Then, through the process of sharing the information publicly (and thus with the state) individual participants became agents of the state. Further, the practice of posting photos of protesters captioned with the misdeeds they are alleged to have committed removes their identity as citizens standing up for their rights and the legitimacy that is afforded by such identity. The photos become a contemporary ‘wanted poster’ framing protesters as criminals who will be sought by the authorities.

Data is weaponized for ‘veillance’

Collecting and revealing personal data has become an important individually executed, yet collective strategy that can be used to challenge or reinforce existing power asymmetries. At one level, these findings highlight the individualized nature of doxing as a contentious performance that originates and is implemented in an individual’s daily life through personal frames. Individuals post the personal information of others, in separate but connected (by virtue of being co-located in a Telegram channel) acts of doxing. The desired outcomes of these acts of doxing also lie at the individual level. Participants in Dadfindboy seek to increase the police’s fears and disrupt or damage the personal lives of police officers and their family members, whereas Youcangotojail aims to trigger legal charges against protesters as private citizens. When viewed as part of a coordinated and collaborative campaign, the individual instances of doxing can be seen as coordinated programs of sousveillance and surveillance in the respective cases of Dadfindboy and Youcangotojail channel. The concentration of doxing, and therefore data, in two locations had the effect of converting the data into a database that was searchable and could be manipulated if additional steps were taken. The data practices in both instances were consistent with surveillance techniques employed by the state. Thus, our findings drawn from LIHKG and the two Telegram channels suggest that the doxing in the Hong Kong example became forms of sousveillance and surveillance that blended more traditional image-based monitoring techniques with more contemporary techniques of datafication.

Conclusion

Situating doxing within the ontologies of data activism comes up against the norms of data release in other contexts, and challenges our understanding of what ethical data practices are and when information releases are justifiable. Our finding that doxing can be a form of sousveillance or surveillance has the effect of conflating several distinct processes each with their own moral codices. Doxing is typically associated with malicious intent and as an unethical practice in light of the damage that can be done to targets. Surveillance has both positive and negative connotations whilst sousveillance is generally understood to be an acceptable practice motivated by a sense of justice. The ethical grey zone in which doxing stands, as a form of data-activism, could pose limits to its use as a protest tactic and negatively influence the legitimacy of movement participants and their claims. The justifiability of doxing as a form of data-activism requires further research. Specific attention should be given to determining the proportion of individuals whose data was released that experienced damage, the nature of the damage experienced, and the perceptions the general public hold of doxing.

Acknowledgements

We would like to thank Shao-Fang (Christine) Kao for scraping the data from LIKHG and Mei Ling (Jay) Chan for data analysis. We appreciate the help of Ruiheng (Peter) Li and Lok Hei (Ken) Liu with manually coding the data from the two Telegram channels. The draft was presented at the paper development workshop organized by Stefania Milan and Davide Beraldo, and work-in-progress workshop by the Society for Hong Kong Studies (SHKS). We appreciate the valuable feedback from the participants, especially Vassilis Charitsis, Cristina Flesher Fominaya, Stefania Milan, Davide Beraldo, Francis Lee, Ching Kwan Lee, and Edmund W. Cheng. We extend our thanks to Samson Yuen, two anonymous reviewers, and the guest editors for their constructive comments and suggestions.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

The work was supported by the Strategic Priority Funding scheme, Arts, Design and Architecture (ADA), University of New South Wales [SPF02].

Notes on contributors

Yao-Tai Li

Yao-Tai Li is a Lecturer of Sociology and Social Policy in the School of Social Sciences at University of New South Wales, Australia. He holds a Ph.D. in Sociology from the University of California, San Diego. His research interests include contentious politics, identity, and social media. His work has been published in several scholarly journals including British Journal of Sociology, The China Quarterly, World Development, Urban Studies, New Media and Society, Big Data & Society, Work, Employment and Society, The Sociological Review, Ethnic and Racial Studies, Discourse & Society, Social Science Computer Review, Journal of Contemporary Asia, among others.

Katherine Whitworth

Katherine Whitworth received her Ph.D. in political economy from the University of Sydney, Australia. Her research interests include political legitimacy, social welfare, contentious politics, Hong Kong and Chinese politics. Her work has been published in scholarly journals such as The China Quarterly, Big Data & Society, Urban Studies, World Development, Journal of Contemporary Asia, Journal of Consumer Culture, Social Science Computer Review, and International Migration. She is currently conducting a project on the Lennon Walls with Yao-Tai Li.

Notes

1. See: https://www.bayarea.gov.hk/filemanager/en/share/pdf/Outline_Development_Plan.pdf.

2. The Office of the Privacy Commissioner for Personal Data (PCPD) defines ‘personal data’ as ‘information which relates to a living individual and can be used to identify that individual.’ See: https://www.pcpd.org.hk/english/data_privacy_law/ordinance_at_a_Glance/ordinance.html.

3. Millions took to the streets of Hong Kong in protest against a proposed amendment to the Extradition Law, that would permit the extradition of individuals within Hong Kong’s jurisdiction to China.

4. The online forum LIHKG is widely regarded as a central communication platform for Anti-ELAB movement supporters (Lee, 2020).

5. See: https://t.me/s/Dadfindboy. It is one of the leading channels created in response to police officers not wearing identification badges. It then evolved to even advocate for harassment and violence against police officers (Maragkou, 2019).

6. See: https://t.me/s/canigosomeplacefun.

7. ‘Fifty cents army’ is responsible for uncovering the state’s extensive efforts to delude negative content, and fostering a more legitimate vision of the party (Repnikova & Fang, 2018).

8. See: https://www.knime.com/.

9. Carrie Lam is the current Chief Executive of Hong Kong since 2017.