Compromising Aid to Protect International Staff: The Politics of Humanitarian Threat Perception after the Arab Uprisings

Scholars expect operational compromises by humanitarian organizations to follow attacks on aid workers. However, in response to the War in Syria, organizations compromised aid and adopted clandestine, cross-border, remote management, and conflict-actor aligned approaches, which best protected international aid workers. This was despite declining rates of attack against them, relative to their national staff counterparts. This article asks why international aid workers were withdrawn and aid was compromised in the wake of the Arab Uprisings by traditional risk-taking organizations: Médecins Sans Frontières (MSF) and the International Committee of the Red Cross (ICRC). Drawing on political ethnography and interviews with aid workers, I show that shocking violent events, everyday insecurity, and changes in the nature of threat have significant effect on threat perception and explain compromises where rates of attack do not. This paper offers a picture of the micro- and field-level foundations of organizational threat perception and decisions about whose security matters. Resumen Syrie, les organisations ont compromis l’aide et ont adopté des approches clandestines transfrontal-ières de gestion à distance alignées sur les acteurs du conflit, ce qui a permis de protéger au mieux les travailleurs humanitaires internationaux. Cela a été fait en dépit de la baisse des taux d’attaque à leur encontre par rapport aux attaques subies par leurs homologues du personnel national. Cet article s’interroge sur les raisons pour lesquelles les organisations prenant traditionnellement des risques ont retiré des travailleurs humanitaires internationaux du terrain et compromis l’aide suite aux soulèvements arabes : Médecins Sans Frontières et Comité international de la Croix-Rouge. Je m’appuie sur l’ethnographie politique et sur des entretiens avec des travailleurs humanitaires et je montre que les événements violents choquants, l’insécurité quotidienne et les changements de nature de la menace ont un effet considérable sur la perception de la menace et expliquent les compromis alors que les taux d’attaques ne le font pas. Cet article offre une image au niveau micro et au niveau du terrain des fondements de la perception des menaces par les organisations et de leurs décisions relatives à ces questions de sécurité.


Introduction
In the years that followed the 2010 self-immolation of Tarek el-Tayeb Mohamed Bouazizi in Tunisia and the Arab Uprisings, international non-governmental organizations (INGOs) and international organizations (IOs) altered their preferred modus operandi. They adopted clandestine, cross-border, remote or blind management, and conflict-actor aligned approaches to aid delivery at an unprecedented level, primarily increasing protections of international workers. From the perspective of IOs and INGOs themselves, these were compromises adopted in a new security environment in the Middle East. In the case of response to the War in Syria, IOs removed international expert support of overwhelmed local systems and global witnesses to targeting and atrocities. This occurred while international workers were less likely to suffer kidnap or killing than the nationally hired colleagues that took their places. This article shows that when organizations withdrew international staff in the wake of the Arab Uprisings, they compromised aid. It asks why new compromises emerged and whose security matters in the aid world.
Taking IOs and INGOs as its unit of analysis, I define compromise as aid operations carried out at a reduced INGO or IO standard. An INGO or IO delivers humanitarian assistance in ways that do not align with its own best practices and standard operating procedures. While circumstances of the War in Syria prohibited measure-ment of the quality of aid received inside of Syria for this study, it is reasonable to assume that where organizations report compromises to their ways of doing business there will be compromises to aid.
Of course, changes not desired by these organizations will not have uniformly negative effects on humanitarian assistance, which is aid meant "to save lives and alleviate suffering of a crisis affected population." (ReliefWeb 2008) There is evidence that local NGOs and civil society groups grew during the Syrian conflict (Ruiz de Elvira 2019) and that local organizations provide more responsive aid than international ones (see Booth and Unsworth 2014). The pulling back of international IOs and INGOs may have created room for local actors to receive funding, build capacity, and direct programs. However, despite commitments to localization made at the World Humanitarian Summit in 2016 (Wall and Hedlund 2016), IOs continue to rely on internationalized approaches to aid delivery ( de Geoffroy, Grunewald, and Ní Chéilleachair 2017;Metcalfe-Hough et al. 2019) and their added value remains cross-context expertise. Removing international aid workers in the midst of conflict response undermines standard aid delivery. What is more, the compromises to aid I explore transferred a great deal of risk and few protections to locally hired staff. This suggests that local staff were hired as risk-takers and means to gain access rather than potential leaders of humanitarian response (Khoury and Scott 2021;Gingerich and Cohen 2015). Finally, compromises may have prevented complete organizational withdrawal. By this logic, increased protection of international staff is a necessary adaptation in an insecure environment. However, it is not clear that the extent of adaptations was justified by the risks. Other options were available to organizations, inclusive of better protecting all staff and sharing risks among international and local staff (see Poole 2014;Stoddard 2020).
International security literatures lead us to expect that operational compromises will follow increased attacks on aid work (Abiew 2012;Burkle 2005;Hoelscher, Miklian, and Nygård 2015) and scholars of aid expect compromises to protect a more threatened class of aid worker (Fast 2014(Fast , 2010Patel et al. 2017;Rubenstein and Bittle 2010;Stoddard and Harmer 2010). An increasing politicization of aid, particularly in Iraq and Afghanistan in the early 2000s-wherein aid is used for political ends and armed combatants do not see aid workers as distinct from military interveners-has been linked to the targeting of aid activities and foreign workers, in particular (De Torrente 2004;Hammond 2008;Donini 2010). Yet, contrary to expectations, rates of international worker killing and kidnapping as a proportion of staff in the field and as compared to their national colleagues have been in relative decline since the 2000s. 1 Incidents ending in death disproportionately affect national staff-staff hired locally-who are employed by international NGOs and IOs. Data from Insecurity Insight's Security in Numbers Database (SiND) show national staff fatalities accounted for 19 percent of aid worker staff fatalities between 1996 and 2000 but 71 percent between 2006 and 2010 (Wille and Fast 2013b, 9). 2 In the same time periods, international staff fatalities as a proportion of all staff fatalities declined from 35 to 15 percent (Wille and Fast 2013b, 3). When the War in Syria broke out, international workers were similarly protected. Fatalities and kidnappings among foreign workers accounted for 5 percent of all aid worker fatalities and kidnappings in Syria in 2012. This rose to 21 percent by 2014 and then fell to below 3 percent through 2018 (Humanitarian Outcomes n.d.). These figures include incidents suffered by local Syrian NGOs and CSOs.
Efforts to protect aid workers from harm are not new. As early as the 1990s, UN General Assembly Resolution 1 Wille and Fast report changes in proportion of attacks over time where reporting biases are likely to hold relatively constant (2013a, 3). See also (Stoddard, Harmer, and DiDomenico 2009, 3). 2 See also Hammond (2008) on sharp increase in national staff incidents.
182/46 emphasized the right of humanitarian agencies to safely access peoples in need. In this period, "corridors of tranquility" in Sudan, "safe havens" in northern Iraq, Bosnia and Herzegovina, and Rwanda, and cross-border operations from Kenya into Somalia were meant to secure humanitarian access (Hoffman and Weiss 2006, 93;UNHCR Central Evaluation Section 1994). During operations in Afghanistan and Iraq, a sense that the humanitarian space was shrinking led to an increasing "bunkerization" of aid (Duffield 2012;Egeland, Harmer, and Stoddard 2011;Hilhorst and Jansen 2010;Smirl 2015)the use of "harder" security measures such as armed guards. In fact, international workers have been protected by a system of legal and political "exceptionalism" (Fast 2014) and a "hierarchy of humanity" that deems some aid worker lives more worthy than others (Fassin 2011;Sweis 2019). What is new is a further ramping up of protections for international workers after 2010 in and around Syria. This was after the pendulum of protection had swung far in international staff favor and appears to have come at the expense of national staff lives. This article explores why aid was compromised in the wake of the Arab Uprisings. It considers why these compromises best protected international aid workers, even as locally hired staff were involved in a greater proportion of violent incidents. Findings have implications for our understanding of how threat perception onthe-ground affects organizational behavior, why risk is transferred from international to nationally hired staff in global humanitarianism, and questions of whose security matters. I find, first, that when shocking violent events occur, there are everyday experiences of insecurity, and an organization places value on risk-taking acts, staff are more likely to believe that threats are on the rise. This is the case even when data indicate otherwise. Second, in the wake of the Arab Uprisings, a perception that the nature of threat changed strongly determined if we saw new or more compromised approaches. New types of threat moved aid workers across a security threshold and toward new or significantly expanded tactics of selfpreservation. Findings tell us that principled humanitarian organizations adopt discriminatory security practices, despite the high standards they set for themselves. In response to particularly violent events, they over protect the international staff they expect to be most threatened and offer insufficient protection to local aid workers. This is the case even when they are presented with evidence that disconfirms expectations surrounding who is under threat.
This study does not aim to resolve these issues. Rather, it unpacks the logics that drive organizations to assess threats and mitigate risks in particular ways, as well as Downloaded from https://academic.oup.com/jogss/article/7/1/ogab024/6372895 by guest on 13 March 2022 their impact. Findings suggest a more critical examination of security practices is needed within humanitarianism. They highlight contradictions between these practices and humanitarian principles that call for solidarity with distant others and the neutral provision of aid, without regard to race or national identity.
In the next section, I discuss the theoretical contributions of this work to scholarly understanding of how micro-and field-level decisions can influence broad organizational behavior and the ways studies that show intuition, narrative, and culture shape threat perception are applicable to aid work. I discuss the possibility that similar compromises may be occurring in contexts outside of Syria. I then review "normal" risk-taking behavior and present a first set of empirics outlining the compromises to aid observable in the midst of the New Arab Wars (Lynch 2016). I consider potential explanations for additional protections of international workers. Next, I outline my case selection and design. I then propose an explanation for diminished risk-taking by humanitarian organizations based on a second set of empirics and, more specifically, on an investigation of two medical humanitarian organizations' security practices after 2010: Médecins Sans Frontières (MSF)/Doctors Without Borders and the International Committee of the Red Cross (ICRC). In a concluding section, I discuss the moral and ethical trade-offs faced by aid workers who assess risk, and point to implications for future research.

Contribution to Theories of Organizational Behavior and Threat Perception
Findings from study of the humanitarian response to the War in Syria are relevant to theories of threat perception and humanitarian organizational behavior in other contexts. Since the 2010 earthquake in Haiti there has been growing interest in data-driven humanitarianism (Meier 2015;Read, Taithe, and Mac Ginty 2016). In line with this trend, the ICRC's Healthcare in Danger Project (ICRC n.d.), Insecurity Insight's Humanitarian Data Exchange (OCHA 2010), and Humanitarian Outcome's Aid Worker Security Database (AWSD) collect data, provide analysis, and make recommendations about threats and risk-taking. However, I find data-driven analysis had little sway over organizational decisions about risk-taking in the Middle East after 2010.
Instead, my research shows that individual experiences and perceptions, and particularly brutal, cruel, or humiliating events shape, not just individual, but organizational behavior. Building on studies of threat perception, I find that humanitarian organizations favor what Slovic et al. describe as instinctive and intuitive reac-tions to danger (Kahneman and Frederick 2002;Slovic, Monahan, and MacGregor 2000) and that their perceptions of risk are most influenced by events with strong "imagery, story and narratives" (Lichtenstein et al. 1978;Loewenstein et al. 2001;Slovic et al. 2004, 317). They base their decisions about where to protect and where to delegate risk on field worker experience and perceptions of security. When attacks on aid work are particularly public or attached to strong imagery (Hendrickx, Vlek, and Oppewal 1989;Tversky and Kahneman 1973) a sense of threat rises, even in the face of contradictory information. Existing narratives and cultural beliefs (draws on Kahan, Jenkins-Smith, and Braman 2011) about threat overwhelm more deliberative analytical systems of judgment. Organizational theorists should therefore expect humanitarian INGO and IO beliefs and behaviors surrounding threat to be particularly hard to shift with new information. This builds on existing aid literatures that show that the politics of the everyday make peacebuilding organizations likely to ignore contradictory information (Autesserre 2014) and that aid workers face barriers to learning in conflict settings (Campbell 2008;Terry 2002;Walkup 1997).
My findings also shed light on the politics of threat in other conflict settings. Since 2010, the number of incidents affecting aid workers has been highest in Afghanistan and then, in descending order, South Sudan, the Syrian Arab Republic, Somalia, and Sudan. International workers accounted for between 7 and 14 percent on average of all aid worker fatalities across all of these contexts (Humanitarian Outcomes n.d.). These rates were lowest in Syria, then Afghanistan, and then Somalia. This suggests that international workers may have been most protected where tactics of killing or kidnapping were widely used or reported. As I discuss in the conclusion of this article, my research suggests that violent actor targeting of international workers is effective in changing IO and INGO behavior. In these ways, my conclusions are likely to be somewhat generalizable.
Interestingly, there is also the possibility that the War in Syria produced unique behaviors that are unlikely to travel, or "a Syria effect." The idea that something is different for humanitarianism in the Middle East is compelling because it captures aid organization discomfort with operating in the Arab world, where there is a history of Islamic philanthropy and limited acceptance of humanitarian organizations as principled actors who are separate from politics (Barnett 2009;Benthall and Bellion-Jourdan 2009;Davey and Svoboda 2014). What is less compelling is the notion that humanitarian worker discomfort with local groups was somehow different than in other contexts. While the murder of George Floyd and growing support for the Black Lives Matter movement have produced some willingness to discuss the role of racism in humanitarianism, and in blocking attempts to "localize" it (Barnett 2021;Cornish 2019;Slim 2020), my research demonstrates just a few of the reasons why this discussion is so badly needed. Humanitarian actors in the Middle East were perceived by IOs and INGOs as particularly deviant, norm defying, or even-with aid workers using the racist, colonial language of humanitarianism's forebears-"barbaric" (Maurer 2017;Shaheen 2016b). There was a willingness to transfer risks to national staff. These phenomena are likely to be at work more broadly.

"Normal" Risk-Taking and Mitigation
Before turning to an empirical look at compromises to aid and what they look like, this section provides a theoretical discussion of "normal" IO and INGO risk-taking and mitigation behaviors, and how they change. Humanitarian organizations adopt principles that act as broad guides for their behavior. These often include providing assistance based on a common humanity and mutual respect (Barnett 2011;Slim 2015) and impartiality, wherein aid is provided based on "need only" and not religion, gender, race, nationality, or political opinion (ICRC 2016). Organizations also commit to solidarity with the people who receive assistance and the national societies or staff they employ or partner (ICRC 2016;Slim 1997). In adopting these principles, organizations set a high standard for their own behavior and the behavior of aid workers, inclusive of risk-taking. To stand in solidarity with those in need, aid workers often need to move closer to the dangers that produce crises.
A range of mitigation strategies are adopted. According to UNOCHA, operations in complex security environments should include "acceptance-building: local outreach and sensitization, promoting community ownership, [and] strengthen[ing] responsibilities of national staff" (Egeland, Harmer, and Stoddard 2011;OCHA 2011, 1). Standard acceptance-based security measures also include relationship-and network-building with communities, promoting local knowledge of organizational principles, standards, and offerings, and working on more than one side of conflict to avoid projecting partiality. Time spent with local actors and potential recipients of aid 3 improves understandings of patterns of con-

3
I use the word "recipient" here for the purposes of clarity. Reasonable critiques of the term point out that those who receive aid are not passive bystanders. See Autesserre (2014, 84-90) and Campbell (2018, 20).
flict, power dynamics, or social service offerings, while allowing for relationship-building and the identification of safe ways of doing business in a particular context. However, these approaches to security also expose IOs and INGOs to potential contextual risks, such as when there is state failure, a return to conflict, a new wave of natural disaster, attack, or illness (Metcalfe, Martin, and Pantuliano 2011). IOs and INGOs tolerate these kinds of dangers to varying degrees, with most organizations recognizing that staff increase their exposure to threats through aid work. A second set of security measures are protectionbased. Protection by way of compound walls, guards, or convoys, maintaining a low profile, or remote work is omnipresent in aid. However, the United Nations and security experts encourage gaining acceptance through "presence, time, and sustained engagement with all parties." (OCHA 2011, 2) This is because when aid workers are made more remote, acceptance is undermined by limits on access, local engagement, and context-specific know-how. This can also contribute to gaps in service and poorer-quality activities. The behaviors discussed in this article are designed to protect specific individuals by hiding them within or removing them from a particular context. Humanitarian organizations have traditionally used protection-based measures less and as a last line of defense before withdrawal, and normally when workers, facilities, or recipients are targets.
However, as is illustrated in the case studies below, these less desired approaches were adopted in the Middle East on a significantly expanded scale after 2010. They best protected international workers who were already facing dramatically reduced risks, heightening contradictions between security practices and principles of solidarity and non-discrimination.

Compromising Aid
Operational compromises to humanitarian activity in the Middle East since 2010 include: clandestine or underground, cross-border, remote or blind management, and conflict actor-aligned operations. In this section, I define these approaches and identify their new or increasing use by drawing on data from interviews, ethnography, as well as organizational reporting. It would be difficult and inadvisable to quantify these operations, which are regularly concealed for security reasons. Based on a number of examples in Lebanon, Jordan, and Syria, I provide a picture of each kind of approach and its increasing or novel use by MSF or the ICRC after 2010. I then explain how the approach compromises aid in the name of security. I focus specifically on compromises during health operations by MSF and the ICRC because organizational data and analysis of attacks on aid were made available in this sector through MSF's Medical Care Under Fire Project and the ICRC's Health Care in Danger Project.
First, clandestine operations are a means to avoid targeting by operating without the knowledge of states and non-state parties to a conflict. In the 1980s, MSF's Founder Bernard Kouchner spoke of a right of doctors to reach populations in need, even against sovereign state wishes (Brauman 2012). Clandestine operations can range from removing identifying emblems during movement to moving hospitals underground. Here, attacks are less likely because visibility is diminished or damage less likely if bombs fall. This kind of operation not only compromises the quality of aid due to difficulty of recipient access, barriers to resupply, and diminished connection to other facilities and experts, but it also undermines transparency and coordination while encouraging duplication. As one respondent stated, "nobody knows who is doing what, where." 4 It also cuts short the abilities of organizations to learn about and negotiate with local networks and power structures.
MSF uses clandestine operations based on moral pragmatism (Rubenstein 2015;Slim 2015). Brauman writes, For MSF, as soon as it becomes thinkable, and hence possible, to set up medical services in an area controlled by an opposition force, it is necessary to do so. If it is deemed useful to set up illegal medical services, then the only considerations that are taken into account are practical ones (whether a neighbouring country will authorise access, whether there is a liberated area, whether credible partners can be identified). Website updated since data were pulled in 2017.
MSF also supported underground and improvised medical facilities in government-controlled areas, with a focus on besieged areas. Operating theaters and entire hospitals were moved "below the earth" while patients in above-ground beds were supposed to be stable enough to move "when the shelling restarts." (MSF 2015) As of 2017, support was provided to hospitals and small health posts in Deraa, Hama, Homs, Idlib, and rural Rif Damascus where, as a general picture, the Syrian state exerted significant control or conducted offensives (MSF 2018). According to the organization, "Almost all of the field hospitals supported by MSF in Homs and elsewhere in Syria have sustained damage from airstrikes and barrel bombs." (MSF 2015) Historically, sharing GPS coordinates was a means to protect medical aid and a cornerstone of MSF relations with warring parties, even in recent conflicts in Iraq and Afghanistan. However, MSF stopped reporting the location of its supported hospitals in Syria as of 2016 (Shaheen 2016a). Between 2012 and 2017, the acknowledged presence of MSF international staff in the country went from limited to zero. A second compromise to standard modes of IO and INGO operation is a cross-border approach. This approach can also be clandestine, without knowledge of the host or receiving state. The novelty of cross-border activity in response to the War in Syria is underlined by the UN adoption of Resolution 2165 in 2014, which authorized the delivery and monitoring of aid across "conflict lines" (Security Council 2014). This occurred after a great deal of lobbying and advocacy by NGOs and was a first for the Security Council because it allowed organizations to enter Syrian territory without the consent of the state. Aid operations were carried out in northern Syria through Turkey and in southern Syria through Jordan, with whispers about some operations through Lebanon. Cross-border activities produce compromises to standards of operation because INGO or IO managers are rarely present or visit project sites infrequently. National staff or local NGO or civil society organization (CSO) partners carry out relief activities, and the organization often lacks the in-country strategic networks necessary to help them oversee activities safely. In addition, resources can be diverted to neighboring host states, which may demand services in exchange for the use of their territory. 6

6
INGO Interview, Anonymous Fieldworker C120, Irbid Jordan, April 2016. Interviews listed as with "fieldworker" are with those who work predominantly at project sites and in the community spaces that surround them. Interviews at the "country-level" were conducted at the highest office in a particular state, generally a Downloaded from https://academic.oup.com/jogss/article/7/1/ogab024/6372895 by guest on 13 March 2022 Cross-border operations into Syria by MSF proceeded with less in-country oversight than had been historically accepted. For example, when MSF withdrew from Somalia in 2013, its management teams based in Kenya had access to established networks and to Somali staff that remained on the ground. 7 However, in Syria the INGO lacked these established networks (Robitaille and Hemily 2014) and could conduct only short and infrequent monitoring trips. It had trouble ensuring funds and supplies reached desired partners or were used for desired ends. The closure of the borders with Syria by Lebanon in 2014 and Jordan in 2016, as well as the tightening of border controls and the blocking of aid routes with Turkey in 2016 (Fordham 2016), further complicated cross-border operations and reduced MSF access to key areas.
The ICRC claims never to use clandestine methods and to act only with state consent (Brauman 2012). While there are historical cases of the organization hiding its emblem as a means to delivering aid (Moorehead 1998), the ICRC treats state sovereignty and the need for state permissions prior to interference as near sacrosanct. This is the case even where that state lacks territorial control. In effect, by adhering to principles of sovereign state territoriality first, the ICRC makes decisions about aid second and only after upholding its moral commitment to state consent. This becomes, on its own, a kind of compromise. However, this approach can sometimes give the ICRC a stronger long-term foothold for aid delivery because sitting governments often hold onto power.
Third, remote management refers to the use of partner organizations or national staff who are not managed in-person by organizational leadership to run projects. Both MSF and the ICRC use remote management (Donini and Maxwell 2013). By 2015, MSF began supporting the underground networks, described above, through management teams based elsewhere. The INGO supports national organizations who have their own networks through funding, in-kind donation, and some management. 8 MSF does not publicly identify those groups due to concerns for their security. In a novel move for the INGO, MSF also began hiring Syrian nationals remotely in 2016 and delegated much of the work of delivering aid to them. 9 Meanwhile, the ICRC worked through the Syrcapital city office. Some interviews are listed only by month, rather than day, to protect anonymity. During response to the War in Syria, loss of Syrian doctors who were targeted or forced to flee along with rising needs, meant that additional international support was required. Blind and remote management approaches reduced substantive and technical support from internationally trained surgeons and doctors. It was limited to mostly financial, in-kind, and/or distanced virtual guidance. Notably, remote or blind management can be combined with clandestine, underground, and cross-border operations. This is likely to compound the negative effects of compromise on the aid delivered.
Fourth, IOs and INGOs committed to neutrality, which refers to not favoring one side in a dispute, compromise when they work with one party to a conflict in order to facilitate operations ( van Mierop 2015). In Syria, the ICRC maintained relations with the Assad regime and accessed non-state-controlled territory only when having negotiated state permission. The ICRC in Syria operates through the SARC, which is arguably aligned with the Syrian regime (Beals and Hopkins 2016; The New Humanitarian 2012; see also Leenders and Mansour 2018). ICRC respondents argue that while an initial refusal to act without state permission limited their services in non-state areas, by maintaining relationships with the state they secured increasing permissions to deliver aid, and to more regions, as the War in Syria evolved. 11 There are examples of ICRC convoys reaching oppositionheld areas as early as 2012 in, for example, Homs and Harasta (ICRC 2012 served opposition-controlled areas. This is the case both because convoys delivered goods and supplies rather than comprehensive care to these areas through "truck and chuck" humanitarianism 12 and because their reach was limited due to their government-controlled starting points. What is more, opposition forces are less likely to negotiate with an actor they see as aligned with the Syrian regime because this is their point-of-entry. As the IS is driven out of its final strongholds in Syria and the Assad regime reasserts control over most of Syria's territory, we see the ICRC's past compromises yielding returns. It has better access than many of its counterparts. In contrast to the ICRC, MSF chooses to seek permissions when they are a means to deliver aid but often ignores them when permissions are barriers. For example, MSF has been open about compromises it made to gain access in negotiation with both the Sri Lankan state between 2006 and 2009 and the Taliban in 2008 (Magone, Neuman, and Weissman 2012). At the beginning of the Syrian War, MSF negotiated for access to oppositioncontrolled zones through non-state actors. In 2012, their negotiations with the Assad regime effectively ended. A member of the team sent by MSF to Damascus reported that MSF was its own "worst enemy for gaining official access to Syria." 13 While some MSF sections were launching activities in opposition-held areas without state permissions, others were negotiating with an Assad regime that viewed these moves as provocations and as "proof" that MSF was untrustworthy. 14 This diminished its capacity to provide aid in regime-controlled zones.
The modes of operation outlined are compromises to the standard ways IOs and INGOs deliver aid. Discussing these compromises, a senior security advisor said, It's not about protecting the institution itself. It's about protecting the capacity of the institution to maintain operations. The problem is that when there is insecurity and specifically [an] incident, it will have, not only an effect on the image of the organization, but on its actual capacity to operate.

Existing Explanations
A first potential explanation for compromise is that aid workers are naturally risk-averse and willing to compromise the value of aid where there are threats. One picture of an aid worker is of a high-paid expatriate or "armchair" humanitarian, 16 whose days are spent safely behind walls. There is some truth to this image. Foreign staff are paid at rates that far exceed their local counterparts and humanitarian workers are often kept at a distance from local people (Smirl 2015). However, aid organizations working in the field or from European or North American headquarters also accept reputational (Ewins et al. 2006) and security risks (Krause 2014). MSF's Charter and Principles state, "As volunteers, members understand the risks and dangers of the missions they carry out and make no claim for themselves or their assigns for any form of compensation other than that which the association might be able to afford them." (MSF n.d., 1997) An ICRC aid worker stated, "Security is high on the agenda of the agency. We lost already a lot of people. We don't want to lose more. We know it's part of our contract. We know that we risk our life, but it's a calculated risk." 17 This calculation is part of crisis management and response, and also key to deciding when to move, where, and for what. Because those in need of humanitarian aid tend to be isolated, persecuted and denied, or targeted for violent attack, humanitarian organizations are wedded to the violence that makes them necessary.
In fact, proximity-or closeness to conflict, crisis, and their victims-is operationally and culturally valued. First, it is necessary for needs assessment. INGOs delivering service require local-level, granular and context-specific information in order to stay relevant to both potential recipient populations and the actors that control them (Honig 2018; On service-delivery see, Murdie 2014; on accountability see, Gizelis and Kosek 2005;Barnett and Walker 2015). This is a cornerstone of acceptance-based security. In conversation with one ICRC project leader, he said of risks faced by his organization, I'm willing to go straight to the beneficiary, and this will save us, if not maybe from the bombs, but at least from the population, will even maybe protect you, because we do good things for them. But if we do less and less that, and more and more politics, then you put yourself in danger. 18 Relatedly, being known to a local population and its leaders for service in one issue area can be essential for future activity in another. For example, direct medical aid to a population that knows the ICRC, facilitates activities in International Humanitarian Law (IHL) under the Geneva Conventions. A senior ICRC delegate in Lebanon explained, Will it or do not will it, like it or do not like it, health activities are a vector to let many other ICRC activities be accepted. No country around the world will accept ICRC detention visits. Putting our nose inside the prisons of a country where things do not work. We are accepted doing many of our activities, including our mandate, because we offer putting something on the plate. … If we keep reducing these activities we will be less and less accepted. 19 Second, proximity is also embedded in the values and culture of humanitarianism (its ethos) and in the principles of right and wrong that guide organizational behavior (its ethics). INGOs and aid workers often describe a desire for "presence," or being operational in an area that is experiencing violence or morally important harm (see Rubenstein 2008, 38;2015, 152). This is not about harm minimization per se, as outlined by political and moral philosophers, such as Singer (2004) and Pogge (2004). Rather, organizations aim to make political statements and establish organizational relevance by acting in areas that they feel deserve a place on the global stage. At times, a desire for presence can take prominence in justifications for aid activity. It is also central to organizational principles that call for solidarity with conflict and crisis-affected populations. As one respondent stated, "In terms of remote approaches, it's completely contradictory to our principle of being on the spot." 20 Thus, while humanitarian organizations seek to mitigate risks, self-preservation has always been weighed against operational and cultural desires to be close to crisis and conflict, as well as potential recipients of aid. This suggests they are not naturally risk averse.
A second and related potential explanation for compromise is that as rates of attack increase, aid organizations will compromise aid activities to protect themselves. International security literatures and scholars of aid observe targeted and large-scale attacks against aid workers (Fast 2010;Patel et al. 2017) and study how security measures protect them (Van Brabant 2000). Where medical aid is concerned, attacks refer to "bombing, shelling, and looting of facilities or transports," violence against workers, takeovers of hospitals by militaries, "fighting in and around hospitals," and "obstruction of access to healthcare, medicine, and essential supplies" ("No Protection, No Respect" 2016). Thus, we are seeing attacks when shells or bombs fall and when aid convoys are not allowed to enter besieged towns. We might expect that security practices will change and that compromises will increase as attacks increase and, at first glance, this seems to be the case. When examining attacks in absolute terms using the AWSD, Stoddard, Harmer, and DiDomenico report that aid workers were killed, kidnapped, or injured at increasing rates between 1997 and 2008 (Stoddard, Harmer, and DiDomenico 2009). Scholars basing findings on the SiND agree that, "Without doubt there has been an increase in the number of aid workers killed, kidnapped, and injured" (Wille and Fast 2013b, 6). Estimating a denominator based on increased aid workers on the ground, they hold that even taking a growing humanitarian footprint into account, rates of attack are on the rise. While there is overreporting of recent incidents due to more common reporting practices, we can say with some confidence that attacks are increasing over time.
Yet, fatalities among international staff as a percentage of staff on the ground dropped well below the rate of those suffered by national staff in the decade before the Arab Uprisings that began in 2010. In fact, Stoddard, Harmer, and DiDomenico conclude, […] A trend in increasing casualty rates for national (locally hired) staff, relative to their number in the field, compared with international (expatriate) staff… was attributed to organisations' increased use of remote management and outsourcing of aid delivery in dangerous environment… (2009, 3) Wille and Fast draw on SiND data to show that over three time periods-1996-2000, 2001-2005, and 2006-2010-international staff fatalities as a proportion of all aid worker deaths declined from 35 percent to 21 percent to 15 percent, respectively, whereas national staff deaths increased from 19 percent to 66 percent to 71 percent of fatalities (2013b, 9). International staff were victims of kidnapping across the same time periods at a rate of 5 percent, then 7 percent, then 14 percent, whereas national staff experienced these incidents during the same periods at rates of 26 percent, 46 percent, and 63 percent (ibid.). Contrary to expectations, new and expanding measures were adopted to protect international aid workers while rates of attack against them were in relative decline.
Thus, by 2010 and with the beginning of the Arab Uprisings, the burden of targeting and incidents in aid delivery was already being borne most by national staff and was only growing. Yet, new and more compromises that best protected international lives were adopted.

Case Selection and Methods
To examine why organizations compromised aid operations and in new or expanded ways during response to the Arab Uprisings and War in Syria, I investigate these behaviors in two cases: the ICRC and MSF after 2010 in Syria and neighboring states. I focus on response to the War in Syria because it has typical features of contemporary complex humanitarianism, including exceptionally high global interest and funding, as well as rapid and growing involvement of state and non-state humanitarian actors in response. It also produced massive forced displacement and refugee need and was the kind of morally and politically important crisis that made continued presence on the ground very desirable to IOs and INGOs. Compromises occurring where there was exceptional interest in being present and after 2010, when rates of attacks against internationals had declined, are particularly interesting.
I examine behaviors by two organizations-the ICRC and MSF-with missions and objectives that make them among the least likely to compromise operations that are proximate to recipient populations. Compromise should be particularly hard in these cases because these organizations are exceptional among humanitarian organizations in their standard willingness to accept risk. The tolerance of both is traditionally high in the midst of conflict and violence because both aim to access populations in hardto-reach spaces and to work on both sides of conflicts. They are a small universe of particularly risk-tolerant cases. As a result, determining what causes this kind of organization to compromise aid can provide insights into behaviors at less risk tolerant organizations and a basis for measured inference (Levy 2008). Variation in ICRC and MSF approaches to delivering assistance also leads to variation in the approaches these organizations are willing to consider, which helps me to identify those mechanisms and factors that lead to different types of compromise.
To evaluate my argument against the dominant explanations related to risk tolerance and rates of attack, I observed the operations of both organizations in response to the Syrian War using political ethnography (Schatz 2013) in Lebanon and Jordan for ten months in 2016 and 2017 when access to Syria for IOs and INGOs from these countries was limited. I tracked the movements and decisions of both organizations as they worked inside Syria through press briefings and through conversations with aid workers in the Middle East and European headquarters and conducted over 120 interviews in 2016 and 2017 with aid workers involved in operational decision-making, as well as security experts at both agencies. Interviews concerned why and how organizations made decisions about the kind of aid delivered to Syrians and where. I focused particularly on the processes and mechanisms driving decision-making, with security and threat as a potential driver. Interlocutors were selected from among both international and national aid workers engaged in operational decision-making surrounding response to the War in Syria and the Syrian refugee crisis, as well as from among organizational and regional security experts. I used the snowball method to expand the sample (Tansey 2007). Interlocutors gave informed and voluntary verbal consent. They were also provided opportunities to withdraw that consent over time, particularly as security conditions and reputational risks could change in the research setting. Participants were given anonymity if they requested it or if I determined their security could be threatened by my revealing their location or identity. I also conducted a thorough review of the literatures and analyses by Insecurity Insight, Humanitarian Outcomes, MSF's Medical Care Under Fire project, and the ICRC's Health Care in Danger Initiative (ICRC 2011). I consulted some of the authors of these reports and studies. 21 In my analysis of primary organizational sources, I identified the factors and traced the processes that led organizations to use compromised modes of operation back to their source. I focused on how new behaviors were justified by aid workers and organizations, and the extent to which these fit into a broader narrative about rates of attack and incidents affecting particular types of aid workers. I also assessed the extent to which these constituted compromises for an INGO or IO. To do so, I looked for moments when workers expressed discomfort with modes of operation, described them as different or new, and/or questioned how justified they were.
Regarding my methods of comparison, I use withincase comparison of changes in ICRC and MSF approaches to aid delivery and adoption of compromises over time. Findings are organized around those mechanisms that lead organizations to compromise, with across-case comparison of MSF and ICRC behaviors used to illustrate how they played out.

Findings
"We are like cockroaches. I mean we are much more numerous now than we used to be in the past. So our exposure to risk is higher…" 22 Aid workers involved in this study expressed beliefs that the threats they faced exceeded those faced by their predecessors and that, under IHL, they should be exceptions to violence as bringers of aid to the theater of war. When international staff were confronted with the possibility that risks to international aid workers were less than those faced by national staff counterparts, they expressed discomfort and confusion. They also maintained that the environment was uniquely dangerous to them. In this section, I show that compromises by MSF and the ICRC that best protected international aid workers, even after they became a more secure and protected class of aid worker, followed from tendencies among aid workers to perceive incidents of violence as if part of a trend, to allow everyday generalized insecurity to magnify threat perception, and to repeat and underline stories of risk and violence. These forces exaggerated a sense that international aid workers were most at risk. I then illustrate that it was not changes in the rate of attack against international workers, but perceived changes in the nature of incidents and violence in the Middle East that triggered new and more frequent compromises and promoted expanded international staff protection.
First, I find that compromises by aid agencies in the Middle East and after 2010 occur because exceptional violent events shape organizational and staff beliefs in a wider trend. Some attacks "shock the conscience" of organizations (Terry 2013, 24) and aid workers interpret threats based on the marks left by one or two events. This observation aligns with the expectations of psychologists who show that human beings are primed to look for change as a means to survival (Manstead, Frijda, and Fischer 2004;Panksepp 2005;Slovic et al. 2004). Things that look different trigger defensive postures because changes in our environments are common sources of threat. Therefore, shocking events become part of the discourse of violence and shared understanding of its prevalence.
For example, in 1996 the ICRC released this statement, "The International Committee of the Red Cross (ICRC) is in shock. Six of its delegates were coldbloodedly shot dead this morning by unidentified gunmen at their quarters at the hospital in Novye Atagi, near Grozny, at 4 a.m. local time." (ICRC 1996) The ICRC withdrew from Chechnya within hours. In 2003, "a flatbed truck loaded with approximately one thousand kilograms of explosives raced down an access road in Baghdad" next to the Canal hotel (Fast 2014, 16). The explosion coated five miles surrounding the area in debris (Power 2008), killed the UN Secretary-General's Special Representative and fourteen others, and wounded 160 people. This incident undermined the assumption that aid workers were protected. In January 2014, the IS in Syria abducted MSF staff members; five international staff members were held for several months. This occurred despite assurances that the organization's operations would not be targeted. One expatriate explained, That fundamentally changed our operational ap-proach… we had had security assurances from them. I've seen the pieces of paper with the stamp and everything, saying we're allowed to work freely, but obviously those assurances were not kept so we made the difficult decision to withdraw from Islamic Statecontrolled areas. 23 Each of these events was raised repeatedly by interlocutors from various aid organizations to illustrate the changing shape of risk during aid operations. Shocking instances primed individual decision-makers to behave based on assumptions about greater risk and exposure. These crowded out alternative views. Second and relatedly, everyday exposure to risks and violence leads humanitarians to feel threat as a baseline. Ample evidence emerged during interviews and observation that regular incidents have deep effects on aid worker senses of threat; these tend not to be reported to media sources or deeply problematized by organizational leadership. Take, for example, the humanitarian worker who was threatened by prisoners while doing protection work at a prison and then told by his employer to return without additional protective measures. 24 I observed medical personnel intervene in a conflict between two patients from two warring groups sharing a hospital room. A fieldworker reported that the IS contacted their organization warning the team to get an expatriate who had
left hospital grounds off the streets, or else. 25 These stories were sometimes told to illustrate the strength of an organization that takes risks. At other times they were condemnations of the insecurity faced in day-to-day humanitarian activities. Aid workers cope with threats as a matter of daily life and tend to draw on personal experience in support of a belief that threats are changing. Importantly, international worker experiences of generalized insecurity are more commonly reported (Stoddard, Harmer, and DiDomenico 2009, 3) and feature more prominently in organizational discourse (Benton 2016), encouraging a sense that they are more threatened. Third, the effects of violence on the behavior of the aid community are made more powerful because organizations reward risk-taking, exposure, and survival. Humanitarian workers gain legitimacy when they can claim to have worked alongside, for example, dangerous, non-state armed groups (NSAGs) such as the Taliban, al-Shabab, Jabhat al-Nusra, or the IS, or strong and defiant states. 26 These claims operate as social currency, connecting staff to in-groups and signaling knowhow. Interlocutors emphasized risk-taking, security constraints, and closeness to danger in discussions with one another and, in particular, when they wished to influence an operational decision. For example, when talking about cross-border operations from Jordan into Southern Syria, where threats from terrorist organizations were a concern, expats illustrated their point using past experience working alongside groups, such as al-Qaeda or the Taliban. 27 This gave weight to their proposal. In these ways the influence of an attack is enhanced in the retelling and by an organizational culture that values risk-taking.
Humanitarian organizations are experiential learners (Kolb 2014), and this makes the intuitive and instinctive reactions to danger that Slovic et al. highlight highly effective in shaping humanitarian behavior (2004). In part because data about medical or humanitarian needs, risks, and outcomes are often unavailable or unreliable during conflict or disaster, organizations and their staff learn and interpret information through their subjective experiences. Honig calls this "navigation by judge-  (2018). ICRC field-based staff described carrying out scouting missions and regular observation of needs as the basis for organizational activities. 28 At MSF the act of witnessing or "témoignage" refers to the role of the fieldworker in observing need among local people and embedded in local networks, in order to shape organizational direction (Redfield 2013, 116). These organizations value proximity and witnessing in knowledge formation and foreground what they learn in these ways in decision-making. In fact, as conditions worsened inside of Syria, an inability to get close enough to learn and gather information "caused a lot of frustration at the level of delegates saying, 'We should stop health activities here. We are blocked. . .'" 29 Staff were concerned by their inability to get a clear picture of need and to respond appropriately in absence of it: "We don't go into where they live. . . we're not walking and drinking tea amongst the community and meeting community elders and talking about what their needs are. . ." 30 In the Middle East and during the New Arab Wars, the rate of attack against international workers did not need to increase, relative to national staff counterparts, for organizations to believe they were most at risk. Instead, the mechanisms outlined-violent or shocking events, insecurity experienced in the everyday, and stories of violence told and retold-allowed this belief to take hold, even where there was evidence to the contrary. What was more, there was a perception that changes in the nature of threat to ICRC and MSF aid workers had changed. Staff believed that threats in the region were more brutal, cruel, and humiliating and that international workers were more likely targets. This moved organizations across security thresholds and toward compromise and is observable along three dimensions.
First, humanitarian actors no longer believed that access and security could be "guaranteed." Interlocutors drew on examples from Iraq and Afghanistan to illustrate that even though aid had been politicized in these contexts, potentially putting humanitarian workers at higher risk of attack (see Hammond 2008), the security situation since the Arab Uprisings had further deteriorated. Numerous aid workers recalled the ways their organization was previously able to gain access to NSAG-controlled spaces and highlighted this as a key organizational competency. For example, the Deputy Head of the ICRC delegation in Jordan said, "Take Afghanistan as a very good example. We were working deep, deep inside Taliban areas where nobody else was working. We were also working, doing trainings for the Afghan army and the international forces as well, the Coalition force over there." 31 However, this ability was challenged after 2010. Acting Head of Mission for MSF Holland in Jordan stated, "A humanitarian worker, if he's more useful dead then he will be killed. . . because our death will make a statement." 32 In 2016, MSF France's Head of Mission in Jordan said, Whatever is the party, it can be a state, it can be opposition group, it can be militia-all the guarantees we have are not reliable. And this is becoming a real concern, and I don't think it's because there is more NGOs in the field. It's because the nature of the conflict and the interest they have, make them -just [not] care about collateral damage. 33 There was a sense that state and non-state actors were increasingly likely to choose to hit a medical target instead of protecting humanitarian aid. Threats in the Middle East could therefore not be mitigated by acceptancebased approaches to security. Launches by the ICRC of its Health Care in Danger project in 2011 and by MSF of its Medical Care Under Fire project in 2013 are illustrative of organizational responses to lost security guarantees. 34 Second, a distinction developed for aid workers between the threat of death or of kidnap and the threat of targeted, cruel, and humiliating treatment, with the latter surpassing their security thresholds. A sense that death could serve a performative purpose for armed actors altered the security calculus of organizations. The kind of experiences a captured or kidnapped international aid worker might have and the kind of death they might suffer were regularly reported as a key concern by aid work- considered irredeemably deviant and norm-defiant (on norm compliance see Cortell and Davis 2000) or by UNmember states and their proxies who were backsliding on IHL commitments. They became embedded in the institutional memories of organizations. Attacks on medical care in Syria were predominantly carried out by the Assad regime (Baker and Brown 2015), but UN Security Council Member States, their proxies, and a range of non-state actors were also perpetrators. The question of "who did it" mattered to IOs and INGOs. They shared perceptions of the norm-defiant group as near-impossible to bring into a global community that respects medical space during war and the backsliding group as having forsaken the same community. Both fundamentally threatened the survival of medical humanitarian aid. When these groups failed to behave as expected, organizations became less capable of anticipating behavior and weighing risks.
In 2015 alone, more than 120 hospital attacks were documented in Syria with many of these facilities subjected to the "double-tap" (Wong and Chen 2018). Forces strike, wait for aid personnel to arrive in the wake of the first attack, and then strike again killing survivors and first responders. Yemen's facilities were hit more than 100 times by Saudi-led coalition and Canadian and American-manufactured weapons, as well as opposition Houthi and allied force shelling. Hospitals in Afghanistan were attacked 122 times. Sixty-one facilities in Iraq were attacked ("No Protection, No Respect" 2016). Also in 2015, bombs fell from an American AC-130 gunship on Kunduz Trauma Center in Afghanistan. MSF interlocutors reported that representatives in Kunduz and around the world were in contact with United States and Afghan authorities. Strikes did not stop. MSF President Joanne Liu recalled, We were under attack … they didn't stop, for an hour. And then our patients burned in their beds. Our colleagues tried to save the lives of their colleagues by operating on a make-shift desk office table, and we lost two of our staff while trying to save their life. (Liu 2016) These strikes are widely believed within the humanitarian community to have been pre-planned attacks on the structure (Donnelly 2016;Lendman 2015). They are seen as indicative of a backsliding by the United States and other UNSC members on commitments to IHL and the Laws of War. They led to calls for an independent criminal investigation (Rosenberg 2016).
As UN-member states and their proxies joined nonstate, norm incompliant groups in using attacks against healthcare as a tactic of war, humanitarians saw fresh moral decline and regress. In February 2016, in Ma'arat Al Numan, Idlib province Syria, an MSF hospital was struck by four missiles killing nine members of the hospital's staff and sixteen patients. Kilometers away two other hospitals were reportedly struck in Azaz city (MSF 2016). An MSF representative stated, "The destruction of the MSF-supported facility appears to be a deliberate attack on a health structure," and ". . . leaves the local population of around 40,000 people without access to medical services in an active zone of conflict." (ibid.) Liu warned in 2015 of a return to "barbarian times" (Shaheen 2016b), demonstrating a common feeling among aid workers that the nature of threat had changed. In 2018, the ICRC President made a speech to the UN Human Rights Council that was titled "The laws of war are our shield against barbarity," which was published in the International Review of the Red Cross' Conflict in Syria issue (Maurer 2017). The speech and both leaders' use of the language of "barbarism" reflect beliefs at both MSF and the ICRC that adherence to IHL is waning and that non-state violent actors in the Middle East are stepping almost irredeemably outside of the law.
As the nature of threat changed during the New Arab Wars, INGOs and IOs increased protections of international aid workers and transferred risks to more threatened national staff.

Conclusion
Violence is a social fact of aid work. MSF's charter states that members "understand the risks and dangers of the missions they carry out" (MSF n.d.). Bernard Kouchner wrote of the organization, "We know there will be a price to pay, because nothing big is ever achieved, nothing gets created or accomplished without risk" (Neuman and Weissman 2016, 6). ICRC delegates report their contracts state that their work could result in violent incident, injury, or even death. 37 Yet, humanitarian organizations regularly ask themselves: What risks are tolerable and what level of self-preservation can we accept, where doing so may undermine recipient wellbeing?
Changes over time in if and for whom risks are tolerated have implications, first, on the question of whose security matters in the context of humanitarian intervention (McDonald 2016). Aid workers have rarely been made equal to those they are deployed to help and regularly weigh recipient lives against their own in order to operate. The Geneva Conventions and customary IHL hold that during armed conflicts "health workers, facilities, ambulances, and people who are wounded or sick" must not be attacked ("No Protection, No Respect" 2016). Aid workers are made an exception to violence-able to stand outside the conflict and act as a "special category of civilians deserving attention and protection." (ibid., 6) This may prevent complete operational withdrawal of important humanitarian organizations and negative subsequent effects on potential recipients. However, humanitarian "exceptionalism" also places the security of mostly white European and North American workers above that of other races and nationalities, and so builds on and perpetuates racism in the aid industry (see, McVeigh 2020).
A second and troubling implication is that national staff insecurity is better tolerated by INGOs and IOs than attacks on international staff. By 2010, national staff workers were already almost five times more likely to suffer fatalities than their international counterparts (Wille and Fast 2013b, 10). Yet, organizations continued to transfer risks to them. This suggests that if the goal of aggressors is to disrupt standard humanitarian operations and reduce organizational influence and footprint, attacks on international workers are an effective strategy. They trigger greater operational compromises by INGOs and IOs. As INGOs and IOs differentiate between international and national workers in their security practices and response to attacks, attackers are more likely to do the same. Violent actors can be savvy enough to track the response of international actors to violence (Autesserre 2012;Brandt and Sandler 2009). Where international staff are marked as the higher value target by INGO and IO behaviors, organizations may-paradoxicallyencourage their targeting and more brutal types of attacks against them.
Third, an assumption that national staff are more secure than international counterparts because of an ability to navigate national contexts, access existing networks, and leverage local know-how is questionable. In fact, national staff who are moved from one region of their country to another or who interact with a different, for example, ethnic or religious group can face heightened risk. National staff are also expected to abide by cultural and security standards that international actors can sidestep. A local staff member often gets away with less where, for example, security authorities may use their discretion in allowing an international vehicle to pass through a checkpoint without a search or interrogation. In addition, the notion that national staff are more willing to take on risks because, as one aid worker put it, "these people are Syrians, they are fighting for their country," 38 may discount the possibility that national staff may take on greater risk because they cannot leave. The evidence in this article shows that three factors led IOs and INGOs to compromise in ways that did not align with rates of attack; staff tended to see incidents as if they were part of a wider trend, were primed to perceive new threats because of experiences of generalized insecurity, and exaggerated violent incidents in the retelling. What is more, it was the nature and not the rate of attack that drove new humanitarian compromises in the Middle East. Humanitarian INGOs believed they were operating in a fundamentally changed security environment after the Arab Uprisings because of unreliable security guarantees, narratives about deviant actors or backsliding states, and, most importantly, fear of brutal and humiliating treatment at the hands of perpetrators. These forces reduced risk tolerance and caused organizations to compromise aid.