In March 2024, the Ministry of Electronics and Information Technology (MeitY) issued a significant advisory aimed at offering guidance to intermediaries governed by the Information Technology Act, 2000 (IT Act) and Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021 (IT Rules). The recent advisory from MeitY stresses on the importance of adhering to legal and ethical norms when deploying AI models. This advisory, following a prior one from December 26, 2023, primarily addresses emerging challenges, particularly concerning the deployment of AI models by intermediaries.
Key Highlights
– The advisory requires intermediaries, which are online platforms and services that enable the sharing of user-created content, such as Facebook, Instagram, WhatsApp, Google/YouTube, Twitter, Snap, Microsoft, and ShareChat, to use AI models , Large Language Models (LLMs), or generative AI models to prevent the posting of illegal content or violations of the IT Act. Additionally, intermediaries must clearly inform users about the potential consequences of hosting such content through user agreements.
– The advisory also addressed a highly contentious issue regarding the use of under-tested and/or unreliable AI models, LLMs, and generative AI models. It mandated that the deployment of these technologies on the Indian internet must receive explicit approval from the Central Government. Following the advisory, a 15-day compliance period was established. It has been proposed that developers may need to conduct a demonstration or stress test of their products after complying and applying for permission. IT Minister Rajeev Chandrasekhar emphasized that this step aims to enhance scrutiny and rigor in the process. Additionally, if such under-tested models are provided to users, a ‘consent popup’ must accompany them to alert users about the potential fallibility of the AI system’s outputs. This requirement faced significant opposition, particularly from startups, arguing that it could hinder AI innovation in India. Many AI startups expressed concerns, criticizing the government advisory as backward-looking and lacking foresight, and raising worries about its negative impact on innovation.
– In response to the backlash, on March 15, 2024, MeitY reportedly issued a revised advisory to eight major intermediaries as mentioned hereinabove. The revised advisory softened the language of the initial one from March 1st and clarified that the requirement for government approval to provide under-tested and/or unreliable AI models to the public would only apply to these eight platforms. If any intermediary allows the creation of synthetic text/audio/visual content that could be deemed as misinformation or deepfake, it must be marked or tagged with a distinct, unalterable metadata or identifier. This is crucial for identifying the origin of the computer resource generating such content. If any modifications are made to the content, the metadata should be updated accordingly to trace the user or computer resource responsible for the alteration.
– The MeitY has urged all intermediaries to refrain from using AI models or Large Language Models (LLMs) that might perpetuate bias or discrimination. Additionally, the ministry has highlighted the potential adverse effects of biased AI systems on a nation’s electoral processes.
– Furthermore, intermediaries and platforms utilizing their computing resources to disseminate audio-visual content were advised to prevent the hosting and display of potentially misleading or deepfake content. To track its origin, all material coming from these computational resources that is shown on a platform has to be permanently identified by metadata or embedding.
Clarifications
Ministers Ashwini Vaishnaw and Rajeev Chandrasekhar have clarified that the advisory is non-binding and primarily targets major platforms, exempting AI start-ups. Compliance with the advisory remains voluntary, as no explicit penalties or enforcement mechanisms have been delineated. It is imperative to delve into the genesis of the opposition and its ramifications for the regulatory milieu.
Compliance Issue
Due to the lack of explicit penalties or enforcement procedures tied to MEITY’s advisories, adherence becomes a matter of choice rather than a legal requirement. This situation amplifies the uncertainty surrounding the legal standing of MEITY’s regulatory directives, raising questions about accountability and procedural fairness in technology regulation.
What is the Way Forward?
A thorough scrutiny and oversight of MeitY’s regulatory measures are imperative to ensure their alignment with democratic governance principles. By adhering to regulatory requirements, implementing robust risk management practices, and upholding ethical standards, businesses can leverage the transformative potential of AI while minimizing legal liabilities and safeguarding fundamental rights. As AI continues to reshape various industries, it is imperative for organizations to embrace responsible AI governance practices guided by legal and ethical principles.
Conclusion
MeitY officials have clarified the legal status of the advisory, stating that it functions as guidance rather than a regulatory framework, stressing the importance of careful and responsible AI implementation. The efficacy of these measures, particularly considering the dynamic nature of AI technology, remains to be evaluated, including its relevance in the digital realm.
The directives outlined in this advisory supplement those provided in the December 26, 2023 advisory, which required intermediaries to effectively communicate prohibited content to users, especially those outlined in Rule 3(1)(b) of the IT Rules. Although MeitY is authorised under Section 13 of the Rules to provide publishers appropriate information and advisories, it is unclear if the provision of advisories on AI governance is covered by the Rules, which casts question on the legality of the authority. The criteria for categorizing “significant/large platforms” versus “start-ups” lacks clarity. Likewise, the standards for assessing “under-tested” and “unreliable” AI are undefined, making voluntary compliance challenging.
Authors: Muskan Modi and Malabika Boruah