Seeing Is No More Believing: India’s IT Amendment Rules of 2026 and Deepfake Governance

This piece is authored by Devansh Awasthi a 3rd Year Student of RMNLU

The New Face of Online Harm

The rise of deepfakes has altered the way government regulators view online harm. Previously, online harm was thought of primarily as the existence of misinformation including false texts, misleading captions, or altered photographs. Today’s technology brings synthetic or machine-made audio and video and images that present the illusion of someone actually saying or doing something that never really occurred. This brings a new dimension of risk. Just a few minutes can ruin a victim’s reputation with a morphed sexual image, or influence the masses with a false video of a politician. It is not only about removing the illegal material from the world wide web; it is more about identifying, labeling, and addressing physical effects (i.e., deepfakes). 

A New Format: Defining Synthetically Generated Information

In an effort to combat the problem of deepfakes, the Ministry of Electronics and Information Technology has created new guidelines when it comes to the use of deepfakes and other forms of synthetically generated content. These new guidelines will be introduced as amendments to the intermediary guidelines of the 2021 Information Technology Act and have become effective as of February 20, 2026.

Purpose of New Guidelines

The primary reason for the introduction of the new guidelines is to put definitions in place for:

The term “synthetically generated information” (SGI), refers to the content (audio, video, etc.) that has been artificially created or altered in a way to look real/authentic/true. This pertains mostly to deepfake/synthetic media applications while the normal multimedia modification methods should not fall within this definition.

Importance of Limited Definition

Having a limited definition of deepfake technology is essential in making sure we comply with these guidelines. Using AI/technology assistance for creation of artificial images, audio or videos will make it very complicated since there will be a large variety of definitions and/or regulations to implement no meaningfully implemented. Looking at the original framework we are focused on synthetic media creation that carries an illusion of reality and thus may lead to certain misleading situations. Therefore, it’s absolutely correct to focus on that. The issue with deepfake use is not that they get created using AI technology rather it’s because of the manipulated image which can successfully pose itself as something real. 

Responsibility of Intermediaries

There are some duties imposed by these new rules for the intermediaries providing the stage for SGI creation, modification, publication, or dissemination.The intermediary must inform the users that there may be penalties, including possible suspension or withdrawal of access, deletion of any content on the site, and possible criminal complaints, in addition to at least one of the above penalties, for dissemination of illegal synthetic material.

Intermediaries must also have reasonable technical means to prevent the use of synthetic materials for any of these unlawful and specific types of content such as children’s sexual abuse, non-consensual intimate images, obscenity, forgery of documents, propaganda related to arms, and material that incorrectly identifies individuals or instances existing in the real world.

Labeling Versus Watermarking

The most significant change is the requirement that all visual forms of SGI have a conspicuous and appropriately sized label while all audio forms of SGI must have a clear and noticeable labeling statement applied first. The use of an inflexible watermarking policy would be disadvantageous, as there are always differences in type and length in which synthetic materials appear. However, an adaptable compliance standard will allow each platform to customize its disclosure requirements on a case-by-case basis, based on the content type used to deliver SGI content.

Difficulties with Standards of Prominence

Although the customizable nature of the compliance standards will allow for creativity in compliance, it may also create difficulties because the word prominence could mean different things on various platforms and devices. What appears to be easily identifiable on a computer may appear blurred out on a mobile phone.Due to the loss of potential usefulness that a warning might provide when included in the beginning of a long video showing this video being created from scratch, a need remains for continuing technical assistance and regular review of these rules. In the absence of greater standardization around the definitions used, it will be difficult for the affected parties to determine whether synthetic media exists.

Under the Regulations, the obligations imposed on the Significant Social Media Intermediary have increased significantly. Those parties now have to obtain a statement from each user about whether or not the content that is being uploaded is synthetic or generated by an SGI and rely on technical methods to verify this statement and properly label the SGI that they identify. As far as significant platforms are concerned, this means that instead of responding retroactively to SGIs they must rely on proactive technology compliance. This means that the issue of SGIs and other synthetic media is no longer strictly a post hoc event but rather an integral part of the entire upload system and necessary technologies in terms of detection, monitoring and governance of SGIs and other synthetic media.

Nevertheless, performing active moderation might present certain challenges. For example, sometimes automatic detection systems are incapable of recognizing high-quality deepfakes.The danger of identification of legal content by platforms as illegal (e.g. satire, parody, journalistic or artistic works) increases due to safe harbour exemptions guaranteed by section 79 of the Information Technology Act, 2000. If platforms are uncertain about their safe harbour exemptions and are worried about losing them, extensive censorship might occur on large private platforms whenever they delete content before assessing it.

Compressed Time Intervals for Actions

The repressed intervals for actions that have been implemented under the IT Guidelines are yet another factor that heightens the chances of private censorship. The time period which is allowed for a platform to respond to a ruling or directive to remove unlawful content has been reduced from 36 hours to three hours. Similarly, the time which is allowed for a platform to respond to sensitive content pertaining to sexual privacy such as non-consensual images or deepfake pornography has diminished from 24 hours to two hours. Although the time allocated for response indicates urgency relating to cases of harm in regard to sexual privacy and dignity, prompt action does not always matter if the delay renders the remedy ineffective.Smaller intermediaries may lack the resources it takes to constantly review these things for it has team, or forensic capacity.

Protecting Against Over-Removal

What we need is not rejecting all expedited timelines. It is important that we meet the urgency for deepfakes and offers a balance of safety and speech. First, it does call for prompt response and different levels of protection concerning takedowns. A takedown request from a government agency must provide enough information to allow people to ask for a review of that request. For example, if a platform gets a request to take down something related to sexual privacy or impersonation, then that request must be handled differently. However, if that same platform had received a request dealing with borderline content such as satire, criticism, or political speech, those requests would have to be placed in context and responded to in a different manner. In addition, platforms are to keep track of all takedown requests made to them and give their reasoning whenever they are able.

Wider implications of the constitutional issue surrounding the balance (or lack thereof) in terms of the protection of safety and free speech must be considered. While synthetic media could be misused to cause harm, that is not the case most of the time. For instance, using labels instead of takedowns may serve to satisfy the requirement that an entity comply with its First Amendment obligations to subsequent public disclosure regarding censorship instead. Because of this, while platforms should at all times keep records of takedown requests, they ought to engage in prohibitive conduct between false statement and false belief. Accordingly, takedowns will be reserved only for the most obviously illegal or harmful content on each service.

Comparative Law: How Two Other Countries Responded to the Same Question

India does not exist in a legislative vacuum, and the first two jurisdictions to respond have come to very different conclusions.

The European Union has chosen disclosure rather than deletion for its responses. Article 50 of the Artificial Intelligence Act comes into effect on 2 August 2026, but this act pursues similar goals and objectives to the Indian Rules but places the burden on others. Specifically, providers of generative systems must have machine-readable information at the output stage, and deployers who create deep fakes (defined under Article 3(60) to mean manipulation or generation of material which makes it look real) must disclose this fact. A detailed guide and set of penalties, including a fine of EUR 15 million or up to three percent of global revenue, have been established as well.The Digital Services Act specifies no time frame for handling notices, only that such notices must be handled expeditiously, with diligence and without arbitrariness. The effect of U.S. law has been the opposite, providing another corrective. After May 19, 2026, the Federal Trade Commission began enforcement of the TAKE IT DOWN Act, requiring the removal of unwanted intimate imagery, as well as its non-consensual AI-generated forgeries, within 48 hours of a legitimate request. In comparison, India’s two-hour requirement for removal of harmful content is 24 times shorter, and occurs without the benefit of the good faith safe harbor that protects U.S. companies that inadvertently remove content later found to be lawful. In this context, speed is not what sets the Indian framework apart; it is the absence of any assurances against the mistakes that result from moving too quickly. 

The more serious concern arises from the case of Kohls v. Bonta. In August 2025, the Eastern District of California granted summary judgment against California’s AB 2839, ruling that the statute imposing limits on materially deceptive campaign media constituted content-based regulation that failed strict scrutiny, thereby concluding that the associated mandate for disclaimers was unconstitutional because the resultant labels created an intolerable burden that would overshadow the parody to which it related. The court reasoned that counter-speech and fact-checking were the alternatives available to the state that were less restrictive. The companion law, AB 2655, forcing services to take down deepfakes, was ruled preempted by Section 230 of the Communications Decency Act.The Ninth Circuit now has both rulings under consideration.

That reasoning undermines an assumption that has permeated much of the Indian debate, including the argument made earlier in this piece, that labelling is a moderate middle ground between inaction and censorship. It is not necessarily the case. When a particular label becomes highly visible, it approaches the point of replacing the expression being labelled. Thus, there is nothing technical, even though “prominence” comes to be understood in terms of proportionality. Seen in light of Article 19(2) and the recent decision that section 79(3)(b) of the Indian IT Act does not apply in the context of the ruling by the Indian Supreme Court in Shreya Singhal v. Union of India, the question must be answered as to whether the disclosures required by the Rule will, in fact, ease any misperceptions, or whether they will instead amount to punitive measures imposed on speaking legally.

Conclusion: Transparency – Not Censorship.

The IT Amendment Rules of 2026 may prove to have an important impact on how India approaches regulating its digital platforms as well as in recognizing that the problems posed by deepfakes cannot easily be addressed using existing content moderation techniques. In defining SGI, the Rule places additional responsibility on a variety of platforms to label content and shorten grievance timelines in the interest of synthetic media. The implementation of the Rules will ultimately determine their effectiveness. It will not be enough to expedite takedowns in the context of India’s deepfake problem. Success, in terms of enhanced transparency with regard to synthetic media, will be measured by the ability to take steps that allow visibility of such content without resulting in the harm of lawful expression.