Everyone struggles with determining whether photos you see online are “real” or whether they might be synthetic media. If you’re feeling overwhelmed by that task, you’re not alone. Even the New York Times struggles to assess the veracity of photos in social media. In this post we’ll take a look at how professional journalists make decisions about what’s real and what’s not. We’ ll also take a look at what you’re up against.
The Facts
On 3 January 2026, the United States executed military strikes against several targets in Venezuela, and President Nicolás Maduro, and his wife, Cilia Flores. President Trump announced this via social media and followed up with the now-famous picture of Maduro blindfolded on board a warship and handcuffed in sweatpants.
Simultaneously dozens of other photos flooded social media supposedly showing Maduro in US custody.
The Challenge: Assess the Veracity of Photos in Social Media
Even for THE news organization in the United States, the task of deciding whether to reprint what Trump had posted as a fact was far from simple. Assessing the veracity of related photos on social media was even more difficult. They had such a hard time, in fact, that they later published a whole article about how they made the decision.
The Steps
President Trump – much like most other people, more than occasionally posts AI-generated content. This was most famously the case with his “Top Gun style” video showing him dumping feces on protestors. So the reporters knew that they didn’t know if it was real. That’s the first step.
The second step was using technical tools to assess how likely the photo was to be a fake. In this case, they used an AI detection tool called “AI or Not“. These kinds of tools scan an image and try to find patterns or watermarks that might have been left behind by an AI image generator. This approach, however, is not reliable and certainly not reliable on its own. When they compared the probability scores for the image to other images of Maduro from social media, they found that the results were largely inconclusive.
The third step is thinking critically. When the editors compared the various photos on social media to the post in question, they found that some were clearly fakes. There were telltale signs of AI in a number of the photos that went viral alongside Trump’s post (which turned out to be legitimate). In one for instance Maduro sat in an Air Force plane with too many windows. Others, however, did not have giveaways visible to the naked eye. Even with technical analysis, it turned out to be very difficult to establish assess the veracity of photos in social media.
The NYT Solution for Assessing the Veracity of Photos
In the end the journalists decided to publish a cropped version of the post on The Times’s home page and a full post on the inside of the newspaper knowing that the front page of the print version would be “sticky” if they got it wrong. They made a judgement call and hedged their bets by calling it what it was – a communication from the president.
Lessons for Everyday Mortals
- Know what you don’t know: Remember any image might be fake.
- Think probabilistically: It’s not a matter of fake or not but what’s your percentage bet based on evidence.
- Gather evidence: Use AI detection tools like AI or Not or SynthID.
- Look for telltale giveaways: Examples are humans with three hands, dogs with five legs, windows or text where it would never be in real life.
- Know the stakes: Remember what’s riding on this decision – if it’s your life savings your confidence level should be near 100% whereas if the stakes are low maybe you can afford a lower confidence level on your bet.
- Make a judgement call: If you need to DO something about the image – only do so after the first five steps. If you don’t need to DO anything – remember that you still don’t know for sure.
If in Doubt: vali.now
Your best defense is healthy skepticism. If something seems just a little off or a bit too good to be true – it probably is. Your best option is to forward the details to help@vali.now. Our cybersecurity professionals have been recognizing and fighting off such attacks for decades. Your first case (up to one hour of research) on our side is free, with affordable rates after that. You have nothing to lose, and you might prevent catastrophic losses by reaching out.