Proteus: Automated Adversarial Robustness Testing for Audio Deepfake Detectors

Proteus is an automated framework developed by Resemble AI for testing the robustness of audio deepfake detectors against everyday audio transformations, which can affect the accuracy of such detectors. Introduced at ISSMAD 2026, Proteus systematically applies combinations of common audio processing techniques such as codec transcoding, added noise, and VoIP simulation to assess their impact on the detector's reliability. The study highlights a significant asymmetry where genuine audio is more easily misclassified as fake, a phenomenon that can be exploited by bad actors in the form of a "liar's dividend." Proteus tested over 17,000 audio transformation chains, with 72% being rejected due to quality degradation, but 4,847 chains successfully manipulated the detector's verdict while maintaining audio intelligibility and speaker identity. The findings, which are integrated back into the model's training data to improve its performance, underscore the security implications of audio deepfake technology and the challenges of maintaining detector accuracy in real-world conditions.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.