India sets a one-month clock on its AI regulation paper
India just put a date on something it has dodged for years: an actual AI regulation framework, starting with a consultation paper due within the month. Meanwhile, a Japanese lab shipped a translation model it claims costs 1/30 of the alternatives.
India's government will publish an AI regulation consultation paper within a month, with AI safety and deepfakes at its center. IT minister Ashwini Vaishnaw announced the timeline at a World Development Report event in New Delhi, saying the paper "will form the base for AI regulations" — a real shift for a country that has long insisted it needed no dedicated AI law, only a techno-legal approach administered across existing statutes. Safety is "the most important point," Vaishnaw said, and he was unusually blunt about deepfakes, calling them a "nuisance which can cause huge problems in society" because "people believe them, and it sets a totally wrong precedent." He also argued that technology "cannot be regulated by law alone" and that industry must shoulder primary responsibility for cybersecurity, user harm and deepfake misuse. Why it matters: this is the first dated regulatory commitment from the world's most populous democracy, and it lands while industry lobby groups are actively pushing for an India-specific safety framework rather than copying the US voluntary pact. A government that spent years saying "not yet" now has a four-week deadline — hold it to that. (New to the terrain? Our explainer AI 101 — What is AI regulation? covers the machinery being built.)
Preferred Networks released PLaMo 3 Translate, a Japanese-first translation model it says costs about 1/30 of comparable translation APIs. The flagship is a 31B-parameter model (an 8B variant sits alongside it), built on the PLaMo 3 base PFN developed with Japan's NICT, and it is already running inside the company's PLaMo translation service. PFN's own benchmarks — treat them as vendor claims — put it above nearly every translation API on Japanese-English, with LLM-judge win rates of 87% over DeepL going English to Japanese and 58% in the reverse direction; the cost comparison against Claude Fable 5.1 works out to 1/85, though the pricing it's measured against is hypothetical since the API isn't public yet. The language list adds up to 53 (14 benchmarked plus 38 more trained). What's notable: neither the weights nor the API are out — PFN says a Hugging Face upload is planned — so for now this is a product, not an open release, and the "1/30 cheaper" headline rests on numbers only PFN has seen.
What to watch: whether Vaishnaw's paper actually arrives by early November, and whether PFN follows through on publishing the weights — right now the cost claim can't be independently tested.
Is a consultation paper with a one-month deadline real progress or regulatory theater? Tell us in the comments.
