India just took a big step on synthetic media

The government has amended the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, to treat AI/deepfake media as “information” under the same compliance framework, and it adds two practical requirements:

• Mandatory labeling of AI-generated / synthetically altered content, with embedded identifiers/metadata where technically feasible, and platforms can’t allow labels/metadata to be removed or suppressed once applied.
• A much faster takedown clock: platforms must act within ~3 hours on government/court directions (down from 36 hours).

This matters because it pushes the conversation from “nice-to-have disclosure” to operational readiness: provenance signals must survive reposting, compression, and the normal chaos of distribution at platform scale.

If you’re building or hosting GenAI content, the message is clear: start implementing durable labeling + provenance now, because this is becoming the new baseline across major markets (and timelines move faster than product cycles).

DeepMark helps GenAI teams comply with these kinds of rules by adding a durable provenance signal directly into audio and video, not just a label that can be stripped. Our watermark survives common real-world transformations (recompression, platform processing, edits, and even many AI “enhancement” pipelines), and our SDK can run in real time on commodity CPUs or on-prem. In practice, that means you can attach an accountable “who/what/version” ID to synthetic content at creation time and still detect it reliably downstream, even after it’s been reshared and modified.