AI voice and deepfakes

AI voices are now good enough that listeners cannot always tell them from people. Lawmakers and courts have responded on three fronts: telling people when they hear an AI, deciding whose voice may be cloned, and stopping deepfakes around elections. Many of these rules were written for social media and ads, so how they apply to phone calls is often unsettled, and each guide says where that is the case.

Start with India's AI labelling rules. Since 20 February 2026 the IT Rules have required platforms that let people make synthetic audio to open it with a spoken label. The rules do not name phone calls, so the safe course is to say at the start of every AI call that the caller is an AI voice. The EU is clearer: Article 50 of the EU AI Act has applied since 2 August 2026 to voice agents that call or answer people in the EU, including agents run by companies based elsewhere, and a breach can cost up to €15 million or 3% of worldwide annual turnover.

Voice cloning law in India follows the court cases, from the Amitabh Bachchan order in 2022 to the Arijit Singh order against AI platforms in 2024. Cloning is not banned and no statute deals with it directly, but High Courts keep restraining commercial use of a recognisable person's voice without consent. Cloning with clear written consent for an agreed use is the low-risk path, and the guide lists what a consent form should cover. For voiceprints under US and EU law, see when a voice counts as biometric data.

Political campaigns have extra rules. The Election Commission's AI guidelines tell parties and candidates to label AI content, not to imitate anyone's voice without consent and to take deepfakes down within three hours. Bulk voice messages are banned in a polling area during the 48-hour silence period, and TRAI's rule that automated calls be declared to the operator applies from 27 November 2026.