Hany Farid, a professor at the UC Berkeley School of Information and senior faculty adviser for the Center for Long-Term Cybersecurity, says artificial intelligence has advanced to the point where even the world's foremost expert in digital forensics can no longer reliably distinguish real images from AI-generated content. Farid, who is widely recognized as the leading authority on image analysis, told The New York Times that he now struggles to identify manipulated videos and images despite decades of experience in the field.
Farid has collaborated with Microsoft to develop PhotoDNA, software used by major tech firms to detect child sexual abuse material, and worked with the U.S. military to create scanning software aimed at identifying deepfake videos before the 2020 election. He said AI technology has improved dramatically in mimicking real shadows, projectile trajectories, plant life, human faces, and sound — eliminating many of the forensic clues that experts like himself once relied upon to detect manipulation.
What the Right Is Saying
Conservative commentators and free-market advocates caution against regulatory approaches that could stifle innovation or grant excessive government power over technology. Critics argue that mandated watermarking requirements would be easily circumvented by bad actors operating outside U.S. jurisdiction and could set a precedent for government control over digital speech.
Tech industry groups have pushed back against what they characterize as premature regulation, noting that detection technology is advancing alongside generation capabilities. The emergence of AI-powered tools like the Deepfake-O-Meter and Farid's own company, GetReal Security, demonstrates that the private sector is developing solutions to combat misuse without government intervention. Some Republican lawmakers have instead advocated for enhanced enforcement of existing fraud and impersonation laws rather than new regulations targeting AI development broadly.
What the Left Is Saying
Progressive voices are using Farid's warnings to renew calls for mandatory regulations on AI companies developing image generation technology. Democratic lawmakers have proposed legislation requiring tech firms to embed watermarks or authentication metadata in AI-generated content, similar to the Content Credentials movement advocating for authenticity pins that would allow viewers to trace a photo's creation and editing history. Consumer protection advocates argue that without such requirements, the public cannot be expected to defend itself against increasingly sophisticated deception.
Senator Amy Klobuchar has previously introduced bills requiring disclosure ofAI-generated political ads, stating that voters deserve to know when they are watching authentic footage versus fabricated content. Organizations like Common Sense Media have argued that tech companies prioritizing profit over safety should face liability for harms caused by their products, pointing to the 3,000% increase in deepfake-related fraud attempts as evidence of systemic negligence.
What the Numbers Show
According to cybersecurity firm DeepStrike, deepfake files increased from approximately 500,000 in 2023 to 8 million in 2025 — a sixteen-fold increase over two years. This surge coincided with a reported 3,000% increase in fraud attempts worldwide that leverage synthetic media.
The AI content detection market has grown correspondingly, with multiple companies now offering tools designed to identify machine-generated images and videos. Detection accuracy rates vary significantly across platforms, with leading services reporting success rates between 85% and 95% on standard test datasets — though experts note these figures often decline when applied to newer generation models.
Major tech companies including Google, Microsoft, and Meta have each committed varying resources to detection technology, though none have disclosed comprehensive data on how many deepfakes their systems successfully identify or miss. The Content Credentials initiative, backed by Adobe and supported by other industry players, has so far been adopted voluntarily rather than as a universal standard.
The Bottom Line
The gap between AI generation capabilities and human detection ability appears to be widening faster than countermeasures can close it. Farid's assessment that "within a year or two, our whole visual system will be utterly useless" reflects broader concerns among security experts about the implications for journalism, legal evidence, financial fraud, and political discourse.
Congress faces mounting pressure to address deepfake regulation ahead of the 2026 midterm elections, with both parties acknowledging the threat but differing on appropriate responses. Whether solutions come through industry self-regulation, mandatory technical standards, or enhanced enforcement of existing law, experts agree that current trajectories make some form of intervention increasingly urgent for protecting democratic processes and public trust in visual media.