Google finds 240,000 years’ worth of AI music with new detector tool

Google has unveiled a new AI detector tool designed to check if an image, text, video or audio file was made using artificial intelligence.

The SynthID Detector draws on imperceptible watermarks on AI-generated media, with Google claiming to have already watermarked over 180 billion images and videos, as well as 240,000 years’ worth of audio content.

The rollout of the detection technology follows the sentencing of a 54-year-old man in the US this week to 18 months in prison for AI-assisted streaming fraud.

Michael Smith from North Carolina ran the scheme for seven years on services like Apple Music and Spotify without being detected, earning millions of dollars from the platforms through hundreds of thousands of AI-generated songs.

An early version of SynthID Detector was launched last year for media professionals to use as a verification tool. The latest version is available for anyone to use.

“As AI tools become better and easier to use, having clear context about the content you encounter online helps you make informed decisions about what to trust,” Pushmeet Kohlo, vice president of science and strategic initiatives at Google DeepMind, wrote in a blog post.

“This tool joins our built-in verification features in Search, the Gemini app, and Chrome, which now regularly handle over 1 million requests daily.”

The watermarks do not change how an image or video appears to human users, and do not impact the quality. They are also designed to still work even after a filter is applied or an image is cropped.

The detection tool currently only works with content generated using Google, OpenAI, Nvidia, Kakao and Apple platforms.

The launch of SynthID Detector comes just days after OpenAI unveiled an invisible watermark to ChatGPT text outputs.

The textGrain system subtly alters word selection so that it reads normally for human users but can be picked up by automated systems that know the secret patterns.

Anthropic has announced similar technology for its chatbot Claude, though both companies warned that the watermarks are not that effective when a human edits or contributes to the AI-generated text.

The tools come in response to the European Union’s AI Act, which requires generative AI products to make their outputs identifiable by automated systems.

Google’s latest detection tool can be found through its AI division DeepMind.