Data for assistance: in emergencies, consent is a transaction more than a choice. As artificial intelligence tests the limits of humanitarian accountability, three specialists argue real safeguards begin where communities can understand, question and challenge what is done in their name.
True accountability is not just humanitarian organisations holding themselves to a higher standard, but ensuring crisis-affected communities are equipped to hold them to account. Equipping humanitarians to be more accountable in their practices has been an important focus of attention, especially as humanitarian services have made increasing use of digital technology. But, as a sector, we have rarely gone beyond that to equip affected people to influence and challenge decisions about how technology is designed and used. We argue here that inclusion is an essential precondition of accountability in the age of artificial intelligence (AI) and that we need to make sure community members understand and can influence tech solutions from the outset.
Organisations are increasingly drawing on the potential of AI to expand the reach, efficiency and proactivity of humanitarian action. As they do so, the issue of accountability becomes both more urgent and more complex because the technology is untransparent and poorly understood. We conducted a scoping study to find out what challenges community members currently face in holding organisations to account on integrating technology into their services, and what they would need to bridge that gap. We used the findings as the basis for community-facing guidance on digital accountability.
This article draws on CLEAR Global’s interviews and co-creation workshops with practitioners from civil society and international non-governmental organisations (NGOs) working across Africa, Latin America, West and South Asia, and Europe, with support from the UK Humanitarian Innovation Hub (UKHIH) and Elrha.[1]Twelve semi-structured key informant interviews and two co-creation workshops were held online in February and March 2026, in English, Spanish and Arabic. The key informants were six women and six … Continue reading
Participation is the missing first step
Community participation in humanitarian decisions on deploying technology is fundamental to digital accountability, yet it remains the exception. This is despite a growing list of commitments to responsible technology use in recent years, and the few examples of participatory design and evaluation of AI-powered and other solutions which study participants provided.
There are several reasons for the lack of meaningful codesign. Constraints of time and money are, of course, prominent in a context of emergency response and shrinking funding. But we also heard of a lack of expectation that communities should be directly involved in technology decisions – on the part of organisations and communities alike. Participants repeatedly told us that if people trust the organisation, they will trust that the technological solutions it brings will be safe, relevant and useful.
In particular, individuals with less power in their own society typically have little expectation that their voices will be heard. Yet they are precisely the ones that technology applications – AI-powered solutions especially – need to take into account to be fair, unbiased and effective: women, people with disabilities, older adults, refugees and internally displaced people, and marginalised groups of all kinds. For example, the vast majority of training data for language AI solutions comes from those relatively privileged sections of society who are online. That means all other languages, cultures, issues, life experiences and contexts are drowned out. Unless those groups are actively involved in creating and evaluating them, AI-powered tools will never work adequately for them, and could cause them harm. And no one may ever know.
We ask people to trust in solutions neither they nor we truly understand
Without the insight that participating in design decisions would give them, community members often do not see that they have a choice in engaging with humanitarian technology. They ask AI chatbots questions on sensitive topics like health or gender-based violence not knowing that the technology may hallucinate[2]In the field of AI, we call a hallucination a false or misleading response presented as a certain fact, for example a bibliographic reference [editor’s note]. or give them inaccurate or harmful answers. They sign up for mobile money transfers or provide fingerprints and other biometric data on the understanding that this is what they have to do to get assistance. In an emergency, people are often understandably more concerned with getting that support than with ideas of accountability.
But community members typically have very little understanding of the systems they interact with in this way. When those systems use AI components, most users are unaware of them. Many do not understand that what they type or say into a phone does not stay in that device. One practitioner described as “almost non-existent” community awareness of what happens to data once it is collected, how it flows between systems, who can access it, how long it is held, and what decisions it may inform. Beyond the immediate purpose of giving data to receive a service, crisis-affected people rarely have any visibility into what happens next.
This gap is not limited to communities. Data collected through a mobile registration form may be stored across multiple servers, shared with partner agencies or government bodies, integrated into third-party platforms, and retained for years. This is infrastructure that implementing organisations do not own and may not fully understand. Several interviewees reported that even the staff operating these systems could not explain how data is processed or what governance arrangements are in place to protect it. And where organisations rely on commercial platforms, data governance may fall entirely outside organisational control.
Without transparency and understanding, consent is more transactional than informed
Consent is the cornerstone of ethical data practice, and humanitarian organisations routinely implement consent procedures. Forms are signed, verbal explanations are given, boxes are ticked. What the study finds consistently is that these procedures rarely translate into genuinely informed choices for crisis-affected people.
Several intersecting factors explain this gap. Privacy policies are technically complex and often inaccessible in any meaningful sense, even for educated users. Certain concepts lack direct equivalents in many languages and do not relate to everyday experience in ways that make risks tangible. These include the very idea of “data”, let alone “data retention”, “secondary use”, and “cross-platform sharing”. This lack of accessibility further discourages individuals already under emergency pressures from asking questions. Since immediate survival takes priority, the most pressing risk is often the prospect of losing access to assistance by refusing to share personal data.
“The result is what might be described as ‘transactional consent’: data for assistance, an exchange governed less by understanding than by necessity.”
The result is what might be described as “transactional consent”: data for assistance, an exchange governed less by understanding than by necessity. In humanitarian settings, that dynamic is compounded by acute power imbalances, limited non-digital alternatives, and the visible authority of institutional actors. Organisations sometimes attempt to make participation more clearly optional through explicit opt-out statements or simplified consent forms, but the depth of understanding achieved remains uncertain.
AI further narrows the space for informed consent. It is not necessarily visible even to organisations, let alone the users themselves, when companies mine their voice data or written queries to train AI models. And communities who cannot meaningfully interrogate a data collection form are considerably less equipped to question an automated decision-making system. There is already documented evidence of humanitarian agencies processing programme-generated data through AI models without the explicit consent of data subjects. In some cases, this was before agency-wide AI governance policies were established.[3]Sarah Spencer, “Humanitarian AId? Considerations for the Future of AI-use in Humanitarian Action”, Reliefweb, Elrha, 17 January 2024, … Continue reading Transactional consent does not provide a basis for challenging an eligibility status decided by algorithm.
Those who suffer digital harms have little recourse
Whether or not those sharing their personal information understand the risks, they have real-world consequences – often for individuals who are already vulnerable.
Sometimes digital visibility carries direct physical risk. In many contexts male ownership of devices means that using apps or messaging services can carry risk for women and girls. This is the case when messages requesting assistance from a gender-based violence support service risk being visible to an abusive partner. Being identified also carries the risk of surveillance and targeting, especially for those living alongside warring parties, forcibly displaced or with precarious legal status. This risk was painfully illustrated when Rohingya refugees’ biometric data, collected to register them for assistance, was then shared with the Myanmar authorities they had fled from.[4]Human Rights Watch, UN Shared Rohingya Data Without Informed Consent, 15 June 2021, https://www.hrw.org/news/2021/06/15/un-shared-rohingya-data-without-informed-consent
In other contexts, digital registration can have critical consequences for future support. We were told of a case in Ukraine where individuals who unsuccessfully applied for international assistance were automatically deemed ineligible for national support. No appeal was possible and they were unable to get their details removed from the international database.
At the organisational level, institutional incentives further undermine accountability. In competitive funding environments, NGOs may fear reputational damage or donor backlash if digital failures are openly acknowledged. Formal complaint mechanisms are also not designed for digital accountability. Hotlines, SMS lines and complaint boxes are widely in place but mainly designed for operational concerns like delays in distribution. Mechanisms specifically addressing digital concerns, data governance disputes or rights relating to collected data are rare.
The accountability implications of more complex and dispersed data systems are also considerable. Where neither communities nor the organisations working with them fully understand how digital systems function, who has responsibility for any harm those systems cause? In many cases, it is not an NGO but a commercial technology company that has most control over how data is processed and governed. Its operations are likely to be global, its presence in crisis settings limited, and its direct accountability to affected communities lacking.
Existing frameworks do not offer accountability for complex uses of technology
Existing feedback and complaint mechanisms were designed for direct relationships between organisations and communities. They are ill adapted to layered systems where the organisation implementing a programme may have no control over the platform processing its data. Data passes through systems that no single actor controls. Even when organisations want to respond to a community member’s request to correct or delete their data, they often lack the technical authority to do so. Local data collectors do not have the credentials to alter backend databases[5]A backend database is where an application or website stores its data permanently, on the “server” side – that is, the part that is invisible to the user [editor’s note].. Data shared with partner agencies or commercial platforms may be beyond the implementing organisation’s reach entirely.
This is the accountability environment into which AI tools are being introduced, for needs assessment, targeting, feedback analysis and natural language processing. AI systems introduce additional barriers to accountability: algorithmic logic that is harder to explain than a data collection form, training data that may embed historical biases, and outputs that are presented as objective while reflecting the values and assumptions built into their design.[6]Giulio Coppi, Rebeca Moreno Jimenez and Sofia Kyriazi, “Explicability of humanitarian AI: a matter of principles”, Journal of International Humanitarian Action, vol. 6, no. 1, 2021.
Accountability in the age of AI needs community participation more than ever; it cannot be reduced to a privacy policy or delivered through a hotline. It is built through sustained relationships between communities, the intermediaries they trust, and the organisations that serve them. This starts with community members’ meaningful participation in decisions about the design and use of technology and with ensuring they are equipped for it.
An accountability resource for communities
With all this in mind, we have developed a resource that we hope can help communities independently question the AI, and other technological solutions, that they are presented with. “How to hold organizations accountable for their use of digital technology: a guide for crisis-affected communities” is an early design, with scope for further participatory development.[7]In open access here: https://clearglobal.org/resources/guide-how-to-hold-organizations-accountable-for-their-use-of-digital-technology But it aims to address the key findings from this study, consolidated through co-creation workshops with practitioners, in practical ways:
“We have developed a resource that we hope can help communities independently question the AI.”
- crisis-affected communities need to know what they are entitled to expect when organisations use digital technology, including their right to be consulted, to receive clear information and to raise concerns. The guide sets out the principles and standards organisations must comply with and explains what human-centred design should look like in practice, so that communities can use this knowledge to hold organisations to account;
- people need practical support for action more than technical explanations of data protection. The guide provides examples of how risks arise in practice, a list of red flags to help communities identify unsafe uses of technology, ready-to-use questions to ask organisations before sharing personal data, and guidance on raising concerns and making complaints safely;
- communities have a right to know not just what organisations do with their data, but what those organisations can and cannot control. Using plain language, the guide equips community members with the questions to ask about data flows, third-party platforms and data-sharing arrangements. It also makes clear that communities have a right to this information even where the honest answer is that the organisation’s control is limited;
- collective action is often safer and more effective than individual complaints, especially when challenging structural features of digital systems. The guide encourages communities to prepare and report problems together, and explains how doing so can increase pressure on organisations, protect individuals and build a stronger basis for escalating complaints to higher authorities when direct reporting fails;
- the guide reflects communities’ consistent demand for access to a real person, particularly in urgent or sensitive situations. It identifies the absence of a human alternative as a red flag in the use of digital technology, highlights that chatbots and automated services cannot apply human judgement and often make mistakes, and recommends that communities ask organisations to provide a human backup for any automated information service. It uses plain language, visual formats, and scenario-based content to reach communities with low literacy or limited connectivity, and it explicitly identifies inaccessibility – whether by language, format, or digital skill requirement – as a red flag in humanitarian use of technology.
It is right that efforts are under way to develop sector-wide AI governance standards.[8]CDAC Network, Alan Turing Institute, Humanitarian Advisory, SAFE AI: Standards and Assurance Framework for Ethical AI, accessed 1 June 2026, https://www.cdacnetwork.org/safe-ai; Wilton Park/UK FCDO, … Continue reading But standards at the organisational or sector level, while important, address only part of the problem. The communities whose data feeds these systems, and whose access to assistance may be shaped by their outputs, need direct accountability. They need to be able to understand, question and exercise meaningful influence over the technologies deployed in their name. “How to hold organizations accountable for their use of digital technology: a guide for crisis-affected communities” is a first step towards that.
Picture credit : ICRC
