Bob's Confetti: Phonetic Memorization Attacks in Music and Video Generation

Researchers discovered a vulnerability in AI music and video generation systems where phonetic prompts can bypass copyright filters. The 'Adversarial PhoneTic Prompting' attack achieves 91% similarity to copyrighted content by using sound-alike phrases that preserve acoustic patterns while evading text-based detection.

• →AI music generators like Suno and YuE are vulnerable to phonetic attacks that bypass copyright protection filters.
• →The APT attack method achieves 91% similarity to original copyrighted songs versus only 13.7% for random lyrics.
• →AI models bind acoustic patterns like phonemes and rhyme to memorized content rather than semantic meaning.
• →The vulnerability extends to video generation, with Veo 3 reconstructing original music video scenes from phonetic prompts alone.
• →Current semantic-based copyright defense systems fail to detect these phonetically-disguised prompts.