Centre revokes AI deployment approval requirement, raises concerns on deepfakes
The Ministry of Electronics and Information Technology (MeitY) has rescinded its previous mandate requiring artificial intelligence (AI) models, large language models (LLMs), and algorithms to obtain explicit government permission before deployment. Issued on Friday, the ministry's new advisory, instead stipulates that these technologies should be made accessible to Indian users only after clearly indicating any potential unreliability in their output. The ministry continues to uphold need for a "consent popup" mechanism to alert users about possible false or unreliable results.
Initial advisory faced criticism
The original advisory, issued on March 1, faced significant criticism from companies worldwide. Several start-ups labeled it as detrimental to innovation. Despite the ministry's clarification that the advisory would not apply to start-ups, the criticism persisted. The initial mandate required all AI models and algorithms in testing or beta stages or deemed unreliable, to seek explicit permission from the government before deployment for Indian users.
Changes stress on due diligence and lawful content
The latest advisory emphasizes that intermediaries and platforms should not neglect their due diligence obligations. It insists that the use of AI models, LLMs, and algorithms should not facilitate the sharing of unlawful content as per Rule 3(1)(b) of the Information Technology (IT) Rules. This rule prohibits certain types of content such as pornography, child sexual abuse material, obscene material, grossly defamatory content or anything deemed unlawful.
Advisory highlights AI's role in electoral integrity
As India prepares for the general elections, the IT ministry's advisory also addresses concerns about AI-generated 'deep fakes.' It mandates that AI models and intermediaries shouldn't allow any bias or discrimination or pose a threat to the electoral process. The ministry has outlined guidelines for labeling such information with unique metadata or identifiers, applicable to audio, visual, text, or audio-visual content. If a user makes changes, the metadata should enable identification of such user or computer resources for tracking purposes.
