A picture of black smoke billowing close to the Pentagon in Washington first surfaced on Twitter on May 22, 2023. The picture appeared authentic. The image featured the visual characteristics of a phone camera rather than a staged production, with a slightly blurred background, constant lighting, and convincing geometry. Pentagon explosions are rare. Markets fluctuated for a few minutes until journalists and fact-checkers linked it to an AI picture generator and the Pentagon verified nothing had happened. The S&P 500 saw a slight decline. Financial markets were affected by a fake image that circulated at scale for around fifteen minutes.
There was no public discussion regarding the incident’s ramifications. After being noted and filed as a cautionary data point, the news cycle resumed. In reality, it demonstrated the verification bottleneck, which is more difficult to understand. Those who needed to know quickly discovered that the photograph was a phony. However, there was enough time between the posting of an AI-generated image and the people who knew it was fraudulent getting to everyone who had seen it to have serious repercussions. Since then, the story of synthetic media has been about that divide, and it has been growing rather than shrinking.
The technology used to create this information has advanced incredibly quickly. In 2019, odd blinking, mismatched skin tones, ears that didn’t quite resolve correctly, and faint flashing at hair boundaries made it easy for skilled eyes to spot deepfakes. In 2026, those tells are essentially nonexistent at the level of quality produced by tools of the current generation. Previously requiring costly hardware and hours of processing, facial synthesis technologies now operate in real time on consumer-grade devices. Systems for voice cloning require thirty seconds of original audio. From text descriptions alone, video generation algorithms are able to create convincing short clips. The difficulty of producing synthetic media that most people cannot tell apart from real film has all but vanished.
In a 2019 Harvard Law Review article, legal experts Robert Chesney and Danielle Citron described the “liar’s dividend” as a phenomena that has progressed from theoretical worry to verified reality. They contended that the availability of reliable deepfake technology would enable people to discount genuine evidence as fake, neutralizing actual recordings of actual wrongdoing by merely asserting that they were AI-generated. This has taken place. The deepfake defense has been used in court cases, political scandals, and in reaction to real audio recordings that upset influential people. Sometimes it succeeds because it creates enough doubt to complicate proceedings or change public opinion before the subject is decided, rather than because jurors or judges are persuaded by it. The prospect of fakery can now be used to defend against real evidence.
This is being felt by the journalism sector in both overt and covert ways. The obvious aspect is the rise in verification overhead; AP, Reuters, and the BBC have all noted significant increases in the amount of time needed to confirm images and videos prior to publication, merely because it is no longer safe to maintain the baseline assumption that a plausible-looking image is probably real. It may seem insignificant, but at newsroom scale with tight budgets, an estimated 15–20 percent more time required on each image verification is a considerable drag on both cost and speed. The less obvious aspect is the decline in public trust in media outlets that misrepresent synthetic media. Any media outlet that publishes an AI-generated image as authentic suffers long-lasting reputational harm, which makes newsrooms more cautious, slower, and, ironically, more susceptible to rivals who move first without sufficient verification.
Although they are limited, the technical solutions being created are genuine. An open standard known as C2PA was created by the Coalition for Content Provenance and Authenticity, a group comprising Adobe, Microsoft, Google, Sony, Nikon, the BBC, and Reuters. It incorporates cryptographic credentials into content at the point of creation, enabling downstream systems to confirm whether an image or video came from a registered device or platform. The uptake is quite encouraging, and the concept makes sense. The drawback is that it only functions on the way in: content created by tools that don’t comply with the standard won’t carry credentials; content that goes thru a camera or platform that complies with C2PA will. The least likely to incorporate verifiable provenance information into their output are the instruments most likely to be used to spread misinformation.
The C2PA standard, the detection technologies, and the suggested media literacy programs don’t adequately address a fundamental issue. Even before photorealistic AI media emerged, public confidence in journalism was already declining. According to the Reuters Institute’s 2024 Digital News Report, the average news trust rate worldwide is 40%, while the US is at 30%. These figures represent years of cumulative complaints, including alleged political bias, corporate ownership issues, and obvious reporting faults, which came ten years before the synthetic media disaster.
An already vulnerable system was made worse by the deepfake issue. As you see the two interact, it’s difficult to ignore the fact that the issue of synthetic reality affects more than just the difficulty of believing particular information. It provides a fresh, technically sound justification for those who were previously inclined to dislike journalists to apply that mistrust to everything they come across.
