A widely shared image showed Pope Francis in a fashionable white puffer coat. Many viewers took it for a photograph, but it was made with an AI image-generation tool. That uncertainty—whether something records an event or was generated—is central to synthetic media. section 4.0
Synthetic media, also called AI-generated media, is content created or modified using artificial intelligence. It includes text, images, video, voice, and data. A deepfake is a more specific, historically deception-associated form, such as a fabricated recording that makes someone appear to say or do something they did not. Not all synthetic media is a deepfake, and not every use is deceptive. section 4.1
The same capability can serve different purposes:
- Creation: Individuals and small teams can produce images, video, and audio without the budgets or specialist teams once required for comparable work. section 4.3
- Protection: Altering faces can preserve a person's privacy; filmmakers used this approach to protect LGBTQ+ activists in Welcome to Chechnya. section 4.3
- Misuse: Impersonation, non-consensual images, and fabricated news-like content can harm people and undermine trust. section 4.1 section 4.3
Identifying synthetic content is difficult and increasingly adversarial. An image may contain blurred edges, inconsistent lighting, or conflicting contextual clues, but improvements in generation can reduce such telltale signs. A discontinued 2023 classifier correctly identified AI-written text only 26% of the time. Detection therefore cannot rest on a single visual clue or tool; authentication, policy, and public education also matter. section 4.4 section 4.15
Questions about who may create or use synthetic media extend beyond whether it looks real. A recreated actor can raise issues of consent and legacy; training models on creative work and generating new work raise unresolved copyright and ownership questions. Licensing and compensation are emerging responses, not settled answers to every dispute. section 4.8 section 4.14 section 4.15
Tracing where content came from may help, but proposed watermarks for generated text have limitations, and rules differ by jurisdiction. China's Deep Synthesis Provisions, for example, call for labels or tags on synthetically generated content. The practical question is not simply whether media was made with AI, but whether its origin, permissions, and possible harms are understood. section 10.23 section 10.31 section 4.14
