How the “Internal Brakes” on Violent Escalation Work and Fail: Toward a Conceptual Framework for Understanding Intra-Group Processes of Restraint in Militant Groups

Abstract This article advances the emergent literature on restraint within militant groups in three ways. First, it offers a framework for situating the “internal brakes on violent escalation”—understood as the practices through which group members shape the outer limits of their action repertoires—in relation to the interplay between conflict dynamics, intra-group processes and individual-level decision making. Second, it develops a basic analytical strategy for examining how such brakes operate at different levels of proximity to potential or actual instances of escalation. Third, it sets out four types of mechanisms through which internal brakes appear to generate or enable restraint.

While terrorism scholars have taken considerable interest in how violence "ends" or deescalates, 4 they have paid less attention to non-escalation or "truncated" processes of escalation. 5 And where they have investigated processes of non-or limited escalation, the primary focus has been at the individual level of analysis. 6 Meanwhile, although peace studies and social movement scholars have explored in detail the transitions between nonviolent and violent tactical repertoires, 7 the dynamics of nonviolent resistance 8 and the upholding of nonviolent discipline, 9 they have dedicated less attention to questions about how those who engage in violence establish and maintain limits on that violence. 10 This article comprises the second of two articles that report on a project in which we sought to respond to this limitation in the literature. Our intention in that project was to develop a descriptive typology of the practices through which members of militant groups themselves contribute to establish and maintain the parameters on their own violence, either in the form of resisting or preempting escalation or by exploring less violent alternatives-thereby inhibiting, albeit sometimes only marginally, the type of escalation processes often described within the terrorism literature. 11 We referred to these practices as the "internal brakes on violent escalation," and understood "militant groups," for the purpose of this project, to refer to groups in which a significant proportion of members have shown a willingness to deploy or support some use of violence. Our aim in developing this typology was to provide researchers and relevant policy and practitioner communities with a vocabulary that they could use to develop more systematic descriptions and analyses of the processes through which members of militant groups themselves contribute to shape the outer limits of their action repertoires.
We developed the typology through a review of relevant literatures and the analysis of three militant milieus, selected using a "most-different case comparative strategy," 12 with cases selected for their variation in terms of ideology and the scope and scale of violence. The cases we selected were the British jihadi milieu from 2001 to 2016, set within its international context; the British extreme right during the 1990s, and the animal liberation movement in the U.K. between the mid-1970s and the end of the 1990s.
The jihadi case provided a case of an international milieu in which a particular form of theological reasoning has been used to advocate and justify mass casualty violence, albeit some actors within the movement have at times sought, often using the same body of theological reasoning, to manage the parameters of that violence. The case study examined how actors within the U.K. responded to efforts, spearheaded by the so-called Islamic State (IS), to expand and escalate jihadi violence. Here we focused on the evolution of the Al-Muhajiroun movement, which had always avoided direct association with terrorism but ended up embracing IS after its declaration of a "caliphate," 13 and an autonomous cell of individuals operating primarily on the social networking site Telegram, who pledged allegiance to IS and sought to carry out attacks in the U.K. 14 While Al-Muhajiroun publicly embraced IS, within the Telegram network some members began to question the validity and efficacy of IS tactics. These dynamics were studied in the wider context of debates concerning the limits of Islamist-inspired violence and responses to its escalation, particularly as articulated by leaders of Al-Qaeda. 15 The extreme right case provided a case of mobilization around a racial-nationalist ideology where, although there was significant interpersonal violence, lethal violence was rare. The empirical focus was on the British National Party (BNP) during the 1990s, and their innovation away from violence as they struggled to achieve electoral legitimacy while wrestling to contain the actions and influence of its radical flank, Combat 18 (C18), a "stewards group" the BNP itself had formed to defend themselves from a "direct action" campaign by Anti-Fascist Action (AFA). Even within the radical flank, while further escalation did take place, there were observable limits on violence, and actions that exceeded the parameters of what militants deemed acceptable violence provoked intramovement opposition, disillusionment and in some cases disengagement. 16 The animal liberation case focused on the radical flank of the animal rights movement, understood as that part of the movement where activists were willing to use illegal tactics to advance campaigns for animal rights. The case study centered on the evolution of the Animal Liberation Front (ALF) and campaigns carried out under other organizational banners, such as the Animal Rights Militia and the Justice Department, which entailed escalation beyond established action repertoires. This case study provided a case of a single issue movement. It was also a movement in which, while activists did perpetrate acts of violence that caused physical and psychological harm to human and non-human animals, interpersonal violence was very rare, and lethal force was never deployed, even when state repression and control significantly inhibited the availability of non-or less violent strategies of action. 17 The cases were selected within a single country to provide geographic and therefore broad contextual consistency in case comparison, while retaining the most-different case comparative strategy. The choice of U.K.-based cases reflects the facts that the U.K. has experienced sustained and varied campaigns of non-state political violence, and that as a result of previous projects, the project team already had access to extensive documentary evidence relating to two of the three case studies. As with any research based on a limited number of cases, there are obvious questions to be considered regarding wider applicability of the analysis. It is important that future research examines and compares restraint dynamics in non-U.K. contexts, and indeed that it examines restraint dynamics within other U.K. cases, the most obvious of which being the violence in Northern Ireland. 18 Within this project, we sought to manage these limitations by engaging with the empirical literature around other cases in order to critically interrogate our emerging analysis. We also limited our research objectives accordingly. Specifically, we did not seek to assess the prevalence or effectiveness of the different brakes, or to provide a comprehensive explanation for the relative absence of greater violence. Rather, we sought simply first to identify practices that appeared to (be intended to) inhibit violent escalation, and then to develop a strategy of description that worked across the three main case studies and beyond.
Each case study was based on analysis of primary and secondary documentary evidence. This included the extant academic literature, movement publications, activist memoirs and court documents. For each case study we looked for and coded instances in which group members either (a) acted to reinforce existing parameters of what they deemed appropriate violence, (b) sought to counteract moves by other members of their group toward the adoption of more violent strategies of action, or (c) sought to innovate toward less violent strategies of action. We understood escalation, and by extension non-escalation and de-escalation, as terms relative to the existing repertoires of action within the milieus under analysis. Such a relative understanding of escalation was adopted to enable comparison across cases with very different outer limits of violence. Such an approach does however raise questions about whether there might be important qualitative differences in how thresholds between "appropriate" and "inappropriate" violence 19 work when the starting point is very different. Such questions warrant further scholarly attention.
In the project report and the first article, 20 we argued that it was indeed possible to develop a typology of the "internal brakes" on violent escalation, and that the typology that we developed enabled us to describe the braking practices observed across the three main case studies and across other cases described in the academic literature.
We organized this typology around five intersecting logics on which the intra-group brakes work: strategic logic; moral logic; ego-maintenance logic; the logic of out-group definition; and organizational logic. Brakes that work on strategic logic work on concerns that certain forms of violent escalation are likely to be ineffective or counterproductive. Brakes that work on moral logic work on concerns that certain forms of violence directed at particular targets contradict their ethical principles. Brakes that work on the logic of ego-maintenance relate to questions about whether violent escalation is commensurate with established in-group identities-whether it is the "sort of thing that we do." Such brakes often feed into but are not always obviously reducible to strategic or moral considerations. Brakes that work on the logic of out-group definition relate to processes of boundary softening-understood as the opposite of the boundary hardening often associated with violent escalation 21 -which, by turning members of out-groups into potential allies, supporters or, at least, not an existential threat, can also give rise to and reinforce brakes that work on moral and strategic logics. Brakes that work on organizational logic are brakes that work through processes of institutionalizing the limits on violent escalation or creating intra-movement conditions more conducive to the application of restraint. 22 By organizing the typology in this way, we were able to identify and describe these intra-group brakes on violent escalation, and to offer some insight as to how they work and fail. Yet even as we wrote the conclusions to the previous publications, we were aware that this descriptive typology arguably generates as many, if not more, questions than it resolves. For example, are there particular constellations of brakes that are especially common, or effective, within groups characterized by particular organizational structures or that mobilize around particular ideologies? To what extent do patterns of braking and brake effectiveness vary across the conflict cycle? What happens to intragroup brakes if and when those groups or movements split or fracture? Are some of the brakes more effective in relation to specific styles or degrees of violence?
These are important questions, and providing satisfactory answers to them will require detailed, systematic empirical research. Yet as Paul Gill 23 observes in relation to the wider field of research on terrorism and threat assessment, it will also require clear and effective conceptualization of these brakes and how they work and sometimes fail. It is this requirement for clear and effective conceptualization to which we attend in this article. We make three specific contributions. First, we provide a simple framework for situating these internal brakes both in relation to wider conflict dynamics and in relation to intra-personal processes of decision-making and reflection. Second, we develop a basic analytical strategy for examining how such brakes operate at different levels of proximity to potential or actual instances of escalation. Third, we set out four types of mechanisms through which internal brakes appear to generate or enable restraint. In doing so, we argue, we lay a more stable foundation for researchers and analysts seeking to incorporate processes of restraint within their analysis and understanding of militant groups.
Locating Intra-Group Processes of Restraint within Multi-Level Ecologies of Conflict While our particular interest within this project was the hitherto largely overlooked intra-group or "internal" brakes on violent escalation, these clearly do not emerge, evolve or function in a vacuum. Rather, intra-group processes of restraint are intimately related both to wider conflict dynamics, and to the lived experience, decision-making and reflection of individuals within those groups.
The actions of security forces, societal elites, opposition groups, rivals and the general public are all likely to have a bearing on what group members conceive of as the parameters of appropriate violence, the types of brakes they apply, and the extent to which those brakes are effective. For example, if group members perceive that political opportunities are opening up, it is likely that some will advocate for the adoption of less violent strategies of action in order to exploit those opportunities. 24 Conversely, when political opportunities close and/or where groups experience significant repression, although this might produce a decline in violence, or even the collapse of a campaign, as the costs of "high risk activism" are raised beyond those that activists are able or willing to sustain, 25 it is also likely that some members will respond to escalated repression with claims that greater violence is now appropriate, or the only available course of action, thereby undermining intra-group processes of restraint. 26 Such interplay between conflict-level and group-level dynamics can be seen in each of our three case studies. In the extreme right case, the BNP's brief electoral success in 1993, when they succeeded in electing a local councilor in east London, encouraged and empowered the party's "modernising" faction, prompting the party leadership to prioritize electoral activity over street activity, the latter of which was phased out from 1994. The BNP leadership also sought to stem the appeal of its small but militant radical flank, C18, by proscribing the group, albeit this in effect led to C18 becoming more violent as the influence of the "modernisers" receded. 27 In the animal liberation case, a perception among activists during the early 1980s that there was widespread public support for and media interest in their cause prompted a major strategic innovation away from violence. This comprised the emergence of regional Animal Liberation Leagues that sought to deprioritise the use of clandestine raids on animal laboratories and storage facilities, in favor of mass daytime invasions, characterized by significantly reduced levels of property damage or theft. 28 These actions resulted in mass arrests and heavy sentences, however, leading those involved to conclude that such tactics carried too high a cost, which resulted in this innovation being largely abandoned as the movement swung back toward more clandestine modes of action. 29 The jihadi case meanwhile provides a useful illustration of how ostensibly similar developments within the broader conflict dynamic can have contrasting effects in terms of intra-group processes. In the U.K. there were long established brakes within Al-Muhajiroun relating to practical concerns about the impact that illegal activity would have on the group's freedom to operate and proselytize. These were swiftly undermined, however, when, after IS's declaration of a caliphate, members decided that being seen to be consistent in their support for forces establishing a new caliphate outweighed such concerns. Yet elsewhere within the jihadi milieu the growing prominence of more violent groups, such as IS, had quite different effects. For example, it actually encouraged the Al-Qaeda leadership to emphasize the moral boundaries of legitimate violence in order to demarcate themselves from their new and increasingly influential rivals. These contrasting views regarding how to interpret and respond to IS's escalation of violence created confusion among the grassroots activists within the Telegram group seeking to support the jihadi cause described above. 30 It is clear therefore that in order to undertake an effective analysis of intra-group processes of restraint, we must situate these within a broader conflict analysis. 31 Yet intra-group processes of restraint also intersect with individual-level processes-with how individual group members think, emote, and reflect on their actions and those of their co-members. And while individual experience is likely to be influenced by their acculturation within the group, it is unlikely to be a mere mirror or extension of the group. Even in groups where members largely break-off external social ties, there will still be differences in group members' lived experience reflecting, among other things, activists' personal biographies 32 and their different pathways into that activist milieu, 33 which are likely to translate into different attitudes and preferences regarding their strategies and tactics.
Again, such intersections between group-level and individual-level processes can be seen in each of our case studies. In June 1989, one extreme right activist took part in a BNP-organized attack upon a community meeting at Welling Library, south London. The attack hospitalized seventeen people. The use of substantial interpersonal violence in itself wasn't significantly outside of the normal, but the fact that the majority of those at the meeting were women represented an important breach of the movement's moral norms, in which violence against women was usually associated with cowardice. It also represented a breach of this individual's sense of self. His participation in the assault gave rise to shame and guilt: I began to realize that this was what race wars were about, the innocent attacked and their dignity destroyed. If my mother had known, she would have disowned me on the spot.
[ … ] I was a fucking coward to have done such a thing. 34 In this case, these feelings did not dissipate-"Still Welling Library played on my mind … Did we really attack a meeting of women and gleefully report and celebrate it?." 35 Eventually, such feelings would propel him out of the movement and lead him to cooperate with the anti-fascists against his former colleagues. In the meantime however, this individual process of reflection and change also impacted on intra-group dynamics, at least within his immediate circle of contacts. The activist continued to be part of the extreme right, and indeed continued to engage in and sometimes enjoy the serious interpersonal violence that entailed, yet he gradually disengaged from such activities and was reluctant to overstep his personal moral boundaries in this way again, thereby contributing to the reimposition of limits on what constituted appropriate violence-at least at an individual level. 36 In the animal liberation case, the letters pages of movement magazines provide a glimpse of how similar personal reflective processes contributed to the brakes on violence within this movement. Particularly in the wake of actions that went beyond established action repertoires, such as the use of explosive devices in Bristol in 1989 and 1990, many letter writers expressed shock, disappointment, frustration, condemnation and disgust, and in doing so helped to maintain or reestablish norms within the milieu regarding the parameters of what was or was not appropriate. 37 In the jihadi case, as IS moved toward greater, more gratuitous and less targeted violence, some members of the Telegram group expressed distinct unease about their willingness to declare other Muslims nonbelievers (a concept known as takfir), and about IS's public execution of people considered to be innocent of any crime that would justify such violence. For example, some members of the group recoiled at the murder of Alan Henning, a British humanitarian aid worker who had set out to help those affected by the war in Syria. In doing so, they shaped the emergent normative order of that particular discursive space, infusing discussion of those particular developments with moral doubt, even as they still moved toward planning acts of terrorism. 38 As such, as well as situating intra-group processes within the context of broader conflict dynamics, it is also important to conceive of how they influence and are influenced by individual-level processes of decision-making and reflection. We propose a simple framework with which to do this ( Figure 1). Within this framework, developments at the more macro levels are conceived of as conditioning evaluations, opening and closing opportunities, structuring incentives and disincentives, or generating expectations at the more micro levels, but not determining what happens there. If new state powers are introduced restricting the opportunities for legal protest, it does not follow automatically that activists will reach for more violent strategies of action, but it is likely to translate into reduced opportunities for movement moderates to advocate for less violent strategies of action, and increased opportunities for radical flank actors to justify escalation as a strategy of last resort. 39 Similarly, intra-group management of the action repertoire does not determine how group members will act, but can raise or lower the costs of, or remove or create incentives for, individuals to engage in escalated violence. If an individual anticipates that the use of escalated violence will result in sanction or moral disapproval from fellow, and especially high-status, group members, they will be more likely to choose to refrain from such violence 40 and indeed to adopt and internalize such norms. 41 Yet processes at the more micro levels also shape practices, or at least intentions, at the more macro levels. Where, for example, an individual has resolved that they do not feel comfortable deploying violence against a particular category of persons, then they are likely to be reluctant to do so, and their nonparticipation in, or even questioning the appropriateness of, such violence in turn shapes emergent normative orders at the group level. Similarly, while the effectiveness of intra-group processes of restraint might be compromised by developments within the wider conflict dynamics, such as the ongoing escalation of repression 42 or the emergence of out-bidding dynamics, 43 intragroup management of the action repertoire is likely to have significant implications for how activists respond to the actions of state actors or opposition groups. This might be, for example, by shaping their perceptions of what comprise appropriate or suitable responses to escalated state repression, or shaping the sorts of capabilities they have to respond to escalated violence by their opponents.
In practice of course the boundaries between these levels are somewhat fuzzy. Returning to the case of the extreme right activist mentioned above, his visceral reaction to the attack in which he participated is partly the outcome of a process of individual moral reflection. It is, however, also shaped by wider group and societal norms about the illegitimacy of violence against women. Similarly, clear distinctions between what comprises intra-group management of the action repertoire and what comprises part of the wider conflict dynamics are likely to be disrupted as in-group and out-group boundaries shift: as allies become rivals or opponents, and so forth. Yet this framework is intended to be orientative rather than definitive: a way of beginning to think through and interrogate how and where processes of restraint emerge, operate and evolve.
We argue that this model, while simple, is useful in two ways in particular. First, it is intentionally agnostic about the origins of processes of restraint: it makes no a priori claims about what comprises the "primary 'location' of causality." 44 This helps to clarify that when we discuss "internal brakes" or "intra-group brakes" this does not imply a claim about the causal roots of these phenomena. Brakes are "internal" or "intra-group" only insofar as they comprise practices through which group members seek to shape the actions of other group members, but it is assumed that the causal roots of these practices are complex and diffuse.
Second, by keeping present the interactions between the conflict-, group-and individual-levels, this basic framework helps to avoid generating an account of restraint that either unduly privileges "internalist" perspectives, 45 or is overly deterministic or topdown and leaves insufficient space for individual agency. 46 While certain configurations of external conditions might make the application of brakes appear "logical" to outside observers, they do not determine that those brakes will be applied: rather, group members must apprehend and internalize those logics. 47 Likewise, the fact that some group members, even influential ones, might try to encourage restraint, does not guarantee that activists across the movement will conform to those requests, or that external actors will perceive and appreciate the restraint being exercised by the group and respond accordingly. 48 How Brakes Operate at Different Levels of Proximity to "The Action" In each of our three cases studies, we found examples of internal brakes being applied across a range of different settings. Within the jihadi case, this included high-level internal strategic debates within Al-Qaeda's international leadership, within public communications, in internal discussions among Al-Muhajiroun activists, and in private conversations as members of the Telegram network exchanged and discussed information about national and international campaigns and planned their own activities. Within the extreme right case, we found brakes operating in the BNP's strategic planning, their official publications, and internal communications, during and immediately after physical confrontations between extreme right militants and their various opponents, during informal one-to-one or small group conversations among activists, and within the context of activists' private reflections. And within the animal liberation case, we found examples of brakes being applied in public communiques, during debates that filled the pages of movement magazines, instruction manuals, in the deliberations and declarations of movement leaders, as well as in the course of actions such as arson attacks, laboratory break-ins and hunt saboteuring.
In order to generate a "thicker" 49 description of these brakes, and how processes of restraint emerge, diffuse and operate or, conversely, where, when and how they fail to diffuse or become embedded, we wanted to find a way of describing and categorizing these different contexts. To do this, we developed a conceptual model comprising four different types of contexts, characterized by their relative temporal and physical distance to instances of escalation beyond the parameters of broadly established action repertoires ( Figure 2). This model is loosely inspired by two related bodies of literature. The first is the literature that develops and deploys the "flashpoints" model to understand and explain instances of public disorder, and draws attention to how such violence emerges across multiple levels of "structuration": structural, political/ideological, cultural, contextual, situational and interactional. 50 The second is the literature that examines the role of micro-situational interactions in shaping the dynamics of political violence. 51 From the flashpoints model, we drew the basic idea that the emergence of violence-or in our case restraint-can be understood as a multi-layered process that both conditions and is constituted by human action. From the research on the role of micro-situational interactions in shaping the dynamics of political violence, we drew the basic ideas that micro-situational interactions can develop their own logics that are sometimes distinct from those of the wider conflict of which they are or will become a part, and that micro-situational encounters can, through a series of feedback mechanisms, alter the subsequent dynamics of the wider conflict.
Within our model, the first and broadest type of contexts we refer to as the movement building contexts. These are all of the settings in which group members are engaged in activities that might be crucial to the development of their movement or struggle, but do not comprise undertaking or actively planning a specific action or campaign. This would include activities-speeches, meetings, publications etc.-in which group members are involved in developing and setting out their general political or ideological positions, strategic positioning, developing general claims about whose interests they represent, articulating their goals, fostering solidarity and enmities, recruitment campaigns, and undertaking general management tasks related to movement organization, discipline, membership and finance. Within these settings, there is likely to be a great deal of discussion and deliberation pertinent to establishing and maintaining the outer limits of their action repertoire, but this will tend to be about the development of principles and norms that are generalizable across multiple actions, and will largely be infused with the logics inscribed within fairly high-level strategic planning, the elaboration of movement philosophies and identities, and jockeying for position among movement factions and their leaders. Much of the discussion among British jihadi sympathizers on Telegram, for instance, revolved around the notion of legitimate targeting and, in particular, excommunication: when it applied, to whom and the broader implications of declaring other believers, including close family, nonbelievers, with members variously inspired by pro-IS or Al-Qaeda sources, or torn by which position to adopt. 52 The second type of contexts we refer to as action planning contexts. These are the settings in which group members attend to the specifics of developing and undertaking a particular action or campaign: identifying targets, considering logistical arrangements in progressively greater detail, allocating roles to different group members, seeking permissions or the blessings of relevant movement authorities, deciding whether or not to inform or engage with public authorities about their plans, discussing thresholds for the abandonment of the action, ruling out certain courses of action, and so forth. Here, broader principles about what comprises appropriate modes of action intersect with practical considerations and the looming reality of the planned action. Taking again the example of the Telegram group of jihadi sympathizers, as some of them moved closer to planning a terrorist attack in London, the more abstract debates about what comprise proportionate and legitimate targets, which had focused attention on members of the police or armed forces, rather than the general public, evolved into more focused discussion about specific targeting decisions and hostile reconnaissance. 53 The third type of settings are the situational contexts in which group members are directly engaged in the action, carrying out their plans, and reacting and adjusting to the situation on-the-ground-as Mike Tyson observed, "everybody has a plan until they are punched in the face." Here, broader principles and established action plans and preparations intersect with the micro-situational logics of the interactions and confrontations that emerge within the actions themselves. This might result in their move toward violence being checked by "situational tension and fear" 54 or accelerated through "forward panic" dynamics, 55 or the weakening or strengthening of "moral disengagement" 56 as individuals are confronted with the immediate reality of their actions. While not part of our case studies, some of the most striking examples of restraint emerging at the situational level can be found in the testimonies of individuals who had planned to undertake mass casualty attacks but found that they were unable or unwilling to go through with it. Kerry Noble, for example, was a leading figure in the Covenant, Sword and Arm of the Lord group who, in 1984, was on the cusp of bombing the Metropolitan Community Church in Kansas City because, as he notes in his memoir, it was a "gay church." He "hesitated" long enough to take stock of his surroundings, which caused him to re-humanise his potential victims. Having come to see them as no different from himself in that split second, he abandoned his plans. 57 Similar examples can be found amongst failed Palestinian suicide bombers, some of whom expressed regret about their involvement after being arrested or talked of their desire to abandon their missions after they were launched. 58 The fourth set of contexts are what we refer to as immediate post-escalation contexts. As the names suggest, these are the contexts in the immediate aftermath of instances of escalation above and beyond "normal" violence-where the brakes in effect have failed-and where group members are reacting to that escalation. In each of our case studies we found a particular proliferation of what look like brakes within these contexts. In the extreme right case we have already described how the sense of shame and disgust experienced by one activist emerged after the attack on the group at Welling Library, and how that effected his and, less directly, his fellow activists subsequent behavior. Similarly, despite his reputation for violence, after participating in an attack on "left-wing paper sellers on Brick Lane," one National Front activist involved with C18 activities, reportedly "threw up down a side alley." He later told a fellow activist that the level of violence used against a target who were unable to defend themselves (i.e. left-wing paper sellers rather than anti-fascist militants) "went too far." 59 And while both of these examples arguably speak largely to processes of individual reflection, within this case study we also find more obviously intra-group processes of criticizing and pulling back from further violence, such as in the immediate aftermath of David Copeland's nail bombing campaign in London in 1999. 60 In the jihadi case study, as noted above, we see how members of the Telegram network recoiled at the killing of a humanitarian aid worker. Internal memos from the Al-Qaeda leadership, seized during the May 2011 raid that killed Usama bin Ladin, also reveal how attacks on places of worship and commercial centers by local affiliates and allies, such as the Pakistani Taliban, prompted mounting concern among senior leaders not only about undermining public support, but also about the legality and proportionality of the attacks, resulting in public communiques intended to rein in these allies. 61 And in the animal liberation case, we can observe in movement magazines the swift and widespread condemnation of the use of explosive devices after incidents that clearly exceeded established repertoires of action.
There are a number of possible explanations as to why we see so much braking taking place after instances of violent escalation. There are likely to be particular legal and reputational incentives at these times for individuals and groups potentially associated with such escalated violence to criticize or distance themselves from it: which raises challenging questions about how one deciphers more or less sincere criticisms of escalated violence within post-escalation contexts. It is also however likely to be a function of the personal and intra-movement turbulence that escalation beyond established repertoires of action can generate as group members, both individually and collectively, seek variously to understand, justify or distance themselves from the actions that have taken place. 62 What is most important for our purposes is that these reactions can have an impact on the subsequent trajectory of violence-a fact that is not surprising if one adopts a processual understanding of conflict "in which antecedent events condition, shape, or 'cause' subsequent events." 63 In the extreme right case, there was no repeat of David Copland's killings; in the animal liberation case, there was no repeat of the use of the type of explosive devices used in 1989 and 1990, even as activists at the radical flank of the movement continued to use incendiary devices; and in the jihadi case the activists in the Telegram network focused their attention on targeting members of the police and armed forces after the killing of Alan Henning had met with condemnation. In other words, far from being simply a case of closing the stable door after the horse has bolted, the post-hoc application of brakes in the immediate post-escalation contexts might actually be key moments in shaping the trajectory of future escalation, de-escalation or non-escalation.
The application of brakes within any of these types of contexts can affect the likelihood of further brakes being applied subsequently in other contexts. Perhaps most obviously, brakes applied within movement building contexts can in effect cascade down into action planning and situational contexts, and into immediate post-escalation contexts. Indeed, within our three case studies the brakes applied within action planning and situational contexts often worked on brakes already developed within movement building contexts, such as when group members invoked the established strategic or moral logics of their wider campaigns when seeking to maintain discipline during event preparation or execution. Yet as the example of Kerry Noble illustrates, restraint is not always or necessarily rooted in broader processes of movement development, but rather can seemingly emerge spontaneously within action planning or situational contexts out of the emergent logics of those particular settings, or through individual responses to the evolving situation. Furthermore, and as noted above, instances of escalation and non-escalation are themselves likely to generate feedback loops that alter how group members prepare for subsequent actions and influence wider movement debates about their aims, objectives, purpose and modus operandi. 64 The way in which brakes might emerge at different levels of proximity to potential acts of escalation, and how these in turn can shape subsequent processes of restraint, can be seen in the animal liberation case. Here, movement building contexts are positively saturated with the deployment of brakes that work on moral, ego-maintenance and out-group definition logics, in the form of repeated insistences on their commitments to nonviolence and the reduction of suffering for non-human and human animals. There is also the frequent application of strategic brakes in the form of discussions about the need for a broad-based under-and over-ground movement in order to achieve their goals. 65 While some radical flank activists loosened these brakes-by, for example, arguing that their campaigns of intimidation did not constitute violence, or that their violence was an action of last resort 66 -the brakes developed and applied within the movement building contexts nonetheless informed how activists across the movement planned and carried out their actions. For example, during their planning, even at the most radical fringes of the movement activists accessed and shared information about how to effectively minimize the risk of inadvertent harm to animals, both non-human and human, during arson attacks and other actions; and when the Animal Liberation Leagues sought to innovate away from clandestine raids to mass daytime invasions, some activists engaged in training exercises to help maintain nonviolent discipline during these actions. 67 Similarly, activists often reminded one another during actions not only of the personal consequences of illdiscipline, but also how it might undermine their cause and advance that of the animal abusers and/or the state security forces. 68 Yet within the same case study there are also examples of restraint emerging out of micro-situational contexts. It seems, for example, that at some hunts, hunt saboteurs and supporters achieved a form of negotiated equilibrium in which both pursued their goals while largely avoiding physical confrontation or maintaining any confrontation at a low level. Indeed, there are anecdotes about hunts at which it was not unheard of for participants and saboteurs to drink peaceably in the same pubs after the hunt, 69 and even of hunt saboteurs receiving a ride home from hunt participants with whom they had become acquainted. 70 The fact that saboteurs perceived there to be less risk of serious physical assault at those hunts meant they were less likely to perceive it necessary to "sab" those hunts en masse in order to protect themselves, or to do so in revenge for previous skirmishes, 71 which in turn reduced the probability of escalated confrontations, thereby generating virtuous cycles that likely go some way to explaining the marked difference in violence across different hunts, even within a few dozen miles of one another.
As before, putting this basic model into practice poses some challenges. One of these again relates to the fuzzy lines between these categories. Even as activists make plans for a specific action, they are likely still to be engaged in processes of constructing the wider frames through which they interpret and organize their struggle. And it can be difficult to identify the point at which action planning actually becomes being "in the action": is it on the morning of the planned action, when they arrive at the place where they planned to undertake the action, or only when they are in the act of tipping over a rival group's newspaper stand, opening the roof of the animal laboratory or arming the bomb at the checkpoint? Indeed, for more detailed analyses of specific cases, further layering might be advantageous, e.g. distinguishing between contexts in which activists are planning specific actions and the slightly higher level planning of campaigns of actions, such as a series of bombings, a series of provocative or intimidatory marches; or a series of raids on animal laboratories.
Another challenge relates to data access. While in most cases researchers or analysts are likely to be able to access a fairly substantial amount of data relevant to the movement building and immediate post-escalation contexts, action planning often takes place in less accessible settings, particularly where the planned actions are illegal or are more likely to be effective if they exploit the element of surprise. It can therefore be difficult to locate data relevant to such practices, even retrospectively. Analysis of the situational contexts can pose a similar problem.
These are not insurmountable problems, however, and in practice researchers and analysts will usually find themselves working with imperfect datasets. What this way of approaching the brakes on violent escalation does is help to capture and give analytical prominence to the dynamic nature of restraint, and to the way that, like violence itself, restraint also "develops in action." 72 This is likely to be important if we are to generate effective explanations for how and why, and the conditions under which, certain brakes work and fail.
It also holds out the possibility of being able to sharpen our ability to assess emergent risks of violent escalation. If we can see clear evidence of brakes being applied within movement building contexts, those seeking to apply brakes closer to the action will have more to work on, and it is likely that, if escalation does happen, the costs in terms of movement cohesion will be higher as it will clearly undermine those who sought to apply the brakes. Similarly, the presence or absence of brakes being applied in the immediate post-escalation contexts is likely to provide valuable clues about whether the escalated violence that has taken place either becomes the "new normal," or remains an outlier in the movement's history.

The Mechanisms through Which Intra-Group Brakes Generate or Enable Restraint
The third contribution we make in this article is to set out four types of mechanisms through which intra-group brakes generate or enable restraint. In doing so we add a further layer to our understanding of how these brakes work and fail. While the descriptive typology we presented previously identifies the logics on which these brakes operate, it tells us little about the mechanisms through which these brakes, once applied, serve to generate restraint among militants within the group in question.
Based on analysis of our three main case studies and on the wider literature on the dynamics of escalation, de-escalation and non-escalation we identify four types of mechanism through which intra-group brakes work. The first are disciplinary mechanisms that work through the making and issuing of rules and guidance about what group members should or should not do, and the enforcement of those rules through sanctions and rewards. Examples of rule-setting were among the most abundant and easiest to find brakes in each of the case studies, observable within multiple public statements, internal communiques and guideline documents. In the jihadi case we see this sort of rule-setting in the issuing of specific "rules" and "guidance" by Al-Qaeda's leaders articulating the boundaries within which fighters were permitted to operate, including the rules of engagement and targeting. 73 These included Ayman al-Zawahiri's General Guidelines for the Jihadi, published in 2013, that included clear limitations in terms of legitimate targets, and were later ridiculed by the IS group for their alleged leniency and deviance from religious law. 74 In the extreme right case, the BNP issued an Activists Handbook which laid down a series of guidelines for personal conduct, and other groups also published their own codes of conduct, though these were often honored in the breach rather than the observance. In the animal liberation case, the ALF Credo, which states that anybody can call themselves an ALF activist provided that they abide by a series of basic rules, is perhaps the most obvious example of such rule-setting.
What were less apparent in our three case studies were examples of the enforcement of those rules. There are a number of factors that might explain this. In part it might reflect the group structures and power dynamics of the groups under analysis. In the jihadi and animal liberation cases we were looking at fairly acephalous groups with limited or very limited formal command and control structures. In the extreme right case, the highly fragmented and politically marginal nature of the movement, and the apparent draw of the radical flank to much of the rank-and-file membership, meant that the authority of BNP leaders was precarious; efforts to impose discipline upon rank-and-file members for transgressing party diktats risked alienating activists, the lifeblood of the party, who could very well de-camp to rival groups if they deemed the sanctions unduly harsh. 75 The relative absence of data pertaining to rule-enforcement might also reflect a methodological challenge: the distribution of sanctions and rewards is likely often to take place behind closed doors, therefore making it more difficult to identify suitable data, and is also likely often to combine more formal disciplinary procedures with more subtle forms, such as informal praise or scorn, prestige or disgrace. 76 As a result such processes can be rather diffuse and difficult to pin-down to specific incidents.
Nonetheless, some examples of (attempts at) control through sanction and reward can be identified. In the jihadi case, leaders admonished-privately and publicly, explicitly and implicitly-subordinates and affiliates, stretching from attempts to control Abu Musab al-Zarqawi's autonomous affiliate in Iraq in 2005 to efforts to limit the violence of Pakistani affiliates and, later, by jihadis fighting on different sides after civil war broke out in Syria. 77 In the animal liberation case, enforcement of the movement's rules often came as much from the grassroots of the movement as it did from movement organizers, such as, for example, in the criticisms and rebukes that appeared within the letters pages of movement magazines described above. In the extreme right, the BNP leadership eventually proscribed C18, although, as described above, this inadvertently undermined the ability of movement moderates to exert any control or influence within the milieu's radical flank, and therefore arguably constitutes an example of a braking attempt that backfired, or at least displaced violence to another part of the "movement." 78 One noteworthy characteristic of these disciplinary mechanisms is how often they appear to work through establishing structures of conditionality around the use of violence. That is, they restrict violence by setting out the conditions under which greater violence would be or is legitimate, but then argue that such conditions have not yet been met. 79 So, for example, violence might be constructed as appropriate if and only if it is undertaken in self-defense or as a last resort, 80 or if it only affects a specific "legitimate" target. 81 These structures of conditionality then enable criticism, and in some cases other forms of disciplinary action, if these conditions are not met: as happened when key figures within the extreme right voiced their disapproval of David Copelands' nail bombings as "misdirected mayhem," or on the grounds that the time simply was not right for armed struggle. 82 The use of such structures of conditionality has a number of potential benefits for movement organizers. By not categorically ruling out more violent courses of action, it potentially facilitates within-group compromise and cohesion whilst also effectively forward-proofing one's legitimacy as a leader, and leaving escalation on the table can serve as an effective device with which to threaten, intimidate and bargain with opponents. 83 It also creates opportunities for escalation, however, if the situation changes and radical flank actors can argue that the conditions for violence have been met, as happened in the case of Al-Muhajiroun's embrace of greater violence after the declaration of a caliphate by IS.
The second type of mechanism through which intra-group brakes work is by shaping group members' perceptions of the nature of their struggle in ways that constrain opportunities for violent escalation. This might be about shaping how group members conceive of themselves (e.g. in the case of ego-maintenance brakes), or their opponents (e.g. in the case of brakes based on out-group definition); about how they perceive what is possible and what is most likely to be effective (e.g. in the case of strategic brakes) or what is or is not at stake (e.g. in the case of moral brakes).
In the animal liberation case for example, this type of mechanism can be found operating in the efforts by movement moderates to inhibit the move toward more radical strategies of action by persuading fellow activists that there was broad public support for their cause, and by attempting to construct a movement theory that emphasized the need for a mass movement: something that often ran contrary to radical flank activists' self-image as members of a revolutionary underground. In the extreme right case, this type of mechanism can be found in BNP organizers' efforts to persuade the rankand-file to re-imagine themselves not as a street movement locked in violent struggle with anti-fascists, but as a viable, if revolutionary, political party able to gain support with a substantial minority of the general population and achieve change through electoral means.
Whether or not such mechanisms work is likely to depend to a large degree on the ability of those articulating these visions of their struggle to persuade other members of the group. This in turn is likely to be a function both of the credibility and influence of that individual within the group and, as discussed above, of wider conflict dynamics 84 : e.g. the BNP organizers' ability to persuade members and supporters that they were a viable political party was given a major boost by their initial electoral success in 1993.
A third type of mechanism through which these brakes work is by shaping group or individual capabilities to undertake violent escalation. This mechanism is most apparent in what we have previously described as "organisational brakes," where group members either invest in, or divest from, capabilities that would enable the group to carry out acts of (greater) violence.
In the extreme right case for example, as the BNP pivoted toward electoral campaigning, they began to lose the support of some of their previously more able and violent street fighters, and began to attract support from people with less ability or inclination for participating in physical confrontations with anti-fascist activists, making the prospect of returning to such modus operandi an unappetizing prospect. In the animal liberation case, even as opportunities for disruptive yet legal protest became increasingly limited, activists did not move to stock-pile weapons or high explosives, thereby reducing further the probability that they would seek to use lethal force.
This mechanism speaks to elements of path dependency whereby strategic or tactical decisions made at one point in time shape the options that group members have in subsequent moments. 85 Applying restraint, particularly at an early stage of an organization's life cycle, can serve to inhibit the acquisition of capabilities for violent escalation 86 or can foster the augmentation of capabilities for less-violent strategies of action, which in turn make the use of more violent strategies of action less appealing as the comparative benefits of deploying violence are diminished.
These first three mechanisms broadly align with the concepts of intent and capability often found in the literature on risk assessment. 87 The first two mechanisms relate to militants' intent to undertake greater violence; the third relates directly to capability. The fourth mechanism that we propose offers what we believe is an important augmentation to thinking about how these brakes might work: this is by constructing opportunities to favor restraint without losing face.
While popular and policy discourse about radical milieus often imagines them comprising individuals with a heightened, sometimes almost all-consuming, appetite for violence, such a view is not well-supported by the empirical evidence. While some undoubtedly are attracted to such milieus by the prospect of violence, Simi and Windisch,88 in their research on why people within white supremacist sub-cultures in the United States do not do as much violence as they might, observe for example that "[c]ontrary to the common perception that extremists are 'crazy' individuals determined to kill as many innocent bystanders as possible, our data suggest extremists struggle with the idea of taking another person's life." As such, while it might be reasonable to expect members of militant groups to be more accustomed to and prepared for violence than the average member of the general public, most will still experience the "situational tension and fear" that arises as acts of violence loom into view, 89 whether that is a result of the corrosion of their moral disengagement, 90 having doubts about the moral logic that would enable them to construct that violence as appropriate or necessary, 91 or a more visceral upsurge of fear or apprehension. However, significant social pressures can make it difficult for individuals to draw back from violence. Indeed, in some groups ongoing participation in violence is a prerequisite for acceptance. 92 We propose that another way that the intra-group brakes on violent escalation can work is by countering pressures to engage in or advocate for greater violence, creating instead opportunities for group members to encourage or exercise restraint without compromising their position with the group or jeopardizing other "critical relationships." 93 For example, when group members have a narrative available to them about the importance of maintaining group discipline in order to achieve their strategic objectives, or about the moral value of restraint and mercy, it becomes easier for them to draw back from violence, regardless of what their actual motives are for doing so. 94 Similarly, Ellefsen and Busher 95 describe how, during the latter stages of the Stop Huntingdon Animal Cruelty campaign, at a time when state repression had generated widespread fear within the activist community, an emergent discourse about the importance of building public support provided a face-and morale-saving way for some animal liberationists to endorse and encourage a "strategic" shift toward less militant modes of activism.
We note that it can be difficult to find evidence of this mechanism at work within movement publications or indeed in activist memoirs, not least because, due to reasons of personal pride and possible reputational costs, most activists are likely to be reluctant to discuss or acknowledge such processes. We can catch enough glimpses of this mechanism however to believe that it is worth our attention, or at least warrants further investigation through, for example, detailed interview-based or ethnographic research. This is also a mechanism described in other contexts of violent escalation and non-escalation. In programs to reduce gang violence in U.S. cities, for example, creating viable "outs" or "offramps" for those who might otherwise be drawn into violence-enabling individuals to identify and choose alternatives to violence while maintaining reputation and honor-has been identified as a particularly effective intervention strategy. 96 Taking into consideration this fourth mechanism would suggest that whether escalation does or does not occur is not just about whether or not group members have an appetite for and capability to undertake greater violence, but also about the extent to which, if and when they begin to feel uncomfortable with the violence in which they are (about to be) engaged, they have available pathways away from violence that do not imply serious loss of face. If this is the case, it suggests that serious consideration should be given both to understanding how group members create and maintain viable pathways away from violence, and to how external interventions might strengthen or erode these pathways.

Conclusions
Developing a stronger understanding of how militant groups and their members develop and deploy restraint, and how these "internal brakes" work and fail, responds to an important limitation in our knowledge regarding the dynamics of political violence, and potentially offers important benefits to scholars, policymakers and practitioners alike. The identification and analysis of instances of "push back" against moves toward greater violence from within the group or movement, or of innovation away from violence, can provide particularly valuable diagnostic moments with which to better understand the extent and nature of the threat that groups pose, the current state of intra-group relationships, and the extent and nature of opportunities to engage with or intervene in such groups in ways that bolster restraint and, ultimately, lead away from violence.
Yet brakes sometimes fail. Even when the leaders of a group encourage restraint, this is no guarantee that rank-and-file members across the movement will conform to those requests, or that efforts to limit escalation will embed across a group or movement at the same pace. If we are to take advantage of the analytical benefits of being attentive to processes of restraint, we must develop a robust conceptualization of how these brakes work and fail.
This article has made three contributions to that end. First, it has offered a framework for situating the "internal brakes on violent escalation" in relation to the interplay between conflict dynamics, intra-group processes and individual-level decision making. Second, it has developed a basic analytical strategy for examining how these brakes operate at different levels of proximity to potential or actual instances of escalation. Third, it has described four types of mechanisms through which internal brakes appear to generate or enable restraint.
Our intention is not to argue that all attempts to examine intra-group brakes need necessarily deploy all three of these frameworks. We would argue however that holding them in mind and using them to structure and inform data collection, collation and analysis, can help move us collectively toward a more detailed, systematic and theorized understanding of processes of restraint within militant groups. In this respect, it is hoped that this article provides a much needed contribution toward establishing the kind of conceptual foundations that are required if this avenue of research into the internal restraints on violence within militant groups is to prove fruitful over the coming years.