An AI-generated image depicting an Immigration and Customs Enforcement raid at Alta Bates Summit Medical Center in Berkeley, Calif. | Source: Courtesy
On Saturday, Juan received a photograph from his girlfriend showing Immigration and Customs Enforcement agents arresting two people outside a Chipotle near San Francisco’s AMC Metreon theater. His girlfriend got it on WhatsApp from a friend who worked nearby.
“I was like, ‘Oh, God,’” said Juan, a restaurant manager and undocumented immigrant from Guanajuato, Mexico, who asked that his last name be withheld for his safety. Juan was driving from San Francisco to work in Oakland when the message came in.Juan quickly warned his colleagues, who passed the alarming photograph along to others.
“I thought, ‘Wow, it looks real,’” Juan said. “I believed it.”
Juan and his coworkers began making plans for what to do if immigration agents arrived at the workplace.
Hours later, Juan saw a post (opens in new tab) from San Francisco’s Rapid Response Network, which runs a hotline and verifies ICE sightings, saying the image had been generated by AI. Juan immediately contacted his girlfriend, urging her to warn the friend who had originally shared it.
An AI-generated image falsely depicting ICE agents by a Chipotle was submitted to ICEout.org moderators, who declined to publish it. | Source: Courtesy ICEout.org
“This is screwing everyone over,” he told her. “We had a deal: If you didn’t see it, don’t share it.”
Increasingly realistic AI-generated images that appear to depict ICE arrests are circulating through tight-knit immigrant communities’ group chats and social media feeds, spreading fear while watchdog groups scramble to verify reports and debunk misinformation. The trend is affecting many regions with large immigrant populations, including the Bay Area.
“These images circulate so quickly,” said Angel Ibarra Nuñez, co-director of the Alameda County Immigration Legal Education Partnership. “They circulate a lot faster than the good news that they are false.”
Tracking down the images is like “some small investigation,” said Celeste, a founder of ICEout.org (opens in new tab), a website that crowdsources reports of ICE activity. Celeste, who asked for her last name to be omitted for fear of political harassment, is the lead moderator for the site’s Bay Area section, deciding whether reports of sightings will be made public.
ICEout received nearly 20 reports of the Chipotle photo Saturday morning, Celeste said.
ICEOut rejected nearly 20 submissions of the same AI-generated image, which sparked fear among immigrant communities Saturday. | Source: Courtesy ICEout.org
She immediately knew it was AI-generated. “The first giveaway is that ICE rarely wears these vests that just say ‘ICE,’ and typically they use yellow letters,” she explained in an Instagram video (opens in new tab) posted Wednesday. “And they certainly don’t use vehicles that say the full ‘U.S. Immigration and Customs Enforcement.’”
Immigrant advocates like Celeste don’t know who is creating the deceptive pictures, which are easy to synthesize. “I do think that the originator of these images is somebody that is trying to cause panic,” she said. “And is doing it out of malicious intent.”
While such images have circulated since last year, Celeste says they have become more prevalent amid reports of “Operation Safe California,” a 45-day immigration enforcement operation targeting sanctuary cities ahead of the midterm elections. On Oct. 7, the Washington Examine (opens in new tab)r reported that ICE agents had been instructed to arrest undocumented immigrants whether or not they have criminal records.
“We are essentially responding to two things at once: actual immigration enforcement activity and false reports that can look extremely convincing,” Ibarra Nuñez said.
Another counterfeit photo popped up Sunday. The Santa Clara County Rapid Response Network debunked (opens in new tab) the AI-generated image, which appeared to show ICE agents outside Pacific Catch, a restaurant in Mountain View.
Ibarra Nuñez said the rapid response network received calls in September about a photograph that appeared to show dozens of ICE officers outside Berkeley’s Alta Bates hospital.
The proliferation of AI-generated images deters immigrants from going to work, school, medical appointments, and other important activities, according to Ibarra Nuñez.
“We had community members essentially calling in to ask us whether they should go to their appointment,” he said. “They were calling ahead because they were afraid.”
Santa Clara’s rapid response network debunked a blurry AI-generated image showing ICE agents stationed in front of a Mountain View restaurant. | Source: Courtesy Santa Clara County Rapid Response Network
While there are AI-verification tools, Ibarra Nuñez said they are unreliable. “It’s essentially using AI to identify AI,” he said.
ICE and the Department of Homeland Security typically don’t comment on their operations. Instead, rapid response networks “play it safe” by dispatching volunteers to look for witnesses who can corroborate or rebut the reported sightings.
“It’s another layer of work, because we have to distinguish credible firsthand reports from the images or the claims that we are seeing be generated or manipulated,” Ibarra Nuñez said. “It consumes verification capacity and creates confusion.”
The manipulated images abound on private Facebook and WhatsApp groups and pop up on crowd-sourced sites like ICEOut, where anyone can submit reports, Ibarra Nuñez said. Occasionally, fake photos get past the gatekeepers, Celeste acknowledged.
The moderators on ICEOut “are human and can miss things,” Celeste said. “When an AI-generated image gets through, we own that. We take it down, look at what went wrong, and train our moderators so it’s less likely to happen again.”
Photos showing outdated ICE activity, those that give incorrect locations, and reports that misidentify other law enforcement officers as immigration agents are another challenge.
“The rapid response networks have gotten very good at confirming and verifying when there is ICE activity or when there are rumors,” said Alex Mensing, communications director of the California Collaborative for Immigrant Justice. He urged people to check with their local rapid response network before sharing unverified information.
Adopting the role of public fact-checker was a jump for the county response networks, which, before President Donald Trump’s second term, worked primarily to connect arrested immigrants with legal assistance.
With fake pics flooding social media, some response networks publish their findings on Facebook and Instagram as a counterweight to the rumors.
Approximately 20% of the sightings reported to Ibarra Nuñez’s group are verified as ICE encounters.
“A lot of the posts that we share are the false ‘No, it was not ICE’ situations,” he said. “Similarly, when it is ICE, we also want folks to be aware that something happened in their communities.”
But by the time a rapid response network publishes a verified report, hours may have passed, and immigrants may have already made decisions based on AI-generated information.
“The harm, the fear, is likely to be created before the proof comes out that it was a fake image,” said Brian Hofer, a Bay Area privacy advocate. “So the damage is already going to be done.”
A potential antidote to online misinformation is decidedly low-tech and hyperlocal, Mensing said.
“Community organizing is the answer to fear,” Mensing said. “When people are able to know each other, be in communication with each other, and trust each other, that’s when people are best able to find out what’s really going on and stay safe.”
For Juan, Saturday’s experience was a reminder of how much influence he has within his community.
“If I post something, people are going to believe me,” he said. “By the time we find out it wasn’t true, 20 or 30 families have already gone into hiding. They’re frightened, they’ve closed their businesses, or they’ve said, ‘I don’t want to work right now.’
“It was the most terrifying hour of my day.”