I Don’t Trust Visual Recordings Anymore. But One Day I Will Again
Due to the rise of generative AI and deepfakes, the author argues for a shift from a "presumption of authenticity" to a "presumption of fake" for digital media, placing the burden of proof on content creators.
Intelligence analysis by Gemini 2.5 Flash

A computer science professor advocates for a fundamental change in how we view digital visual content, suggesting that all recordings should be considered fake until proven otherwise. This paradigm shift is necessitated by advanced AI technologies that make it increasingly difficult to discern real from fabricated media, as illustrated by a case involving a suspicious missile strike v…
Imagine you're watching a video of your friend doing something silly, but then you wonder if it's really them or if someone used a clever computer trick to make it look like them. Because computers are getting so good at making fake videos, we now have to assume that any video we see might be fake until someone proves it's real. It's like when you play a game, and you have to show your cards to prove you have a winning hand, instead of everyone just believing you right away.
Analysis
Presumption of Fake
The proliferation of generative AI has fundamentally altered the landscape of digital media, eroding public trust in visual recordings. Hany Farid, a computer science professor, argues that society must transition from a "presumption of authenticity" to a "presumption of fake" when encountering online content. This means that instead of skeptics bearing the burden of proving a recording is manipulated, creators will now be responsible for demonstrating the veracity of their content. This shift, while potentially challenging for legitimate creators, is presented as a necessary adaptation to the current technological reality.
The article posits that this new stance is not merely a reflection of the current state of deepfake technology but also a pathway toward a future where trust in online media can be re-established. By mandating proof of authenticity, the hope is to incentivize the development and adoption of robust provenance signals and verification methods. This proactive approach aims to restore a baseline of credibility to digital information, which is increasingly vital in an era where misinformation can have significant real-world consequences.
Forensic Approach
Farid's work over three decades has focused on developing mathematical and computational techniques to analyze and detect manipulation in digital media. His methodology involves a meticulous forensic examination of suspicious videos, looking for inconsistencies that betray digital alteration. This approach was exemplified when he was asked to analyze a video purportedly showing a U.S.-made Tomahawk missile hitting a school, which had emerged weeks after the alleged event and was distributed by a news agency with a questionable record.
His analysis began by considering if a digitally added missile would exhibit physically implausible movement if camera motion wasn't accounted for. By aligning video frames using a stationary background landmark, he could observe the missile's trajectory. Furthermore, he assessed the plausibility of the missile's size in pixels under reasonable assumptions, indicating a systematic process to uncover digital tampering. This detailed, scientific scrutiny is crucial in a world where visual evidence can be easily fabricated, providing a counter-measure to the deceptive capabilities of generative AI.
Tomahawk Missile
The specific case of the alleged Tomahawk missile strike video highlights the immediate and critical need for deepfake detection. The video, claiming over 175 casualties, including many children, sparked international concern and denials from the U.S. president and military. The low quality of the video, its delayed emergence, and its distribution source immediately raised suspicions, prompting reporters to seek Farid's expertise.
Farid's initial assessment, even before detailed analysis, leaned towards skepticism due to these contextual factors. His subsequent forensic examination involved aligning video frames to analyze the missile's trajectory and evaluating its pixel size for plausibility. While the article cuts off before revealing the full conclusion of his analysis, the example underscores the practical application of his detection techniques in high-stakes scenarios where the authenticity of visual evidence can influence geopolitical narratives and public perception.
Key points
- The rise of generative AI necessitates a shift from presuming authenticity to presuming falsity for digital visual recordings.
- Content creators will increasingly bear the burden of proving the authenticity of their media.
- Computer science professor Hany Farid develops mathematical and computational techniques for detecting digital media manipulation.
- Forensic analysis involves examining video for physically implausible movements and inconsistent object sizes.
- A suspicious video of a Tomahawk missile strike highlighted the immediate need for robust deepfake detection methods.
The development of advanced detection techniques and the push for content labeling could lead to new standards for digital media, restoring a degree of trust in online visual information. This shift could foster innovation in authentication technologies, ultimately making it harder for malicious actors to spread misinformation effectively.
The widespread adoption of a "presumption of fake" could lead to pervasive distrust in all visual media, making it difficult to distinguish legitimate news and evidence from fabrications. This could further polarize society and undermine critical decision-making processes, as the burden of proof becomes increasingly complex and costly for creators.



