The infrastructure of trust: AI, information integrity and accountability in crisis
In a crisis, information is not background noise. It shapes whether people move towards safety or danger, trust a vaccine advisory or dismiss it, register for aid or stay out of the system. Information is what allows people to ask questions, make choices, challenge decisions and exercise agency over their own lives.
This has always been at the heart of CDAC Network’s work. But the environment in which this work happens has changed.
People affected by crisis no longer get information only from official announcements, community meetings, radio broadcasts or humanitarian feedback channels. They also navigate social media platforms, messaging apps, search engines, influencers, automated translation, synthetic content and, increasingly, AI-generated answers. Aid organisations themselves use digital tools to analyse feedback, monitor rumours, translate messages, detect harmful narratives and make decisions.
CDAC’s three workstreams – Community Engagement and Accountability, Information Integrity, and AI – are not three separate agendas. They are three aspects of the same underlying question that has always driven CDAC’s work: who has power over information, decisions and technology in crises?
Information integrity is part of accountability
Information integrity is often discussed as a problem of false or misleading content. That framing is too narrow, especially in humanitarian contexts.
Crises are environments of uncertainty. Rumours may circulate because people are afraid, because official information is absent, or because communities are trying to make sense of danger with the information available to them. At the same time, armed actors, political groups, scammers and other adversarial actors deliberately manipulate the information space.
In this environment, harmful information is not only a communications problem. It can affect whether people access services, whether staff and local responders are safe and whether people can make informed decisions about their own protection and survival.
Accountability depends on information integrity: an environment in which people can ask questions, compare claims, receive timely answers and challenge what they are told. AI is now reshaping that environment in ways that threaten both.
AI shapes the humanitarian information environment
The public debate about AI and information integrity has focused on synthetic content: fake images, impersonated voices, automated propaganda and misleading text. Those risks are real, but they are only part of the story.
More fundamental is the embedding of AI in the systems people use to search, translate, summarise, moderate, verify and act on information. AI is not only changing what information is produced: it is changing how people come to know what they know.
Consider the shift from search to answer, enabled by generative AI. Traditional search engines present users with a set of sources to compare, whereas AI-powered search and chatbot interfaces offer a single, synthesised response.
This matters acutely in crisis contexts, where information is fragmented, contested, time-sensitive and unevenly accessible, and where questions rarely have one stable answer. A road may be safe in the morning and unsafe by afternoon. A phrase that looks harmless in translation may have political, ethnic or security implications locally.
In these settings, the danger is not only that AI may produce false information. It is that confident answers produced by an AI system make contested, uncertain or evolving information appear settled.
Trust cannot be automated
AI can support humanitarian action, but it cannot automate trust. Trust is built socially through relationships, language, presence, transparency and repeated evidence that information providers understand the realities people face. It is earned when communities can ask questions, challenge information, see how decisions are made and observe whether their feedback changes anything.
An AI model can summarise a rumour, but it cannot understand why that rumour matters unless the system is grounded in local knowledge. A dashboard can show a spike in harmful narratives, but it alone cannot determine whether those narratives reflect fear, manipulation, sarcasm or lived experience.
Community journalists, local responders, civil society groups, translators, diaspora networks and trusted intermediaries understand the signals that automated systems miss: slang, coded language, humour, stigma and the social significance of who is speaking. AI systems deployed without this contextual knowledge may misread the environments they are meant to help navigate, presenting a cross-cutting risk to protection and accountability.
Participation must shape the system
Too often, participation in technology projects is treated as consultation: communities are asked for feedback late in the process, after key decisions have been made.
For AI systems in crisis contexts, participation needs to be understood as governance intelligence. Crisis-affected communities, local responders, journalists and civil society can identify risks that system-level assessments may not catch: mistranslation, exclusion errors, harmful assumptions, unintended consequences for access to assistance, or ways a tool may be interpreted in a low-trust environment.
As established in CDAC’s briefing paper on the governance gap in humanitarian AI, participation only functions as governance if it can influence decisions: whether a system is selected, adjusted, limited or withdrawn. Anything less is ‘participation washing’.
The right to know when AI has been involved in communications, analysis or decisions that affect people’s lives should be a minimum standard, as set out in the SAFE AI Framework. People should know when they are interacting with a chatbot rather than a human, how to correct information, challenge harm or refuse a system without losing access to assistance. Without this, AI risks creating new forms of distance between people and the institutions meant to serve them.
A shared agenda
This is why CDAC’s three workstreams form a shared agenda:
Community Engagement and Accountability is the foundation: people affected by crisis must have the information, channels and influence to shape the decisions that affect their lives.
Information Integrity is the condition for the foundation to hold – people cannot participate meaningly if the information environment is polluted, manipulated or fragmented.
AI is now part of the infrastructure underlying both: the systems through which information is produced, filtered, translated and trusted.
The question is not whether AI will be part of humanitarian information ecosystems – it already is. Crisis-affected people encounter AI through search engines, messaging platforms, translation tools, chatbots, social media feeds, automated moderation systems and synthetic content online. Humanitarian organisations, as well as using AI tools themselves, operate in environments increasingly shaped by them.
The real question is whether these systems weaken people’s ability to understand their situation and make critical decisions – or strengthen it. In crises, people do not only need answers. They need to know where those answers come from, what remains uncertain, whose knowledge counts, and how to challenge information that affects their lives.
That is the standard humanitarian information systems should be held to, whether they are analogue, digital or AI-enabled.
For humanitarian organisations, donors, technology companies and regulators, this means AI governance is an information integrity issue; information integrity is an accountability issue; and community engagement and accountability must be central to responsible AI.