The End of Blind Trust: Hany Farid Argues for a 'Presumption of Fake' in the Age of Generative AI
Hany Farid argues that the rise of generative AI necessitates a shift from 'presumption of authenticity' to 'presumption of fake,' advocating for cryptographic provenance signals like C2PA and SynthID to restore trust in digital media.
S
··2 min readSciTech Editorial Team
Newsroom

In a pivotal shift in digital forensics, Hany Farid, a leading expert in media authentication, argues that the era of 'presumption of authenticity' for visual recordings has ended. Following a detailed forensic analysis of a controversial video depicting a U.S. Tomahawk missile strike on an Iranian school in March, Farid concluded that the burden of proof must now shift to the creator. In the age of sophisticated generative AI, he posits that we must adopt a 'presumption of fake,' where content is considered inauthentic until proven otherwise through cryptographic or technical verification. This stance reflects the growing difficulty in distinguishing genuine footage from deepfakes, a challenge that has eroded public trust in visual evidence.
The analysis of the missile video served as a turning point for Farid, who has spent three decades developing techniques to detect digital manipulation. Initially skeptical due to the video's low quality and delayed release by a state-affiliated agency, Farid subjected the footage to rigorous mathematical scrutiny. He aligned frames using stationary landmarks to verify the missile’s trajectory, which appeared physically plausible, and calculated the object's size and distance based on pixel data and the speed of sound. Despite these consistent indicators, the lack of obvious artifacts did not immediately convince him of its authenticity, mirroring the findings of a recent study from UC Berkeley. The study revealed that observers are now so wary of digital media that they struggle to accept real videos as genuine, highlighting a profound psychological shift in how we perceive visual truth.
To combat this crisis of trust, Farid advocates for the widespread adoption of provenance signals, such as content credentials and watermarks, which operate at the speed and scale of the internet. Unlike forensic analysis, which is slow and resource-intensive, these signals are embedded at the moment of creation. For instance, the C2PA standard attaches a cryptographically signed manifest to media files, recording the device, time, location, and any subsequent edits, allowing machines to detect tampering. Similarly, technologies like Google’s SynthID embed imperceptible signals into pixels or audio to identify the origin of AI-generated content. Apple has also moved in this direction, integrating reference images and metadata into its latest iPhones to enhance content verification.
The implications of this shift are profound for journalism, law, and public discourse. As generative AI becomes more accessible, the ability to verify the origin of media will be critical for maintaining societal trust. Farid’s work underscores the need for a new infrastructure of truth, where technical provenance replaces subjective skepticism. While the transition to a 'presumption of fake' may seem burdensome for legitimate creators, it offers a path toward restoring confidence in digital media. By embedding verifiable data into the creation process, we can create a future where visual recordings are once again trusted, not by default, but by design.



