The Deepfake Threat Moved Faster. So Did We.

In the rapidly advancing field of voice AI, the indistinguishability between real and AI-generated media has become a significant challenge, as demonstrated by a game played at MWC Barcelona where 70% of participants, including experts, couldn't tell the difference. In response, the team has developed a watermarking API that now covers audio, images, and video, aiming to prove the authenticity of media at the moment of creation. Alongside this, significant updates have been made to their detection stack, including improvements in image detection accuracy and the introduction of reverse image search, which provides provenance data even for new or untrained synthetic media. The team also addressed the need for accurate AI-generated content without faces and introduced features like Zero Retention Mode for enhanced privacy. On the generation side, advancements in custom vocabulary and collaboration with NVIDIA for expressive NPC voices in games were highlighted. The company has made these detection tools accessible to the public through a free bot and browser extension, aiming to democratize access to deepfake detection. These developments reflect a proactive approach to the increasing threat of synthetic media, with ongoing efforts to enhance the capabilities of their tools.