Even deepfake experts can't tell what's real anymore

Sam Gregory, an expert in deepfake analysis and member of WITNESS's Deepfake Rapid Response Force, warns that rapid advancements in generative AI have created a landscape where even professionals struggle to distinguish between real and synthetic media. As detection tools fail to keep pace with technological improvements, the integrity of visual and audio evidence is increasingly being undermined by misinformation campaigns.

What happened

A notable incident occurred in 2024 when a scam network circulated altered videos of former British Prime Ministers Keir Starmer and Rishi Sunak on Facebook and Instagram. These videos depicted the two men discussing the details of a high-return investment scheme. The deceptive content was viewed millions of times by people in the UK alone.

This incident highlights a broader trend where malicious actors leverage AI to misrepresent public figures. Beyond financial scams, the difficulty in verifying content has created a environment where political figures and legal entities can weaponize skepticism. By falsely claiming that genuine footage is AI-generated, individuals are able to cast doubt on authentic evidence. Gregory notes that detection tools are often unreliable, particularly with compressed content or languages they were not trained on, and can sometimes incorrectly flag real footage as fake. This creates a "dream scenario" for those wishing to deny the truth, as disproving a false claim takes significantly longer than the time it takes for misinformation to spread across social media. Gregory argues that this erosion of trust threatens the public's capacity to act, think, and judge accurately.