This House would require AI-generated content to be visibly labeled.

tech

Generative AI now produces tens of millions of images and articles daily. Without disclosure, users cannot tell what they are consuming. With disclosure, every output carries a friction tax that erodes the technology's value.

Background

The EU AI Act, effective 2026, requires watermarking of synthetic media and clear disclosure of chatbots. California's AB 730 mandates labels on AI-altered political content. Adobe, Microsoft, and OpenAI co-developed C2PA, a cryptographic provenance standard that embeds origin data in image metadata. None of these regimes survive a single screenshot, which strips the metadata.

The scale challenge: an estimated 30% of new web content in 2025 is at least partially AI-generated, per a Europol study. Mandatory labeling at that volume creates a definitional problem (what counts as "AI-generated" when a human edits an AI draft?) and an enforcement problem (most platforms cannot verify watermarks at scale). The strongest case for the policy is that labels work for the use case that matters most (explicit deepfakes and impersonation), even if they fail for the marginal case of "human plus AI assist."

Government opens with

Asymmetric information markets always favor the producer; labeling restores consumer choice.

Opposition responds with

Mandatory labels create false confidence in what is unlabeled and discriminate against legitimate AI use.