Only 36% of AI image generators include any machine-readable watermark and 0 major social platforms offer API-based watermark scanning, making synthetic media provenance unenforceable

Despite the EU AI Act requiring AI-generated content to be labeled and the voluntary commitments made by AI companies at the White House in July 2023, only 36% of image generators include any machine-readable watermark, only 16% provide visible deepfake disclosures, and just 8% fully meet the EU AI Act's visible-labeling criterion. No YouTube, Meta, or TikTok API exists for automated watermark scanning as of February 2026, and no standardized benchmarks exist for detection accuracy after platform-specific image transformations like compression, cropping, or screenshot capture. Why it matters: AI-generated images circulate without provenance metadata, so platforms cannot programmatically distinguish synthetic from authentic media at scale, so content moderators must rely on manual reporting which catches only a fraction of harmful synthetic content, so deepfake election interference and nonconsensual imagery proliferate faster than human review can contain them, so the entire premise of 'AI content labeling' as a governance strategy collapses because the technical infrastructure to enforce it does not exist. The structural root cause is that watermarking and provenance standards like C2PA are voluntary, no entity has authority to mandate adoption across the fragmented ecosystem of open-source and commercial generators, and the fundamental computer science problem remains unsolved -- no watermark has been demonstrated to be simultaneously robust to adversarial removal, unforgeable, and publicly detectable.