Do You Have to Label AI Content? The Aug 2 Rules

Two AI labeling laws, Article 50 of the EU AI Act and California's AI Transparency Act, went into effect on August 2, 2026. For freelancers, marketers, small publishers, and those sharing AI-generated content online, the key question is which obligations apply to them and which belong to AI model providers like OpenAI, Google, and Adobe.

The answer is that most obligations fall on model providers, with few specific duties for users. The critical rule is whether a user is a provider or deployer. Under the EU AI Act, providers are responsible for ensuring AI-generated content carries a machine-readable label identifying it as AI-created. This involves watermarking, content credentials, or cryptographic signatures.

Deployers have three obligations: disclosing when interacting with people, disclosing AI-generated or manipulated images/audio/video resembling real things, and disclosing AI-written public interest text if human editorial review has occurred. The transparency obligation has a crucial exception for text edited by humans with editorial responsibility.

The AI Act also mandates disclosure for text intended to inform the public on matters of public interest, with a notable human-review exception. The California law, CAITA, imposes similar obligations on providers with over one million monthly users but focuses on images, video, and audio. Both laws take effect on August 2, 2026, and extend to non-EU businesses whose AI output is used in the EU.