ERIS: Evolutionary Real-world Interference Scheme for Jailbreaking Audio Large Models

Researchers developed ERIS, a new framework that uses genetic algorithms to exploit Audio Large Models (ALMs) by disguising malicious instructions as natural speech with background noise. The system can bypass safety filters by embedding harmful content in real-world audio interference that appears harmless to humans and security systems.

• →ERIS uses genetic algorithms to optimize real-world audio interference as a carrier for jailbreaking Audio Large Models.
• →The framework can disguise malicious instructions as natural speech with harmless background noise to bypass safety filters.
• →Testing shows ERIS significantly outperforms existing text and audio jailbreak methods across multiple ALMs.
• →The research reveals that innocuous real-world audio interference can be weaponized to circumvent AI safety constraints.
• →These findings highlight critical security vulnerabilities in current Audio Large Model alignment strategies.