AI-powered disinformation spreads online amid Los Angeles protests

France 24

20 hours ago

Since June 6, Los Angeles has been rocked by protests and clashes with law enforcement, after more than 40 migrants were arrested downtown by the Immigration and Customs Enforcement (ICE).
The events have escalated into a confrontation between Donald Trump and the California leaders after the US President deployed National Guard troops against protestors on June 7, overriding the objections of California Governor Gavin Newsom.
In this context, several internet users have circulated misleading or erroneous content using artificial intelligence tools.
AI-generated selfie video of National Guard soldier
On June 9, an alleged National Guard soldier, who identified himself as Bob, posted a selfie video on TikTok that garnered more than 770,000 views.
The 'soldier' is seen smiling and offering a 'behind-the-scenes' look at troops preparing for deployment to 'gas' protesters.
The second scene of the video features 'Bob' during the clashes, decrying the throwing of 'oil-filled balloons' at law enforcement.
It turns out that the video is fake. There are several indications that it was generated using artificial intelligence.
For instance, in the video's first scene, one of the badges that should display the soldier's rank shows nonsensical characters. AI still struggles with creating coherent text.
Another inconsistency is a strangely placed traffic light on the left side of the image, which is partially obscuring another traffic light.
We found another anomaly in the video's second scene. A police car displays an incorrect acronym, LAPC. The Los Angeles Police Department's actual acronym is LAPD.
The TikTok account that posted the video, 'maybenotquitereal', is a parody. Its bio states that the user produces "satirical" content.
However, this disclaimer isn't visible in the video featuring the supposed National Guard soldier.
As seen in the comments, many users believe the video is genuine. Some urge "Bob" to "stay safe" and thank him for 'protecting the country', while another pro-protester online user criticises him for "hurting the citizens" he was supposed to "protect".
Did California's Governor share fake National Guard photos? An AI fact-check error
On June 9, California Governor Gavin Newsom posted photos on social media showing National Guard soldiers seemingly asleep on the ground.
"You sent your troops here without fuel, food, water or a place to sleep. Here they are — being forced to sleep on the floor, piled on top of one another. If anyone is treating our troops disrespectfully, it is you, Donald Trump,' Newsom wrote.
However, a user quickly challenged Newsom's claim in the comments, asking, "Why are you posting fake news?"
The user backed up this assertion with a screenshot of an image verification performed by ChatGPT. According to the AI, the photograph was taken on August 19, 2021, at Hamid Karzai Airport in Kabul, Afghanistan, during the fall of the Afghan capital to the Taliban.
The image was reportedly part of a series titled "Afghanistan Evacuation [Image 1 of 3]," published on Dvids, a US military image bank.
As explained by BBC Verify journalist Shayan Sardarizadeh, ChatGPT's analysis is inaccurate. A reverse image search (here's how to do one) traces the origin of the images to an exclusive publication by a local media, the San Francisco Chronicle, on June 9, 2025. No instances of these photos appear on Google before this date.
Furthermore, our team conducted a search on the Dvids website, where ChatGPT claimed the photos come from, using the AI's suggested title, "Afghanistan Evacuation [Image 1 of 3]." The search yields photographs unrelated to those published by Gavin Newsom.
The reliability of image origin verification by chatbots like ChatGPT and Grok varies significantly. Search results from these AI can change depending on when the query is made, as demonstrated by a recent X thread where several users asked Elon Musk's Grok AI to identify the source of the photos.
In an analysis initiated on June 9 at 22:55, Grok determined that photographs shared by Gavin Newsom were taken in Kabul on August 19, 2021, just like ChatGPT. In this analysis, the chatbot even deemed the use of these photos in 2025 to criticise the troop situation in Los Angeles as "misleading".
However, the same analysis performed at midnight on June 10 yielded a different outcome. While Grok still suggested the photo was taken in Afghanistan in 2021, it no longer entirely ruled out the possibility that it could date from 2025.
Finally, an analysis run on June 10 at 5:23 provided the correct result, attributing the photograph's origin to the San Francisco Chronicle.
A search conducted by our team in English on ChatGPT on June 10 at 17:46 also yielded an accurate result, crediting the San Francisco Chronicle as the source of the photographs.