Overview
- India’s DPIIT working paper proposes an automatic licence to train on all lawfully accessed, non‑paywalled content with no opt‑out, with royalties payable only upon commercialisation.
- The proposal creates a nonprofit Copyright Royalties Collective for AI Training to collect and distribute payments, enforce compliance, and run a Works Database for payout eligibility.
- A government Rate Setting Committee would fix flat revenue‑linked royalty rates reviewed every three years, with obligations applying retroactively to already‑trained models.
- Nasscom and several big tech firms formally dissented, urging a text‑and‑data‑mining exception or machine‑readable opt‑outs and warning of costs, enforcement hurdles, and risks to startups.
- Developers would disclose only high‑level training categories, and the paper acknowledges gaps where sectors lack CMOs, raising questions about auditing global revenues and cross‑border datasets.