Google has opened its SynthID Detector to the public, allowing users to check if an image, video, or audio file was generated or edited using supported AI systems. This move comes as AI-generated content, such as deepfake videos, becomes increasingly difficult to distinguish from authentic recordings.
The SynthID Detector embeds an imperceptible digital watermark into AI-generated content, which remains detectable even when media is modified. According to Pushmeet Kohli, vice president of Science and Strategic Initiatives at Google DeepMind, the technology has already been used to watermark over 180 billion images and videos and 240,000 years of audio.
The detector checks content for a SynthID watermark and indicates whether supported AI technology was involved in creating or modifying the media. This can help journalists, fact-checkers, businesses, and social media users verify the authenticity of digital content.
A recent investigation by DUBAWA found multiple Facebook pages publishing AI-generated videos portraying fictitious security incidents in Nigeria. One video falsely depicting American troops heading to Nigeria recorded over 966,000 views.
AI-generated footage can exploit existing public anxiety around insecurity and make an event that never happened appear to have been captured on camera. An AI expert cited in the investigation stated that realistic-looking footage should not automatically be treated as evidence of an event.
Deepfake videos impersonating prominent Nigerians, such as Senator Natasha Akpoti-Uduaghan, Peter Obi, and Ngozi Okonjo-Iweala, have been used to promote targeted investment schemes. A manipulated video appeared to show Peter Obi promoting an investment scheme promising users millions of naira.
The use of deepfakes in online fraud means that users can no longer rely solely on a familiar face or recognizable voice to determine whether a message is authentic. The SynthID Detector can help users identify AI-generated content and make more informed decisions about the media they consume.
The expanded SynthID Detector supports watermarks from partners including OpenAI, NVIDIA, and Kakao, with Apple expected to join the ecosystem. However, there is an important limitation: a failure to detect a SynthID watermark does not prove that a piece of content is authentic or human-created.
Google’s SynthID Detector provides a tool for identifying AI-generated content.
