India is walking back its recent advisory opinion on artificial intelligence after receiving criticism from many local and global entrepreneurs and investors.
The Ministry of Electronics and Information Technology on Friday shared its latest artificial intelligence recommendations with industry stakeholders, no longer requiring them to obtain government approval before rolling out or deploying artificial intelligence models to users in South Asian markets.
Under the revised guidance, companies are advised to label untested and unreliable AI models to inform users of their potential errors or unreliability.
The revision comes after India’s IT ministry came under heavy criticism from many prominent figures earlier this month. Martin Casado, a partner at venture capital firm Andreessen Horowitz, called India’s move “a travesty.”
The March 1 consultation also marked a reversal from India’s previous hands-off approach to AI regulation. Less than a year ago, the department refused to regulate the growth of artificial intelligence, arguing that the industry was vital to India’s strategic interests.
Like the original advisory earlier this month, the new advisory has not yet been posted online, but TechCrunch has reviewed a copy of it.
The ministry said earlier this month that while the advice was not legally binding, it showed it was “the future of regulation” and the government was required to comply.
The recommendation emphasizes that under Indian law, artificial intelligence models should not be used to share illegal content and should not allow bias, discrimination or threats to the integrity of the electoral process. It is also recommended that intermediaries use “consent pop-ups” or similar mechanisms to clearly inform users of the unreliability of output generated by artificial intelligence.
The ministry remains focused on ensuring that deepfakes and misinformation are easily identifiable, advising intermediaries to use unique metadata or identifiers to tag or embed content. It no longer requires companies to design a technology to identify the “originator” of any particular message.
