British startup Surf Security has created a browser that detects dipfakes

London-based startup Surf Security has unveiled a beta version of its browser, which the company claims is the first in the world with built-in detection of dipfakes created by artificial intelligence.

Artificial Intelligence vs. Artificial Intelligence

The tool, available in the Surf browser or as an extension, can determine with 98% accuracy whether a voice in an online communication is real or AI-generated.

The system is based on a military-grade neural network and uses State Space Models to analyze audio frames, identifying unnatural features common to AI clones. It supports working with a variety of languages and accents.

«We focused on speed and accuracy,» says Ziv Jankowitz, CTO of Surf Security. The neural network has been trained using the best voice cloning platforms, which allows it to work quickly, taking less than 2 seconds to analyze an audio file.

The neural network has been trained using the best voice cloning platforms, which allows it to work quickly, taking less than 2 seconds to analyze an audio file.

Expanded analytics capabilities

The tool is capable of verifying audio files from online video or communication apps such as WhatsApp, Slack, Zoom, Google Meet, and supports both recorded and live broadcasts. All the user needs to do is press a button to authenticate the voice.

WhatsApp, Slack, Zoom, Google Meet.

In the future, Surf Security plans to add a feature to detect dipfakes in images, expanding the browser’s capabilities.

Dipfake threat grows

Dipfakes have become a serious threat. This week, for example, BBC researchers discovered audio clips purportedly recorded by David Attenborough. In fact, they turned out to be dipfake recordings created to spread false information about politics and other topics.

Dipfake recordings have become a serious threat.

These examples are just the tip of the iceberg. Dipfakes are actively used to commit fraud, create political provocations and destroy reputations.

These examples are just the tip of the iceberg.

A tool to protect the public

Surf Security says the new detector will help protect businesses, media, law enforcement and even the military from the growing threat of AI cloning.

The fight against dipfakes, however, is a never-ending battle: «Voice cloning software is evolving every day,» Yankowitz acknowledges. – «As in any cybersecurity field, we have to win this ever-changing arms race. »

We have to win this ever-changing arms race.

A full version of the dipfake detection tool is expected early next year.