In a world where AI-generated content is becoming increasingly prevalent, concerns about misinformation and deception are growing. The sheer volume of AI-generated imagery and content has already had significant ramifications, with historians having to view the internet post-2023 as something entirely different from what came before. The rise of deepfakes, counterfeit content created using machine learning algorithms, has only exacerbated the problem. Tech giants such as Meta, Google, and OpenAI have proposed solutions to mark AI-generated content, incorporating visible markers, invisible watermarks, and detailed metadata to indicate their artificial origins. These efforts are supported by the Coalition for Content Provenance and Authenticity (C2PA), a group aiming to trace digital files’ origins and distinguish between genuine and manipulated content. While these initiatives aim to foster transparency and accountability, questions arise about who gets to decide what is real and if detection can be universal without empowering those with access to exploit it. Public trust in technology and innovation is already low, with skepticism towards institutions’ management of innovations and concerns about the rapid pace of technological change not benefiting society at large. However, reversing the lack of trust is crucial if watermarking is to be effective. Incidents like Google’s Gemini, which shadow-prompted biased images, and OpenAI’s CTO’s uncertain response about training data quality for Sora highlight the need for standards and public education about these tools. Consistency in labeling, accountability measures, open communication when things go wrong, and transparency throughout the process are essential. Without these measures, watermarking may serve as little more than a token gesture, easily circumvented or ignored. As deepfake election interference becomes a significant concern, solving the problem of trust in synthetic content is crucial before expecting watermarking to be successful.
