Lies spread fast, but AI-generated lies packaged with deepfake video and hyper-targeted algorithms move at an entirely different speed. In humanitarian emergencies, that speed kills. We aren't talking about abstract political debates or internet trolls messing around in online forums. We are talking about deepfakes of aid workers used to incite riots, and AI-spun rumors that get refugees killed or pushed right back into the gunfire they just escaped.
The United Nations High Commissioner for Refugees (UNHCR) recently spotlighted this crisis at the AI for Good Global Summit in Geneva. They made it clear that today's global displacement crises are fundamentally linked to severe information crises. When bad actors weaponize generative technology, the targets aren't just systems. They are real people who have lost everything.
People searching for the intersection of technology and human displacement usually want to know if AI is genuinely making the world more dangerous for vulnerable groups. The answer is a definitive yes. While tech companies brag about synthetic media and automated content generation, humanitarians on the ground are scrambling to deal with the fallout of localized, weaponized disinformation.
Deepfakes and Human Targets: The New Danger Zone
Gisella Lomax, the UNHCR senior advisor on information integrity, pointed out a horrifying trend. The swirl of coordinated online lies has driven a sharp increase in false, hostile narratives specifically targeting humanitarian operations and field staff.
Imagine an automated account generating hundreds of high-quality, translated articles claiming an aid organization is smuggling weapons instead of delivering clean water. Within hours, local groups read these fake stories, look at a realistic but entirely fabricated deepfake video of an aid worker confirming the lie, and march on a refugee camp with weapons. This isn't a hypothetical movie plot. It's happening in active conflict zones where information risks are incredibly sharp and immediately translate into physical violence.
When automated tools distort information, the damage echoes across every part of a displaced person's life. It isn't just about avoiding physical attacks. Synthetic hate campaigns erode social cohesion in host communities. They make it harder for refugees to get jobs, secure housing, or enroll their kids in schools. Online rumors, false accusations, and dehumanizing speech act as direct triggers for local protests. In the absolute worst scenarios, these coordinated digital campaigns cause physical killings and force families to flee for a second or third time.
The Numbers Behind the Displacement Crisis
To truly understand the scale of what's at stake, you have to look at the sheer volume of human beings currently caught in these digital crosshairs. The raw numbers are staggering.
- Total displaced people: 117.8 million individuals globally were forcibly displaced by the end of 2025.
- Internally displaced: 68.7 million people forced to flee their homes but remaining within their own country's borders.
- Refugees under UN mandate: 35.6 million people who have crossed international borders to escape violence or persecution.
The concentration of this crisis makes it incredibly easy for malicious actors to target specific populations. Roughly two-thirds of all global refugees come from just five countries: Venezuela, Ukraine, Syria, Afghanistan, and Sudan.
Because the originating points are so concentrated, bad actors can build highly specific, culturally tuned AI models to generate targeted propaganda. They don't need a massive team of writers anymore. One person with a cheap, off-the-shelf generative model can flood the digital ecosystems of these specific communities with terrifying efficiency.
How Bad Actors and Smugglers Weaponize the Tech
We often view AI safety through the lens of Western elections or corporate deepfakes. That's a massive oversight. Human traffickers and smugglers are some of the most sophisticated adopters of new tech platforms.
Traffickers use generative tools to clone legitimate humanitarian websites and build highly convincing social media profiles. They use automated translation models to write perfect, persuasive messages in local dialects, offering desperate families guaranteed jobs, safe passage across borders, or fake visas.
Once a family buys into the lie, they end up trapped. They face extortion, forced labor, or physical violence. The platforms hosting these ads often miss them entirely because their automated content moderation systems are built to catch English-language scams, completely ignoring rare languages or regional dialects used in East Africa or parts of Asia.
The digital platforms hosting this content are failing to keep up. When an algorithm rewards engagement above accuracy, a sensationalized, hateful post about refugees will always outperform a dry, verified fact-sheet from an aid agency. The systemic incentives of modern social media companies directly subsidize the spread of life-threatening information.
The Flip Side: Using AI to Actually Help
It's easy to look at these disasters and want to ban the technology entirely. But that's impossible, and frankly, it would do a disservice to the tools that actually work. When deployed properly, automated systems can save lives in humanitarian crises.
For instance, machine learning tools can parse satellite imagery and local emergency radio traffic to trace missing persons significantly faster than human teams. During major displacements, tracking down separated family members is an agonizingly slow process. Predictive algorithms can match scattered data points to reconnect parents with their kids in a fraction of the time.
Data analysis tools can also predict mass movements of people fleeing sudden violence, allowing aid agencies to position food, clean water, and medical supplies at border crossings before the crowds arrive. The technology itself isn't the enemy. The problem is a massive lack of accountability, investment, and localized protection.
What Tech Platforms Must Do Next
We can't rely on generic corporate statements or performative ethics boards anymore. If technology companies want to prove their tools are built for good, they need to take immediate, systemic action to protect vulnerable populations.
First, tech giants and generative developers must partner directly with humanitarian organizations. This means investing real capital to co-develop early-warning systems that spot coordinated hate campaigns before they trigger real-world massacres.
Second, content moderation tools have to be dramatically overhauled to work effectively in humanitarian contexts and less-common languages. It's unacceptable for a multi-billion-dollar tech firm to have flawless automated moderation for English text while leaving local languages in conflict zones completely unmonitored.
Finally, digital platforms must change their algorithmic incentives. Posts flagging humanitarian emergencies or providing verified safety information should be prioritized, while unverified accounts churning out high-volume, racially charged synthetic media should face instant demonetization and account suspension.
Your Practical Action Plan
If you want to stop the weaponization of information in your own circles and support digital safety for displaced people, stop overthinking and start doing the basic work.
- Starve the algorithm: Never share, quote-tweet, or comment on an inflammatory post about a vulnerable group, even if you are doing it to debunk it. Engagement is engagement. Report it directly to the platform and move on.
- Verify before you fund: Before donating to an emergency relief cause based on a viral video or an emotional social media story, check the organization's official website independently. Make sure the media wasn't generated to siphon funds into scam operations.
- Support localized digital rights: Donate to or amplify organizations like the Information Integrity Lab or local digital rights groups in conflict areas. They do the heavy lifting of translating, tracking, and fighting localized digital hate when big tech fails to look.
