Hostile actors pose risks by exploiting AI technologies in humanitarian contexts, with negative impacts on the information space including cyberattacks, disinformation, and fraud as well as growing surveillance. Participants reflected on the increasing risks to humanitarians, the eroding of trust between humanitarian actors and affected populations as well as steps to mitigate potential harms.
AI opens easy opportunities for disinformation, criminal activity, and the sowing of social discord. It is now possible to deploy deeply sophisticated scams at scale; for example, LLMs can write 20 persuasive tweets in five minutes and fake content is easy to create.
The information environment is polluted, overloaded, and manipulated, particularly in conflict and crises. Trends are emerging of weaponizing civilians to harm others in the real world. Both state and non-state actors can use AI to push conspiracy theories and misinformation.
Trust and safety teams are being laid off on big platforms, content moderation has worsened, and underrepresented languages are not moderated at all.
One consultation in Kenya revealed that people were very concerned about audio scams during crises and had knowledge about the risks. How is the humanitarian sector supporting communities who are dealing with misinformation and scams? An analytical framework is needed that thinks about these risks from the perspective of the community. How does the sector seek useful information and intelligence about how to do this effectively from AI and tech experts? It is necessary to reframe the problem. Disinformation is organised crime and criminals are making inroads into places that the humanitarian sector cares about deeply. Platforms and frameworks alone will not solve this problem.
“If a hostile actor’s goal is to undermine trust in a humanitarian organisation, it could change lots of things – this is not science fiction.”