India has made a significant alteration to its recent AI advisory, following a wave of criticism from both local and global entrepreneurs and investors.
The Ministry of Electronics and IT announced an updated guideline, reversing a controversial requirement for government approval before launching or deploying AI models in the South Asian market.
Under the revised guidelines, companies are now advised to implement measures such as labeling under-tested and unreliable AI models, effectively informing users about potential inaccuracies or uncertainties. This change comes in response to widespread disapproval received by the ministry earlier this month, notably from Martin Casado, a partner at venture firm Andreessen Horowitz, who labeled India’s initial move as “a travesty.”
The shift in stance represents a departure from India’s previous laissez-faire approach to AI regulation, which had been evident less than a year ago when the ministry declined to regulate AI growth, citing its importance to India’s strategic interests.
Although the revised AI advisory has not been officially published online, a reviewed copy confirms the alterations, signaling a move towards a more progressive regulatory framework. The ministry clarified that while the advisory is not legally binding, it sets the tone for future regulation, emphasizing the necessity for compliance.
Key points reiterated in the AI advisory include guidelines prohibiting the use of AI models to disseminate unlawful content under Indian law and preventing bias, discrimination, or threats to the integrity of the electoral process.
Moreover, intermediaries are instructed to employ mechanisms such as “consent popups” to explicitly inform users about the potential unreliability of AI-generated output.
In line with the objective of combating misinformation, the ministry maintains its focus on ensuring the identification of deepfakes and misleading content. Intermediaries are now encouraged to label or embed content with unique metadata or identifiers, eliminating the previous requirement to devise techniques for identifying the “originator” of specific messages.