Laws track and differentiate real from fake

There isn't one single global law, but rather a growing number of laws and regulations around the world that require AI-generated content to be tracked and labelled. The most prominent are in the European Union, China, and the United States (California).

Let us understand them
Requires AI-generated text, images, audio, and video to have machine-readable markings (watermarking).
Users must be informed they are interacting with AI if they are using chatbots.
Providers must publish summaries of training data.

China
Requires a dual "explicit + implicit" labelling system.
Explicit labels are visible warnings (watermarks, text).
Implicit labels are invisible metadata or digital watermarks embedded in the file for traceability.
United States (California)
California AI Transparency Act (AB 853)
Requires providers of generative AI systems with over 1 million monthly users to embed traceable provenance data (e.g., digital signatures) in generated images, video, and audio. They must also provide free AI detection tools to users.

India
Draft Amendments to IT Rules, 2021
Proposed (not yet final)
Proposes mandatory visual and audio watermarks on all AI-generated content to combat deepfakes and enhance traceability.

How the Tracking Works in Practice
The "law" is often implemented through specific technical methods. A major example is how companies like Anthropic comply with the EU AI Act.
Embedded Invisible Watermarks (for text): Anthropic has begun embedding a watermark directly into the text generated by its Claude models. The watermark is invisible to human readers and doesn't change the meaning, but it is detectable by machines. Crucially, it travels with the text when copied and pasted, and may persist through some editing.
Cryptographic Provenance Metadata (for files): For image files (like .svg, .png, or .jpg), the system attaches signed metadata using the C2PA (Coalition for Content Provenance and Authenticity) open standard. This acts like a digital signature, indicating the file was generated or processed by an AI and whether it has been tampered with since.
Important Note: These tracking methods have limitations. For instance, heavily edited text may lose its watermark. The presence of a watermark is a signal, not an absolute, unremovable guarantee of origin.
Many major AI companies like Google, Meta, Microsoft, and OpenAI have signed on to these transparency commitments.

With AI or artificial intelligence entering all areas of life, we do need to understand it completely and differentiate the real from the fake. We hope the whole world recognises the real from the fake soon.

We hope the whole world recognises the real from the fake soon.

By Jamuna Rangachari