Soon, Everything Will Be a Deepfake. That’s Not the Real Problem.
Updated: Oct 8
Detecting that something was manipulated may not tell you very much on its own. The question we think will matter more is: Why was it manipulated? and to exemplify this point we will share a recent example from Brazil’s general election.
Is manipulated media necessarily a sign of malicious intent?
We’ve spent the last few years asking whether an image or video is real.
It made sense. Deepfakes were relatively unusual, and detecting manipulation was the hard part. If you could identify that pixels had been altered, a face had been swapped, or a video had been generated, you had already learned something important.
But that’s changing quickly.
We’re heading toward an internet where manipulated and AI-generated content is normal. Brands are already using generated images in campaigns. People are creating AI videos for entertainment, satire, memes, political content, and pretty much everything else.
Soon, detecting that something was manipulated may not tell you very much on its own.
The question we think will matter more is: Why was it manipulated?
There’s a big difference between an AI-generated image made as a joke and a fabricated video of your CEO announcing something that never happened. Technically, both may be synthetic. From the perspective of the organization being targeted, they’re obviously not the same problem.|
That’s the thinking behind Malicious Intent Probability in Brinker.
Why understanding intent reveals more about manipulated media than analyzing pixels.
A recent example from Brazil’s general election detected by Brinker shows why this matters.
A video began circulating claiming that Lula threw the Brazilian flag on the ground after voting.
There was just one problem: the video itself wasn’t manipulated. The context was.
The footage actually shows Lula handing the flag to another person, without it touching the ground. But presented with a different caption and narrative, the same footage told a very different story.
The claim began circulating on Sunday, October 4, with posts appearing across X, TikTok, Instagram, Facebook, and Threads and gaining thousands of views.
A court ordered the removal of posts containing the false claim on October 5. But by then, the narrative was already spreading.|
This is an important distinction. Content doesn’t have to be AI-generated, photoshopped, or technically altered to be misleading or harmful.
And as the volume of synthetic content grows, that context will become even more important.
What questions should we be asking when looking at deepfakes?
Deepfake detection needs to become a prioritization problem.
Which content is harmless? Which deserves a second look? Which is connected to a narrative that is already spreading? And which looks like someone is deliberately trying to damage your reputation, manipulate your community, or influence how people perceive your organization?
According to a Europol report, by the end of 2026, 90% of online content will be AI-generated. Investigating every manipulated image or video with the same level of urgency simply won’t scale.
Deepfake detection tells you that something may have been manipulated.
Malicious Intent Probability helps you understand whether that manipulation actually matters.
