Skip to main content
Editorial pencil sketch of a magnifying glass examining two identical sound waveforms on a screen, unable to distinguish original from processed audio
Surface
MIDWATER

Standard Audio Processing Defeats State-of-the-Art Deepfake Detectors

VERIFIEDConfidence: 80%

The most widely deployed method for detecting audio deepfakes cannot tell the difference between a fabricated voice and a real one that has been cleaned up. That is the central finding of an unreviewed arXiv preprint from KTH Royal Institute of Technology (arXiv:2603.14033v1), submitted to Interspeech 2026 and not yet...

Create an account to read this article

Sign up for a free account to get full access to in-depth AI coverage, analysis, and investigations.

Related

BRIEFSurface

Google's Lyria 3 Pro extends AI music from a jingle to an actual song

Google announced Lyria 3 Pro on March 25, 2026, raising its AI music generator's output limit from 30 seconds to 3 minutes — six times longer than the base model released five weeks ago. The upgrade also teaches the model how songs are structured. That combination moves Lyria from novelty into something closer to a production tool.

BRIEFSurface

OpenAI's next model is nearly ready, and Altman says it can move the economy

OpenAI has completed pre-training on a new model codenamed "Spud," which CEO Sam Altman described to employees as "very strong" and capable of meaningfully accelerating the economy. The release is expected within weeks. The announcement comes with a consequential governance shift: Altman is stepping back from direct oversight of safety and security teams to focus on fundraising and infrastructure.

SIGNALSurface

OpenAI shutters AI video generator Sora

OpenAI has shut down Sora, its AI video generation platform, discontinuing both the consumer app and developer API with no replacement or migration path announced.