Why China’s New AI Content ID Rule and Claude’s Data Policy Signal a Turning Point

From September 1, 2025, China will require AI‑generated content to carry an “identity card” – a mandatory label that marks text, images, audio or video as AI‑created.

The “Artificial Intelligence Generated Synthetic Content Identification Measures” defines explicit labeling (visible tags such as “AI generated”) and implicit labeling (metadata hidden tags detectable only by machines).

At the same time, Anthropic’s Claude model changed its user‑data policy: it now defaults to using chat and code conversations for model training unless users opt out before September 28, 2025, and offers a 30‑day deletion window for those who refuse.

These developments highlight a causal relationship: massive data feeds power AI models, while content identification attempts to expose the AI origin of the output.

Large language models ingest vast, diverse data from the public web, turning it into a “super‑nutrient” that improves their ability to generate human‑like text, images, audio and video. This data hunger can lead to privacy‑invasive practices, as illustrated by Claude’s policy change.

When AI‑generated content becomes indistinguishable from human‑made material, deep‑fake risks and misinformation surge, prompting regulators to introduce the identification measures as a “post‑hoc” safeguard.

The article argues that effective governance must start at the data source, establishing stricter rules for collection, processing and training to prevent AI from “running off the rails” and eroding public trust.