An overview of synthetic content risk controls published by NIST notes that a single detector or watermark is insufficient to reliably prove audio authenticity, that watermarks may fail due to compression, resampling or malicious attacks, and that metadata can be stripped or forged.