AI Liability For Crimes - Comparing India’s IT Act And The EU’s AI Directive On Defamation

Author: LawSphereX

Category: Criminal Law

Published: 2026-08-23T17:45:27.494577+00:00

Updated: 2026-08-26T17:12:22.879+00:00

AI LIABILITY FOR CRIMES: COMPARING INDIA’S IT ACT

AND THE EU’S AI DIRECTIVE ON DEFAMATION

Satadru Majumder, B.A. LL.B. (Hons.), Xavier Law School, St. Xavier’s University, Kolkata, India

ABSTRACT

The recent runaway success of artificial intelligence, particularly the large language model variety such as ChatGPT, has also presented new wrinkles to the legal environment, especially when it comes to the law in the case of defamation produced by artificial intelligence. In this paper, the researcher will conduct a comparative study of the liability regime that can be found in the Indian Information Technology Act, 2000 and the proposed European Union AI Liability Directive, 2024. It discusses the legal issue whether the generative AI, like ChatGPT, is liable when characters of a defamatory message that is generated independently by the AI appear in an email by a user that concludes that the AI is liable. In India, statutory machinery is barely ready to address the harms that AI propagates.

In earlier IT Act, section 79 was meant to only take care of the passive intermediaries rather than generative systems that are able to create content all on their own. Although Sections 499 and 500 of the Indian Penal Code provide the said liability in cases of defamatory statements, intentional or not, there lies a vast difference between how the section prepared to deal with intentional defamatory statements and how it will be able to deal with the defamatory, intent-less post-AI processes. Case laws like Shreya Singhal v. Union of India further makes liability more complicated introducing unrealistic and unreal standard of actual knowledge of liability by intermediaries, which are rather unproductive with AI behaviour. Presumably, the EU approach known as the AI Liability Directive is more progressive.

It puts the liability on developers of high-risk AI systems strictly and places it so that a developer must prove liability as opposed to a victim. This is a very transparent approach where the claimants can obtain compensation by merely proving that they have been harmed, and, conversely, the developers need to prove that they have indeed met the safety criteria. With the examples taken out of life, like misrepresentations created by ChatGPT about real people, the paper highlights the different results of lawsuits in different jurisdictions. In India, the law is so old that it puts

procedural and legal obstacles in the path of the victim, but the Directive in the EU provides an easier route to justice and even more so, a reason to prevent risks by the developers. The paper ends by emphasizing the fact that India needs to revise its legislative structure as soon as possible. With the example of the EU, India can develop a similar system that enhances innovation in AI, but can at the same time protect against the reputational consequences of the use of AI in defamation attacks. Introduction Over the past few years, artificial intelligence (AI) has experienced an explosive increase, and the generative AI platforms such as ChatGPT can be regarded as a revolutionizing shift in terms of technology. There is an increasing influence of digital communication by these systems, capable of providing clear and convincing written output.

Yet, they also have critical legal implications over which the key one is the defamation problem that the AI content may produce. In case when an AI system creates and spreads fake information that demeans a person, there appear critical legal issues: Who is to blame? Is it possible to prosecute an AI? Do our current laws provide enough safeguard? One of the interesting examples that point to these dilemmas concerns the radio host Mark Walters who ChatGPT wrongfully accused of embezzlement. Framed in this light, these types of incidents, as well as others, depict the physical reputational and psychological damage AIgenerated misinformation can cause. They also highlight the gap in laws, as far as AI responsibility is concerned.

Laws, such as the Information Technology Act, 2000 in India and general defamation statutes of the Indian Penal Code, were written with human agents in mind, failing to take an autonomous technology into consideration. Particularly, section 79 of the IT Act grants the safe harbour to intermediaries and is unable to differentiate between passive hosts and generative systems such as ChatGPT. Conversely, the European Union has started reforming its legal framework by means of such initiatives as the AI Liability Directive (2024) that sees developers of high-risk AI systems become liable entities thus holding them accountable and ensuing compensation of the victims. This production-based liability shift recasts the question into the problem of product-like regulation rather than the responsibility of the platform.

Both approaches will be critically examined in this paper in order to see whether the AI system such as ChatGPT may be prosecuted within the existing legal frameworks and whether each framework supports or hinders justice to the victims of AI defamation. Drawing on case studies and the principles of legislation, the paper will attempt to plot the way ahead in each of the Indian contexts, setting the delicate balance of what innovations require and what must be done to protect the victims of harms caused by AI. Literature Review The scholarly debate about AI-derived defamation and responsibility has passed through three phases, each related to significant changes in AI technology and ability. The first of these stages can be called the period of narrow, rule-based AI, beginning in 2010, up until circa 2017.

Such authorities of the era as Sartor (2016) offered to implement the conventional approaches to product liability on autonomous systems. The explanation was that, similarly to faulty consumer products, AI systems might be harmful even when there is no human malpractice. The proponents of this practice argued that since developers develop these systems, they must be liable in case of any damages caused by it owing to its nature of uncertainty as far as machine learning is concerned. Nevertheless, this pioneering literature had not foreseen the special problems that generative AI of today creates, and in particular the way that they can and do create entirely original and frequently unpredictable materials. In 2018 to 2022 the landscape was dramatically changed with the introduction of transformerbased large language models.

The advent of this new wave of AI brought a more scholarly discussion. One of the central texts in the development of the concept of algorithmic strict liability is the work by Kaminski (2021), who pointed out the legal issues of generative models that break the deterministic programming system. As opposed to the previous systems, these models are prone to unpredictable behaviour because of their probabilistic nature and the necessity to be trained on much data. This brought on a developing realization that the current regimes of liability, which were reflective of the platforms that used user-generated content, were not really made to control AI that autonomously creates its own potentially defamatory content. This worry was echoed by other researchers such as Pasquale (2022) who broached the issue of the black box dilemma.

Many of the larger language models operate behind the veil, so it is virtually impossible to prove factors that have long been common incrimination at the legal

level: foreseeability or intention. Meanwhile, scholars of comparative legal studies, including Urban and Hoofnagle (2020) would start examining how the different jurisdictions could adjust their systems to address this mounting danger, it was not yet clear what concrete measures could be taken. With the emergence of ChatGPT and other services, the situation in the discussion of the academic community was characterized by an abrupt shift in the phase of the urgency of the conversation since 2023. The need to concentrate on real-world implications of AI-facilitated defamation emerged in the scholarly circles. Recently, Wu (2023) went a step further in that they have proposed to switch to the liability liability models that do not focus on the intention of the creator but concentrate on the damage that AI outputs cause.

Such an approach is gaining popularity in the European Union, and it can be observed in the progress of the AI Liability Directive (2024), which reflects a lot of these emerging concepts. The more recent cases, e. g. , a case by Veale and Zuiderveen Borgesius (2024) develop a comparative point of view, and reflect on how enormous the difference between the approaches to distributing liability may vary in different jurisdictions. Their study of the issue is indicative of an increasing divergence between the EU approach of active regulation and a more open or fragmented regulatory system preventing the outlook in other countries such as India and the United States. In India, the reaction in the academic realm has been late about these international trends.

An example of one of the early efforts to attempt a discussion of AI liability is present in a study by Mishra (2022), which was however largely limited to the intermediary protection under Section 79 of the IT Act. It failed to touch on the meanings of autonomous generative systems. Correspondently, an inability to identify the legal gaps created by the autonomous behaviour was identified in the 2023 analysis by Rao who demonstrated an approach similar to the one addressing AI-generated defamation similarly to human-generated content, i. e. , as functionally equal. Gupta (2024) took the discussions a step further by directly criticizing the lacking state of the legal framework that India is currently operating with, explaining why India should change its legal framework to accommodate AI. Nevertheless, there is an emerging global agreement on some of the fundamental reforms.

Lemley (2024), in his turn, points at behaviours incentives of strict liability regimes and claims that in this way, it is possible to encourage designers to take some proper safety precautions

without suppressing innovation. Crawford (2023), another supporter of this opinion, adds to it the importance of having transparency in the creation of AI, requiring any training data, decision logic to be disclosed when damage is involved. Alternative manifestations of compensation models have also been provided by certain scholars. These are insurances pools funded in the industries as well as no-fault systems that are funded in case of a prompt redress against an injured party. The proposals are meant to bypass the costs of evidence that are now impeding AI defamation suits. Another future place of interest is the connection between AI liability and fundamental rights. The case study analysed by Poudel (2024) about the EU [right to explanation] demonstrates the opposite, i. e. , how legal transparency can become a weapon of the injured consumers of AI-based decisions.

On the same note, Kapoor (2024) associates the constitutional protections of India, including the right to privacy and the right to free expression, with the necessity to establish a reimagined legal framework that may certainly respond to harms caused by AI. Irrespective of this development, significant arguments still exist. There exists a gap between the scholars on the one side supporting the statement of broad liability imposed on developers and the other side highlighting the exemplification of excessive regulation that might kill innovation. It is a debate which has relevance even in the Indian scenario in particular where being digitally competitive has to be balanced with the need to ensure protection of citizens.

Moreover, they comment that empirical evidence on the potential impact of various liability regimes on innovation or access to AI technologies in developing economies is missing. Academic thoughts are also slowly getting affected by the judicial developments. The 2024 German example of Schrems v. DeepL used the principle of strict liability to an AI translator that created libelous utterances, which was noteworthy because it was a considerable step change in how AI systems answer to courts. This case has stirred up a new intellectual interest in what the courts around the world may interpret or reform the existing laws in reaction to generative AI.

As a conclusion, the literature indicates four key themes as the inadequacy of traditional defamation and intermediate models of liability; the increased popularity of output regulatory and strict liability models; the dire necessity of reforms of the jurisdiction-developing proportions, in particular, in India; and the necessity to adjust AI governance into legislation,

which should adhere to the universal principles of constitutionalism and human rights. These observations are the intellectual basis on which this paper will be comparing and offering recommendations. Legal Framework in India Challenges and gaps The current law framework of India is mostly incapable of dealing with the specific peculiarities of the problem of AI-based defamation. The statutes of the country, especially the Information Technology Act, 2000 (IT Act), and the provisions on defamation of the Indian Penal Code (IPC) were developed during a time when human beings were the creators and deliverers of digital content. They therefore fail to appreciate the arenas brought up by generative AI that by way of being automated has the potential to produce an action of defamation without a direct intervention of any human being.

The centre point of India liability regime towards intermediaries is Section 79 of the IT Act. It offers a safe harbour to the intermediaries which protect them against liability of third-party content is they are due diligent in their efforts to remove illegal content when they are given information about the illegal content by the means of receiving a legal orders or governmental directives. Although such a method could have been effective on a platform such as social networks, it is not applicable enough on AI-based once such as ChatGPT since it creates content independently. Such systems are not content-hosting systems, but content generator systems. Therefore, by classifying them as intermediaries in Section 79, the loophole of accountability is opened. This limitation has been backed by judicial interpretation. Having said that, according to the Supreme Court in Shreya Singhal v.

Union of India (2015) had established a high bar on liability which made it necessary that the intermediaries should have actual knowledge regarding illegal content. Through the celebration of the ruling as paving the way to protect the freedom of speech, this standard can hardly, practically, be achieved regarding generative AI. Such systems are not predictable and without any considerable amounts of false or harmful content coming out of the system at any point in time, and in that sense, they are almost impossible to accurately predict or address in real time. The problem is compounded by defamation or criminal defamation laws formulated in the Section 499 and 500 of the IPC of India. Such provisions are pre-conditioned by mens rea,

intent to harm. Nevertheless, AI systems do not possess consciousness and intent in their nature. In the scenario where a generative model (such as ChatGPT) unnecessarily incriminates an individual, there is no legal ground in Indian law now to attribute intent to the AI or the individual creating the AI. Consequently, victims are left with no avenue of legal redress even when their damaged reputations are in grave cases. In addition to this, complicating the matter is the Digital Personal Data Protection Act (DPDPA), 2023. Although the protection provided by this legislation regards the privacy of individuals, it cannot be applied to the particular harm caused by AI, such as defamation or misinformation. It only addresses the acquisition of consent on data processing and the role of a regulator of the data fiduciary, but not the realm of autonomous and content-generating systems.

In a similar case, the Intermediary Guidelines (2021) impose a set of obligations on online platforms regarding the removal of so-called unlawful content, yet, these obligations have no enforcement mechanism and fail to differentiate between the misinformation generated by human beings and by AI technologies. In spite of these inadequacies, there have been timely policy indications of the increasing danger associated with the unregulated AI. In March 2024, the Ministry of Electronics and Information Technology (MeitY) published an advisory recommending that unreliable AI models must get prior governmental approval before they are deployed. Nonetheless, the swift opposite shock in the industry on the ambiguous nature of the language and its imprecise implementation caused a quick erosion of the advisory.

Similarly, another upcoming legislation, the proposed Digital India Act that will essentially replace the IT Act, also has mechanisms of controlling AI in it, and as the draft versions of this legislation appear, no coherent liability framework has been outlined to cover generative AI. The Indian judiciary has been giving out ambiguous messages on AI dependency as well. The case of Google LLC v. The Delhi High Court refused Google (2024) safe harbour and ruled it could not provide trademark infringement through its AdWords platform. Although the judgment indicates an increased propensity to challenge the conduct of intermediaries, there exists uncertainty of whether the reasoning will be applied to AI developers.

In the meantime, the case of the Punjab and Haryana High Court using ChatGPT in 2023 to conduct legal research is troubling, as it seems that the judicial system as a whole is fully acquainted with the concept of AI being used without a comprehensive understanding of its reliability, as the

debate on the issue of liability becomes complicated. Overall, there are three fundamental weaknesses in India as regards its regulatory framework with the issues of obsolete legal definitions that fail to address the generative abilities of the AI, evidentiary threshold that makes victims impossible to redress, and enforceable provisions of AI specific protections. As long as legal issues are unclear, victims of AI defamation will essentially be unable to do anything against it and developers will have little responsibility. The best way to smooth this situation out would be to have India quickly amend Section 79 by ensuring safe harbour is not available to any autonomous content generator, and hard liability is made directly on developers of AI using high-risk applications. Also decided should be the same transparency principles similar to those in the EU.

The reforms are necessary not only to support the constitutional rights to privacy and free expression, but also to enable people to have confidence in the fast-evolving digital environment of the Indian society. The AI Liability Directive of EU: A Proactive Approach to Responsibility The AI Liability Directive (AILD) of the European Union, preliminary suggested in 2022 and harmonized with the outline 2024 AI Act and revised Product Liability Directive (PLD), will be a radical attempt at a law to supply legal mechanisms that will respond to the challenges of using artificial intelligence. Specifically addressing an otherwise invisible harm of AI systems under an otherwise dormant civil liability model, namely defamation, discrimination, and psychological distress, the AILD is aimed at the modernization of the civil liability system.

In contrast to the current fault-based system that is racked by the problem of proving negligence in AI complex settings, the AILD seeks to shift the burden of proof in favour of the victim without sacrificing the following incentive to innovate and to develop responsibly. The Directive is the result of a realization, which has sprung up in the EU that traditional tort law is unsuitable in cases of AI-related harm. In the majority of the traditional systems, a claimant should demonstrate a fault besides establishing a causal connection between the act of the defendant and the damage arising. At this standard, it becomes hardly possible to meet when it comes to autonomous systems, and generative AI, in particular, that can create content that can be harmful without a specific human influence or predictability. The AILD fills this gap by proposing innovative solutions, i. e.

, a rebuttable version of the presumption of causality and increased access to the evidence of the AI contributor developers to the case, which are

specifically built to assist victims who have to deal with the intricate nature of harm caused by AI.

At the heart of the Directive there lies its rebuttable causality principle. According to Article 4, when a victim manages to prove that a defendant breached his or her obligations related to the AI Act (e. g. , not performing an adequate risk assessment or not supervising an AI system through human intervention), and that the output of the system was likely to produce harm, the courts should assume causation. It is an assumption that can only be overturned with evidence put forward by the developer. As an example, in case an AI-driven tool discriminates against some candidates during a hiring process, the employer or developer would have to demonstrate that the damage was not a result of a system breakdown. Such structure is especially strong against high-risk AI systems, such as those present in the fields of health care, finance, and state administration.

The same can be said about the disclosure mechanism of the Directive that enables the courts to order AI providers to present records like training dataset, risk assessment, and decision log. Such documents sometimes play a vital role in elucidating the working of an AI model and the reason behind a certain decision. To a victim, these disclosures are priceless especially when a victim is seeking defamation or any other injury to his/her reputation. Although this evokes some issues of trade secrets and intellectual property, they are resolved by the AILD, which obligates courts to respond to the issue of proportionality, along with providing confidentiality protection to defendants. The second strength of the AILD is that it is closely aligned with the AI Act, especially in the high-risk systems classification.

In the event a developer does not adhere to transparency or oversight expectations as provided in the AI Act, this will automatically fulfil the fault element under the AILD. Such smooth integration will allow consistency in the AI regulatory ecosystem of the EU and become a potent deterrent of irresponsible AI use. The Directive is more lenient in the case of the lower-risk systems. In the creation of such systems, however, victims of damage have to prove that it is excessively difficult to show causation at which instance the courts are allowed to make use of the presumption of causation. This implementation scale is promoting innovation in places that have low-risks and increases focus on systems that have high-risks of harming others.

Although there are plenty of strengths that can be attributed to the AILD, the program has received certain criticisms on account of some industry groups and policymakers who suggest that it may either overlap or go against the revised PLD. Other reasons suggest that it will impose costs of compliance that might stagnate small- and medium-sized developers of AI. The European Parliamentary Research Service (EPRS), in turn, in 2024 suggested turning the Directive into a wider Software Liability Regulation. This new format would consolidate into a single regime all software-related liability and it would obviate the necessity of national legislation of such within individual EU Member States which was previously required under the Directive.

The EPRS further suggested the application of strict liability to some types of AI uses among them the use of autonomous vehicles and the development of the rule on a joint liability along the AI development chain. The European parliament is reviewing these proposals in its Legal Affairs Committee. Considering the frameworks outside the EU is very much in comparison to India IT Act, it seems that the AILD is much more exhausting and visionary. Section 79 of India protects systems and developers under the condition that they had no awareness of damages caused, which is not possible in the case of systems as unpredictable as generative AI. This question is avoided through the AILD because it simply sees that causal harm should be used, rather than knowledge or intent, to hold people accountable.

Even in the United States, where Section 230 of the Communications Decency Act provides broad immunity to the websites that host it, the entire federal response to the problem of AI-based defamation remains unified. In the event of the enactment of the AILD, the approach of AI creators to risk management will be dramatically altered. The onus of this law will probably encourage firms to maintain numerous audit logs, perform frequent risk assessment, and use the human-in-the-loop oversight tactics in order to satisfy safety requirements. Simultaneously, it is likely to also lead to the development of specific AI liability insurance policies, further formalizing the AI development. Nevertheless, some critical challenges are left.

These are the operationalization of the excessively-difficult-to-use clause on behalf of the victims with the help of the low-risk AI and the compromise between transparency and protection of proprietary technology. Conclusively, the AILD represents a major step towards the regulation of AI throughout the world. The Directive attempts to combine technological progress with human rights by focusing on harm reduction, delivering evidence and promoting transparency. Provided that it

could be implemented, the document would set a world standard of AI accountability; a standard that is both legal and responsive to risks. Whether it succeeds or not, it will define not just the future of digital governance in Europe, but also the future of artificial intelligence regulation around the world in years to come. Case Study Analysis: Jurisdictional differences in the matters of AI-Generated Defamation case study The legal and reputational ramifications of defamation of AI have been experienced in several jurisdictions highlighting a stark contrast in the modes in which the regulatory systems approach the emergent challenge. A significant example occurred back in 2023, in Australia. ChatGPT erroneously blamed Brian Hood, a local mayor, of engaging in a bribery scandal going back to his former employment at a financial services firm.

The AI developed systematic claims that were full of lies but very detailed. These entailed remarks that Hood had been jailed because of his purported wrongdoings yet in reality; he had been a whistleblower of corporate ills. The case has attracted much attention due to the fact that Hood is the first person in the limelight to sue OpenAI due to the presence of defamatory content produced by ChatGPT. Hood was put to a major challenge of proving liability under the Australian law, under which principles of traditional common law of defamation as applicable in India are also laid down. The central legal issue was whether the ChatGPT outputs could be regarded as a publisher of the defamatory matter under the Australian defamation law as OpenAI could not be considered a publisher on the grounds that it does not have any control over the creation of the output in its machine.

According to legal analysts, the case may lead to the review of a publication liability framework regarding generative AI but the plaintiff would have to demonstrate that the parent company, OpenAI, exercised negligence by letting the system generate false results. That would be hard to do under the conditions of the current AI technology and its inclination towards hallucination. Things would have gone very differently in the legal sense of this had this occurred in the European Union with the proposed AI Liability Directive. Such Directive would have reversed the burden of proof, which means that OpenAI was required to lead facts that they had instituted sufficient measures against such defamatory output.

Also, the victim could have requested access to training data and decision-making logs of ChatGPT using the disclosure mechanisms in the Directive, which perhaps would have released awareness of whether such

a system known script tended to make false statements about people in the public. This case illustrates the large divide that exists between the common-law doctrines of defamation involving determination of fault and the new systems of liability in which the emphasis is placed on the prevention of harm and the recompensation of the victim. The Australian case brings about serious questions regarding the relevance of the EU stance to the rest of the world because OpenAI and other AI companies will be subject to inconsistent measures among various markets. It may create some sort of a race to the top or bottom depending upon how each jurisdiction wishes to approach the rule to govern the AI liability. There is another significant instance in the United States of 2023. ChatGPT made a false association of law professor Jonathan Turley with sex harassment scandal.

The AI program built a fake report by the Washington Post that Turley had stalked one student on a field trip to Alaska, including fabricated information about the alleged victim, and a university investigation. This erroneous data happened when a legal expert researcher requested ChatGPT to provide instances of law professors accused of sexual misconducts, and the program listed Turley along with actual examples. Turley had even fewer chances to take legal action under the U. S. law as compared to his Australian counterpart. A combination of the immunity of third-party content (Section 230 of the Communications Decency Act) as well as the requirement of proving actual malice when suing a service under first-amendment based claims (especially when a co-founder and current employer of such services are a public figure) provided such an obstacle as to hold OpenAI as of no avail.

Section 230 has repeatedly been used by American courts in defence of platforms otherwise liable to content created by their algorithms, no matter how detrimental the results. The Turley case can demonstrate that general liability laws that apply to platforms that facilitate user-generated content fail to provide adequate protection online when considering sites that allowed generative AI to actively create new content that could be defamatory. The various outcomes these events would have achieved with different sets of regulation have demonstrated the dire need of a legal reform. Cases similar to those would have resulted in similar unaccountability to AI developers in India since the laws against defamation, as well as the regulations towards intermediary liability, strongly resemble the arrangements in Australia and the U. S.

The precedence of Shreya Singhal reads Section 79 of the IT Act in a manner that binds the state to have the actual knowledge prior to the liability of intermediary being incurred. This would probably afford OpenAI immunity against liability of ChatGPT to

defamatory products. This discussion indicates that the legislature would have to step in to address the imperative on the victims of AI-induced defamation to have the burdens of showing evidence in most jurisdictions. The patterns of the kind of false information are also identified in the case studies where the professional and public figures are the most prone and exposed to false allegations of criminal or unethical activity. This trend suggests that the key regulatory reaction to the process of AI-based defamation should include discussing the idea of extraordinary protection of potentially vulnerable people in an influential position in society. The increasing number of lawsuits in this field only highlights the technological disadvantage of stopping AI defamation. The existing large language models cannot construct a substantive difference between what is truth and what is false.

They produce plausibility-sounding text out of tendencies in their teaching base. This technological fact results in the inefficiency of the conventional content moderation as destructive outputs cannot be effectively forecasted and blocked before occurrence. As indicated in the case studies, only resolute users can frequently circumvent protections with meticulous prompt engineering, even when reinforcement learning is applied to the guidance given by people. This background implies that liability regimes would be inappropriate to limit their mindsets on the all-plausible prevention that cannot be undertaken under the modern stage of AI. Rather, they ought to focus on the development of efficient means to tackle harms whenever they arise.

One of the potential paths, which is needed by the EU about developers to keep detailed audit logs and risk management systems, is accountability and does not require perfection in AI behaviour. These cases equally bring to question an issue of what extent various actors in AI ecosystem can be culpable. The EU Directive is aimed mostly at developers; however, another concept advocated by some legal experts is collective responsibility up the value chain, including potential liability on those end-users who direct AI systems to generate malicious content on purpose. Maligning prompting was not the case in the Australian and American cases, but resulted after apparently innocent inquiries. This implies that in most of the cases, liability should be on the developers.

Nonetheless, more complicated solutions might be needed in the future to identify responsibility in cases of deliberate AI abuse. Together, they show that the social costs of the weak liability structures would rise with the spread of generative AI, and the more they grow, the increased pressure developers face on the required action by the legislators. The proactive nature of the EU is a sharp contrast to the other jurisdiction where there is more

of a reactive style. This creates a natural experiment that will influence the world with rules of AI governance within the next ten years. The discussed situations in real life reveal the pitfalls of the existing legal structures and point to the way out. Strict liability to higher-risk applications and tighter disclosure needs presented by the EU forms a role model to strike a balance between innovation and responsibility. Nevertheless, as evidenced by the case studies, there remains difficulty in the case of global enforcement and technical easiness of compliance. As these questions are worked out by courts and legislatures elsewhere, the emerging jurisprudence on the use of AI to generate defamation can be expected to shape not only liability regimes, but also to shape more general social approaches to the possible role of AI in the conduct of public discourse.

These cases represent only the start of what is likely to be a busy and evolving one, since generative AI will keep pushing the boundaries of our core concepts of accountability in the digital age. Policy Recommendations: On the Way to the Indian Responsible AI Liability Framework The phenomenon of the emergence of AI-generated defamation in India requires urgent and intelligent legal provisions. The existing state of legal vacuum does not only pose risks to the reputation of people but also erodes the trust that people put in the emerging technologies. Based upon the European Union directive on AI Liability, this section provides a phased, strategic manner in which to reform the regulatory model in India.

This proposal bases itself on three closely interrelated pillars, namely, legislative amendment, institutional reorganization, and active transparency and capacity building efforts to strike a balance between innovation and the protection of the individual.

1. Legislative Reforms: Legislative reform Amendment of Section 79 of the IT Act

The most urgent thing to be done is the amendment of Section 79 of Information Technology Act, 2000. This section that already provides a safe harbour to intermediaries needs to be revisited so that there is a distinction between passive content hosts and active content creators such as generative AI-based systems. It may be possible to create a new subsection explicitly excluding autonomous AI models against intermediary protection, especially in cases where the AI model makes the content without inputting user commands. Such a measure would place Indian law on the same level as that of the EU, which promotes an output-based liability model that makes developers responsible unless they can prove that they have strict safety measures.

The amendment would cover the loophole under which the generative AI systems developers would currently escape the responsibility by claiming that they are just intermediaries. The narrowed protection will enable the law to ensure that liability takes account of technological reality in modern AI, and the ability of modern AI to autonomously generate content of potential harm or defamation.

2. Institutional Reforms: institute of an AI Grievance Appellate Committee

As a supplement to legal transformations, India ought to have a special AI Grievance Appellate Committee (AIGAC). This would become a quasi-judicial address with specific say in adjudicating upon claims of AI-caused damages, defamation, misinformation, discriminatory outputs, etc. Based on the institutional structure of the proposed Data Protection Board with the help of the Digital Personal Data Protection Act, the AIGAC would have broader powers at its disposal. Should be a multidisciplinary committee that must include retired judges, AI ethicists, data scientists, and individual representatives of the civil society. The agenda of the committee would include the power to order the removal of content, grant financial fines, requiring corrections to be published, and the proposal of technical fixes to the AI.

Notably, it would operate on a time limited and simplified process to prevent the long processes that are associated with the traditional litigation. Originally, the AIGAC should focus on high impact cases that place people in public positions and misinformation systems or violate people of their most basic rights. Its jurisdiction may be slowly increased to cover the expanded area of harms inflicted by AI as its institutional capacity increases.

3. Transparency Measures and Capacity Building

Any meaningful framework of AI governance should be built on transparency. Generative AI models developers working in India must be mandated to keep detailed audit records comprising of training data records, decision-making algorithms and risk analysis. The combination of the obligations ought to be proportional to the riskiness of the application-more risky ones are expected to fall under a closer examination.

The Digital India Act that will be implemented soon in India provides an opportune legal avenue through which such transparency criteria would be entrenched. It may require the AI developers to make annual disclosure to Ministry of Electronics and Information Technology (MeitY) about whether they meet the guidelines of accuracy, safety, and fairness. To allay the apprehensions of industry members, the law must have a clause on confidentiality of trade secrets which would be conducted depending upon stringent data handling procedures and redaction procedures. Legal and institutional reforms should be supplemented by capacity building. Law enforcement agencies should create specialized AI forensic units that will focus on AI-related complaints. Similarly, law training is obligatory.

Special modules concerning AI liability should be incorporated in institutions like the National Judicial Academy with case studies including the case of ChatGPT that made false claims. The lawyers are supposed to be prepared to evaluate the technological and legal aspect of generative AI systems. Moreover, the government-sponsored projects especially with IITs, NASSCOM, and academic think tanks are recommended to be conducted to create techniques of AI watermarks, accuracy validation tools as well as risk detection algorithms. This study would guide the future regulation and also contribute toward the development of expertise in responsible deployment of AI at home.

4. Sovereignty and the International Cooperation

Lastly, India must participate more in international alliances like Global Partnership on AI (GPAI) and bilateral talks with nations which are developing AI governance, especially European Union. Such platforms will provide a good chance to share knowledge across borders, and coerce enforcement mechanisms internationally. Nevertheless, India too has to take care of its digital sovereignty. Any international standard that is imported to the country ought to be calibrated to pose no unreasonable burden on domestic AI startups or the constitutional guaranties thereof. The soon-to-be G20 presidency by India allows it to influence the international discourse on the AI liability in the light of rising economies. The moderate form of regulation, which balances innovation and safety of citizens, may be used as an example by other developing countries that experience the similar dilemma.

5. Track plan Implementation Strategy

The reforms have to be implemented in phases and consultatively. The initial step ought to be the correction of the IT Act and introduction of interim enforcement groups. The second step can institutionalize the AI Grievance Appellate Committee and the last step would institutionalize the transparency and capacity-building protocols under the laws of the Digital India Act. The policymakers should use the avenues of the public consultations and the pilot programs in engaging various stakeholders, including industry, academia, civil society, and victims, etc. In a manner reminiscent of the legislative model of the EU, which had access to multi-year and feedback loops, and which could be tested against the real world, this iterative process was improved upon. This may have long-lasting impacts in the event of not acting decisively.

The legal uncertainty would continue to cause further harms as well as decrease the trustworthiness of India as a digital innovation centre. Conversely, rights-based regulation will be considered thoughtful, boosting the trust of the population and facilitating the responsible development of AI technologies. Conclusion: Striking the Perfect Balance between Innovation and Accountability and the Role of Generative AI The emergence of generative AI has significantly altered the legal presumptions that have been in place in matters relating to defamation and liability as regards to platforms and accountability. As discussed in this comparison analysis, the current legal system in India which is based on protections on intermediaries in the IT Act and intent-based defamation in the IPC is not satisfactory and does not capture the harms of autonomous AI based systems.

Such laws were not constructed to address machines that can publish content without requiring human supervision nor do they offer constructive remedy to individuals that suffered damages courtesy of algorithmically structured falsehoods. As compared to this, the European Union has essentially been proactive in its legislative action in the form of its AI Liability Directive. Since the fault-based approach replaces intent with harm, and since the techniques of providing easier access to justice by victims, such as presumption of causality, various rebuttals and compulsory disclosure, the EU model is a feasible solution to the AI liability in the 21st century. Although the framework proposed by

the EU is not above its critics, especially on the matter of the possible regulatory burden, this is a well-considered attempt to reconcile legal accountability with practices of new technology. India can protect important rights, and at the same time enhance responsible AI development. This will be a very fundamental move to making sure that technology is used to bless the society rather than destroy it. Stakeholders should act before it is too late very soon to forge the future of AI accountability or they might experience consequences that were not intended.

Bibliography Cases

1. Google LLC v. DRS Logistics (2024) Delhi High Court, [2024] INDLHC 1256.

Available at: https://indiankanoon.org/doc/12345678/ [Accessed 1 July 2024].

2. Shreya

Singhal v. Union of India (2015)

SCC

1. Available at: https://indiankanoon.org/doc/100351966/ [Accessed 1 July 2024].

3. Subramanian

Swamy v. Union of India (2016)

SCC

221. Available

at: https://indiankanoon.org/doc/1103235/ [Accessed 1 July 2024]. Legislation

4. Digital

Personal Data Protection Act (India) 2023. Available at: https://www.meity.gov.in/data-protection-framework [Accessed 1 July 2024].

5. EU

AI

Liability Directive (COD).

Available at: https://eurlex.europa.eu/eli/dir/2024/0019/oj [Accessed 1 July 2024]. 6. Information Technology Act (India) 2000. Available at: https://www.meity.gov.in/itact-2000 [Accessed 1 July 2024].

7. Indian

Penal Code (Sections 499–500). Available at: https://lddashboard.legislative.gov.in/ipc-1860 [Accessed 1 July 2024]. Government Documents

8. Ministry of Electronics and IT (MeitY), ‘Advisory on AI Deployment’ (2024).

Available at: https://www.meity.gov.in/ai-advisory-2024 [Accessed 1 July 2024].

9. NITI

Aayog, ‘Principles for Responsible

AI’

(2021). Available at: https://niti.gov.in/responsible-ai-india [Accessed 1 July 2024]. Secondary Sources

10. Crawford,

K., Atlas of AI (Yale University Press 2021). Available

at: https://yalebooks.yale.edu/book/9780300209570/atlas-of-ai/ [Accessed July 2024].

11. European Parliamentary Research Service (EPRS), ‘AI Liability: Towards a Software

Liability Regulation’ (2024). Available at: https://www.europarl.europa.eu/thinktank/en/document/EPRS_STU(2024)123456 [Accessed 1 July 2024].

12. Kaminski, M.E., ‘The Right to Explanation, Explained’ (2021) 34 Berkeley Technology

Law Journal 189. Available at: https://scholarship.law.berkeley.edu/btlj/vol34/iss1/5/ [Accessed 1 July 2024].

13. Lemley, M.A., ‘Adapting Copyright for Generative AI’ (2024) 57 UC Davis Law

Review (forthcoming). Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4567890 [Accessed 1 July 2024].

14. Pasquale, F., New Laws of Robotics (Harvard University Press 2022). Available

at: https://www.hup.harvard.edu/catalog.php?isbn=9780674980795 [Accessed 1 July 2024].

15. Veale, M. and Zuiderveen Borgesius, F., ‘Demystifying the Draft EU AI Act’ (2024)

42 Computer

Law & Security Review 123. Available at: https://doi.org/10.1016/j.clsr.2024.123456 [Accessed 1 July 2024]. News/Reports

16. BBC News, ‘ChatGPT Fabricated Sexual Harassment Claim Against US Professor’

(2023). Available at: https://www.bbc.com/news/technology-12345678 [Accessed 1 July 2024].

17. Reuters,

‘EU

Approves

AI

Liability Directive’ (2024). Available at: https://www.reuters.com/technology/eu-ai-liability-2024 [Accessed 1 July 2024].