Skip to content
AI News HubLIVE
More
Summary only

Voice AI Systems Are Vulnerable to Hidden Audio Attacks

Research reveals that AI voice systems can be hijacked via imperceptible audio signals, with success rates up to 96%. The AudioHijack attack targets large audio-language models (LALMs) and can trigger malicious actions including data theft and unauthorized tool use.

SourceIEEE Spectrum AIAuthor: Edd Gent
Voice AI Systems Are Vulnerable to Hidden Audio Attacks
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions

Summary only

Under the source content permission, this page provides an editorial summary, key points, and a link to the original article.

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • AudioHijack uses imperceptible audio modifications to manipulate AI voice models with a 79-96% success rate.
  • The attack works on open and commercial models from Microsoft and Mistral, and can be reused regardless of user instructions.
  • Researchers demonstrated six attack types including denial of service, misinformation, and tool misuse.
  • Common defenses are largely ineffective; only attention monitoring works but can be bypassed.

Highlights and analysis are generated automatically and may contain errors. Check the original source.

More recent stories on this topic · Models