AI-Generated Crimes - A Critical Analysis Of Law In India
Author: LawSphereX
Category: Criminal Law
Published:
Updated:
AI-GENERATED CRIMES: A CRITICAL ANALYSIS OF LAW
IN INDIA
Dr. Sufiya Ahmed, Department of Law, Babasaheb Bhimrao Ambedkar University, Lucknow Sunidhi Singh, Department of Law, Babasaheb Bhimrao Ambedkar University, Lucknow
ABSTRACT
The artificial intelligence is rapidly transforming various sectors including the field of law. As with the rapid growth of artificial intelligence in various fields raises a query regarding the lawful implications of artificial intelligence and resulting into the crimes committed through it. The emergence of technology has also led to the increase in malpractices of artificial intelligence however raising a test for the government and judiciary to combat these challenges. Therefore, there is a need to revise the old criminal laws and addition of definition of new crimes in the IT Act of 2000 with the appropriate relief. Consequently, legislative steps are required to be taken by the government in order to contest with unlawful activities committed through artificial intelligence.
The paper explores the meaning of artificial intelligence and also provide with different AI crimes such as face swapping, deepfakes, phishing attacks and AI-powered cyber-attacks etc. It suggest upon the requirement of AI personhood, the obligation of AI creators. This paper also deals with the ethical concerns and also highlights the need for establishment of statutory body at national and international domains to combat AI -crimes. Keywords: Artificial intelligence, cyber-crimes, Machine learning, legal challenges, deep fake, cyber security.
Introduction Artificial intelligence can be described as a machine that involves the functions of thinking and learning like a human but doesn't have a natural body, unlike humans. This machine can perform several intelligent tasks, and decision-making and assist people in various tasks. The machine is like a supercomputer program, however, it is not made up of natural body cells but can perform intellectual functions of thinking and learning easily from the information as human beings do. Computer programs utilize various data and calculations to make decisions. Artificial intelligence acquires information like human beings learns from the available books, experiences, and videos. Artificial intelligence is a complex field that attempts to match with the human intelligence. One of the critical facets of human intelligence is thinking.
1 Artificial intelligence can be commonly defined as a technology that permits machines to emulate several multifaceted skills, unlike human beings. 2 . Explaining artificial intelligence is not an easy task as there is no general explanation of artificial intelligence. 3 Artificial intelligence (AI) and machine learning (ML) technologies have advanced significantly in competence, accessibility, and mainstream adoption, with no signs of slowing down. Although AI technology is commonly advertised as 'personal assistants' like Google Home, Alexa, Apple Siri, Amazon, and Alexa, learning-based methods are routinely used for similar purposes. AI has a wide range of applications, including route-finding, language translation, biometric identification, political campaigning, industrial process management, and food supply logistics (Benaich and Hogarth, 2019).
4 AI has greatly benefited humanity over the past decade and will continue to do so as it becomes more integrated into digital services. Governments worldwide are exploring the use of AI systems and apps to enhance their operations, particularly in detecting and predicting crime. National security and intelligence services recognize the capabilities of AI technologies to achieve the aim of national and public welfare objectives. 5 AI technologies, such as facial recognition in criminal justice, drones, selfsufficient weapons, and self-driving vehicles, can be misused and harm individuals' rights and
1 Saptarsi Goswami, Amit Kumar Das, et.al., AI for Everyone 8-9 (Pearson, Noida, 2024).
2 Sheikh, H., Prins, C., & Schrijvers, E., "Artificial Intelligence: Definition and Background" in *Mission AI. Research for Policy* (Springer, Cham, 2023), https://doi.org/10.1007/978-3-031-21448-6_2.**
3 Russell, S. J., & Norvig, P., Artificial Intelligence: A Modern Approach, 4th ed. (Pearson, 2020).
4 Caldwell, M., Andrews, J.T.A., Tanay, T., et al. "AI-enabled Future Crime." Crime Science 9, no. 14 (2020). https://doi.org/10.1186/s40163-020-00123-8.
5 Velasco, C., "Cybercrime and Artificial Intelligence: An overview of the work of international
organizations on criminal justice and the international applicable instruments," ERA Forum 23 (2022): 109–126. https://doi.org/10.1007/s12027-022-00702-z.
freedoms if not properly managed. AI and machine learning can learn and answer to cyberattacks which may result into targets at critical infrastructure sectors like hydropower, energy, and power supply. They can also help manage cybersecurity solutions to reduce security risks. Despite limited expenditures, small and medium firms nevertheless face difficult cybersecurity concerns. The COVID-19 pandemic restricted connectivity for a significant portion of the global population. AI-based systems, technologies, and applications have become rapidly important for inaccessible work, M-learning, E-payments, and entertainment (e. g. , OTT and video on-demand services). Organized criminal gangs reorganized their activities to target specific stakeholders, such as international organizations, research and health sector entities, supply chain corporations, and individuals.
6 Different forms of AI-generated crimes Phishing Attacks Phishing attacks are among the most prevalent cybersecurity threats. These attacks utilize social engineering techniques to send deceptive communications with the aim of tricking victims into disclosing their credentials. Often, phishing is associated with other malicious activities, such as deploying ransomware or infecting a computer system. Such attacks are highly common, and individuals are frequently advised to receive comprehensive training on recognizing and mitigating phishing threats. Phishing can be executed through various methods. Attackers may conduct mass phishing campaigns that target large groups of people, thereby increasing the likelihood of reaching vulnerable individuals.
To effectively prevent and respond to phishing attacks, it is crucial to remain vigilant and well-prepared, incorporating robust identification and avoidance strategies as essential components of cybersecurity practices. 7 According to Cisco (2022), phishing attacks involve sending deceptive messages that appear to originate from a legitimate source, typically via email. The primary objective of such attacks is to obtain sensitive information, such as credentials and credit card details. Additionally,
6 Velasco, C., "Cybercrime and Artificial Intelligence: An overview of the work of international
organizations on criminal justice and the international applicable instruments," ERA Forum 23 (2022): 109–126. https://doi.org/10.1007/s12027-022-00702-z 7 Ansari, Meraj Farheen, Sharma, Pawan Kumar & Dash, Bibhu, "Prevention of Phishing Attacks Using AI- Based Cybersecurity Awareness Training", (2022) 3 IJSSAN 61-72, doi: 10.47893/IJSSAN.2022.1221.
phishing attacks may include the deployment of malware on victims' devices, with the attackers subsequently demanding a ransom for the removal of the infection. An illustrative example of a phishing attack involves a user being misled into clicking a fraudulent URL contained in an email. The malicious website then prompts the user to enter login credentials, which the user unwittingly provides, believing they are interacting with a legitimate and trustworthy site. Phishing attacks, which encompass various methods such as email, spear, voice, and social media phishing, remain among the most prevalent tactics used by hackers and scammers. Recent research by Verizon indicates that phishing accounts for 36% of all breach cases, reflecting its widespread use and growing sophistication.
The increasing reliance on the internet has heightened vulnerability to such attacks, making phishing a persistent threat in contemporary society. Moreover, advancements in AI and machine learning have further augmented the capabilities of phishing attackers, enabling them to execute more sophisticated and targeted attacks. 8 AI-Powered Cyberattacks: Cyberattacks have become a significant global risk, posing severe financial threats to individuals and organizations. According to Dixon and Eagan (2019), these attacks can lead to catastrophic losses, with data breaches in the United States averaging $8. 19 million per incident and global economic impacts reaching approximately $400 billion annually (Fischer 2016; Kirat, Jang, and Stoecklin 2018). The rise in sophisticated cyberattack techniques, including the use of Artificial Intelligence (AI), has intensified the threat landscape.
AI-driven cyberattacks utilize advanced algorithms to outpace traditional detection methods, enabling cybercriminals to execute attacks with unprecedented speed and scale. These attacks can disrupt operations, compromise critical data, and damage reputations, presenting challenges that exceed the capabilities of conventional cybersecurity measures (Chakkaravarthy et al. , 2018; DarkTrace, 2021). The growing sophistication of AI-driven attacks necessitates innovative defensive strategies to protect against increasingly autonomous and adaptable threats (Thanh and Zelinka 2019; Hamadah & Aqel, 2020). 9
8 ibid
9 Blessing, Guembe, Azeta, Ambrose, Misra, Sanjay, Osamor, Victor, Fernandez-Sanz, Luis & Pospelova, Vera, "The Emerging Threat of AI-driven Cyber Attacks: A Review", (2022) 36 Applied Artificial Intelligence 1-34, doi: 10.1080/08839514.2022.2037254.
Voice Cloning with AI Technology Voice cloning is a kind of deep learning technique used to synthesize a voice that closely mimics the original. Applications such as iSpeech, LyreBird, and CereVoice Me provide access to the public for various deepfake technologies. These programs require only a few minutes of audio recordings to generate a similar voice, capable of reading any text aloud. According to Kinnunen et al. (2017), AI-generated synthetic voices can sound identical to the unique audio samples. Training the system necessitates a collection of matching audio recordings and corresponding texts. Brief samples of spoken conversation can produce sounds comparable to the human voice. Major internet companies such as Google and Amazon are investing significantly in the development of this technology, even though it remains in its early stages.
Neural network-generated synthetic speech has numerous applications, including voice greetings, audiobooks, and voice cloning for parents. This technology is also used for reading fairytales, audio and video training sessions, promotional videos, audio advertisements, voice bots, and personalized voice assistants. Law enforcement officers may also utilize this technology for impersonation to avoid recognition. However, AI technology can be exploited for criminal purposes such as fraud, prank calls, and falsification. Consequently, it is essential to develop tools that prevent the unauthorized use of voice cloning technology. There is a need to identify gaps in the existing legislation of the Russian Federation to address the risks associated with the development and dissemination of illegal and fraudulent audio in the media sphere.
10 AI -Genocide The research devolves into the profound implications of generative AI on the memorialization of mass atrocities, highlighting its dual potential to enhance remembrance and pose significant risks, such as the distortion of historical facts and the proliferation of denialism. It underscores the transformative capacity of generative AI to create new narratives for commemorating mass atrocities by processing vast amounts of data rapidly, thereby enabling innovative and creative approaches to memory practices. This technological advancement can democratize memorialization by amplifying the voices of underrepresented communities and making historical events more accessible. However, it also raises critical ethical concerns. The non-
10 Galyashina, Elena and Nikishin, Vladimir, "AI-Generated Fake Audio as a New Threat to Information
Security: Legal and Forensic Aspects" (2021) 17-21 https://doi.org/10.5220/0010616700003170
transparency of AI-generated content sources can lead to misrepresentations and the creation of false evidence. The potential for AI to trivialize or distort past suffering is significant, especially given the ease with which digital content can be manipulated. It emphasizes the importance of distinguishing between human and AI-generated content to preserve the integrity of historical facts. The integration of generative AI into memorialization practices presents challenges in how societies perceive and interact with digital content. The caution against the possibility of AI-generated content influencing public beliefs and attitudes, potentially undermining trust in digital media and memorialization efforts. They call for careful consideration and regulation of AI technologies in this sensitive field to ensure ethical and sustainable practices.
At last, it concludes by advocating for a balanced approach to incorporating generative AI in memorialization, recognizing its potential to enhance memory practices while stressing the need for stringent ethical guidelines and transparency to mitigate the risks associated with AI-generated content. The ultimate goal is to harness AI's capabilities while safeguarding the dignity of victims and the accuracy of historical narratives. 11 Deepfakes A deepfake is a digitally forged creation made using deep learning, a branch of artificial
intelligence. Deepfakes can produce completely new content or modify remaining content, including videos, images, audio, and text. These digital forgeries can be used to defame individuals, impersonate or blackmail public figures, and support cybercrime activities. The initial public examples of deepfakes emerged in November 2017, when operators on the popular online forum Reddit employed AI-based ‘face swap’ tools to portray celebrities' faces in pornographic videos. Since then, deepfake technology has become widely accessible and user-friendly. Free software and popular smartphone applications like Face Swap also enable general users to create and disseminate deepfake content. Additionally, low-cost services such as Lyrebird’s voice generation offer subscription packages for their tools. Consequently, deepfake technology has been democratized.
The cost and accessibility of deepfake software are expected to decrease further due to advancements in computing power and AI techniques that streamline the training of deepfake 11 Makhortykh M. , Zucker E. M. , Simon D. J. , et al. , "Shall Androids Dream of Genocides? How Generative AI Can Change the Future of Memorialization of Mass Atrocities," Discov Artif Intell 3 (2023): 28, https://doi. org/10. 1007/s44163-023-00072-6.
algorithms. Generative Adversarial Networks (GANs) exemplify such advancements by reducing and automating the AI training process. In GANs, two neural networks compete: the 'generator' network creates fake content, while the 'discriminator' network evaluates the authenticity of the content. This competitive process continues over thousands or millions of cycles until the generated outputs are indistinguishable from genuine ones. GAN models are now widely accessible, with many available for free online. 12 There are following types of deepfakes which are as follows: - • Face swapping Users can superpose the face of a target onto another body, a technique applicable to both still images and video. Basic versions of this technology are readily accessible online through specialized applications.
13 Facial swapping generates images by superimposing an individual's facial identity and attributes (such as pose, expression, lighting, and background) onto another body. This process typically involves extensive preprocessing and postprocessing to achieve high perceptual quality. Algorithm developers employ two primary approaches: the targetbased method, which involves extracting and inserting the person's identity from the source image into the target image, and the source-based method, which edits the source image based on attributes from the target image. The latter approach lacks control over the environmental context, leading to a focus on target-based algorithms for broader applicability. Advances in algorithmic techniques are enhancing quality and efficiency, though the best results often come from models trained on specific examples rather than generalized data.
This discussion will emphasize identity conversion pipelines and the influence of preprocessing and postprocessing tools on their outcomes. 14 The term preprocessing includes Effective data processing is crucial for both training and application of facial swapping technology, particularly in locating human faces within images. Face swap software, which operates on a one-to-one basis with target and source individuals, requires extensive data such as photographs or videos. The software processes this data through 12 Smith Hannah and Mansted Katherine, "What's a Deep Fake? " Weaponized Deep Fakes: National Security and Democracy, Australian Strategic Policy Institute (2020): 05-10, http://www. jstor. org/stable/resrep25129. 6 accessed 5 August 2024.
13 Ibid
14 Walczyna T. and Piotrowski Z., "Quick Overview of Face Swap Deep Fakes," Appl. Sci. 13 (2023): 6711, https://doi.org/10.3390/app13116711. Academic Editors: Junhui Huang, Qi Xue, and Zhao Wang.
detection, alignment, and mask generation. Face detection identifies face-like regions, alignment locates key facial landmarks and mask generation isolates the face from the background. Users can select from various detection models, including CV2 DNN, MTCNNs, and S3FD, the latter being noted for its superior ability to detect faces of varying sizes and minimize false positives through advanced scaling and anchor-matching strategies. The process outlined emphasizes the complexity and resource demands of facial swapping technologies. While the S3FD detector’s advanced capabilities in handling diverse face sizes and reducing false positives are notable, its high resource requirements may limit accessibility and practical implementation for many users.
Additionally, the lack of transparency regarding the datasets used by Faceswap raises significant concerns about data privacy and ethical usage. The reliance on publicly available training sets without detailed descriptions poses risks related to data security and the potential for misuse. To address these issues, greater transparency in dataset usage and stricter ethical guidelines are essential for mitigating risks associated with facial swapping technology. 15 • Re-enactment The face from a target source is superimposed onto a user, enabling the manipulator to control and modify the target's facial movements and expressions. 16 Generation Puppet-master, called as facial reenactment, is a type of deepfake technology that manipulates facial expressions by transferring gestures, eye movements, and head movements from a source actor to a target individual in a video.
This technique has diverse applications, including altering expressions in multilingual video conferences, dubbing in film production, and creating photorealistic animations for entertainment. Initial methods relied on 3D facial modeling for accurate geometry and photorealism, with advancements such as Face2Face enabling real-time expression transfer using RGB video streams. Recent developments include the use of GANs for high-resolution, photorealistic results, with models like Pix2pixHD and approaches such as act GAN and Ganimation enhancing facial reenactment capabilities. Additionally, few-shot and one-shot methods, like X2face and Zakharov et al. 's meta-transfer learning, allow reenactment using minimal source images. Real-time solutions, including FSGAN, provide facial replacement and reenactment with occlusion handling at 30fps. Future advancements may
15 Ibid
16 Smith Hannah and Mansted Katherine, "What's a Deep Fake?" Weaponized Deep Fakes: National Security and Democracy, Australian Strategic Policy Institute (2020): 05-10, http://www.jstor.org/stable/resrep25129.6 accessed 5 August 2024.
extend to full-body reenactment and synthesized audio, presenting significant challenges in distinguishing real from fabricated content. The progress in facial reenactment technology highlights impressive advancements in photorealism and real-time processing. However, these capabilities also raise critical concerns. The ability to convincingly manipulate facial expressions and integrate synthetic audio could exacerbate issues related to misinformation and digital deception. The reliance on extensive training data and the need for real-time processing can limit the accessibility and ethical use of these technologies. Additionally, while few-shot learning approaches represent a significant advancement, they may struggle with preserving identity and handling large pose differences.
As these technologies continue to evolve, ensuring robust safeguards and ethical standards will be crucial to mitigate potential misuse and maintain public trust. 17 • Lip syncing Users can transfer mouth movements onto a target video, and when combined with audio generation, this technique can create the illusion that the target is articulating false content. 18 The significant advancements in deepfake technology, emphasize the evolution from basic face-swapping techniques to sophisticated implementations achieved by institutions like DeepMind. Initially, deepfake creation demanded extensive computational resources and expertise, but advancements in consumer-level software, such as Face App, have democratized access, making the technology more user-friendly and accessible.
The technical achievement of accurately lip-syncing dialogue to a target's facial movements involves training neural networks on extensive datasets to map and replicate facial expressions. Although early deepfakes exhibited visible artifacts and lacked realism, improvements in deep-learning algorithms have enhanced the fidelity of lip-syncing, effectively bridging the "uncanny valley" that previously characterized CGI and motion-capture techniques. The transition from high-resource requirements to consumer-level applications is crucial, as it enables individuals with limited technical expertise to produce convincing deepfakes. Tools like Fake App have expanded the scope of potential uses and misuses, blurring the line between 17 Masood M. , Nawaz M. , Malik K. M. , et al. , "Deepfakes Generation and Detection: State-of-the-Art, Open Challenges, Countermeasures, and Way Forward," Appl.
Intell. 53 (2023): 3974–4026, 18 Smith Hannah and Mansted Katherine, "What's a Deep Fake? " Weaponized Deep Fakes: National Security and Democracy, Australian Strategic Policy Institute (2020): 05-10, http://www. jstor. org/stable/resrep25129. 6 accessed 5 August 2024
genuine and fabricated content and raising concerns about the authenticity of digital media. The ethical implications, particularly in the context of non-consensual pornography, are significant. The use of an individual's likeness without consent, especially in explicit content, presents severe privacy and rights issues. The argument that a public figure's prior consent to appear on screen justifies their use in deepfake pornography disregards personal agency and the right to control one's image. The widespread distribution of such content underscores the need for robust legal and ethical safeguards. Regulatory responses, such as Reddit's ban on involuntary pornography, highlight both efforts to address the issue and the limitations of existing legal frameworks.
The rapid pace of technological advancement outstrips current regulations, necessitating comprehensive and adaptive approaches to mitigate misuse while balancing innovation and privacy rights. Future developments in deepfake technology raise additional concerns. As the technology evolves, distinguishing between authentic and fabricated content may become increasingly challenging, with profound implications for misinformation, digital security, and public trust. Moreover, the potential use of advanced deepfake technology by governments, corporations, and other institutions introduces risks of propaganda and cyber warfare, presenting significant ethical and security challenges.
Addressing these issues requires ongoing vigilance, the adaptation of legal frameworks, and heightened public awareness to mitigate the potential harms associated with increasingly realistic and accessible deepfake content. 19 The Information Technology Act and Deepfake laws: - India currently lacks specific legislation that directly addresses the challenges posed by deepfake technology. However, existing legal frameworks offer some recourse in cases involving deepfakes, drawing upon various acts such as the Indian Penal Code (IPC), the Information Technology Act, of 2000 (IT Act), and the Copyright Act, of 1957. In cases of deception and identity theft, Section 66D of the IT Act, 2000, penalizes the use of communication devices or computer resources with the intent to deceive, with offenders facing imprisonment for up to three years and fines.
Additionally, Sections 66, 66-C of the IT Act and 19 Fletcher John, "Deepfakes, Artificial Intelligence, and Some Kind of Dystopia: The New Faces of Online Post-Fact Performance," Theatre J. 70, no. 4 (2018): 455-71, https://www. jstor. org/stable/48586871 accessed 5 August 2024.
Sections 420, 468 of the IPC, 1860, can be invoked depending on the nature of the deception and the harm caused. To combat misinformation and cyberterrorism, Section 66-F of the IT Act, 2000, along with the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2022, provide mechanisms to curb the dissemination of false information, particularly when it threatens the government. Furthermore, Sections 121-A and 124-A of the IPC, 1860, address offenses impacting national sovereignty and integrity, including sedition. The legal framework also offers protection against hate speech and defamation through Sections 153-A, 153-B, and 499 of the IPC, 1860, which allow for legal action against the spread of hateful or defamatory content via deepfakes.
In the context of elections and privacy, the Representation of the People Act, of 1951, contains provisions to tackle electoral malpractices, which could include the use of deepfakes to manipulate public opinion. Additionally, Sections 66-E, 67, 67-A, and 67-B of the IT Act, 2000, alongside Sections 292 and 294 of the IPC, 1860, address privacy violations and the dissemination of obscene content, potential uses for deepfakes. Intellectual property rights are safeguarded under the Copyright Act, of 1957 (Section 51), which protects against the unauthorized use of copyrighted material in deepfakes. Looking forward, the upcoming Data Protection Bill, 2021, is anticipated to further enhance protections against cybercrimes, including those involving deepfakes. Despite these provisions, there are notable limitations in India's legal approach to deepfakes.
The absence of specific legislation tailored to the unique challenges posed by this technology leaves gaps in the legal framework. The current reactive approach, relying on the application of existing laws, may not fully cover all aspects of deepfakes, especially considering the rapid pace of technological advancement. This situation places a significant burden on law enforcement agencies to detect and prove the authenticity of content, and on the judiciary to assess such evidence. Furthermore, there is a lack of regulatory clarity, with no clear guidelines or standards for online platforms and content creators regarding the creation and dissemination of deepfakes. 20
20 Kolakkal, Aldrin. Deep Fake Technology: Analysis of Legal Framework and the Way Forward (2024)
10.13140/RG.2.2.11565.09441.
Conclusion and Recommendations. The connection of artificial intelligence and human beings will be the key reason for technological crimes in the future, however with the increasing dynamism in the field of technology also stresses the need for a comprehensive legal regulatory framework to govern them in India which also raises a demand for cautious considerations from policymakers, technologists, etc. Due to the advancement in Technology, it also poses a responsibility to curb the challenges. There are certain recommendations to deal with crimes done through artificial intelligence which are as follows: - • There is a need for legal personhood which means there is an urgent requirement for the formation of laws that deal with the accountability of AI creators, operators, etc.
• There is a huge challenge in assigning a criminal liability of act done through AI, this liability may vary in terms of ethics, legal, and technological considerations. • There is a further prerequisite of fairness, maintaining the transparency and accountability of AI developers and also deployment is the crucial need for curbing the intentional misuse of technology. • Adapting with technological advancement, also raises a mandate for making legal framework which deals with AI crimes and also at the same time raising a necessity of keeping a check on relevancy and effectiveness of laws in addressing the issues and complexities of these crimes. • The legal system has to be quite alert to the speedy technological advancement as the complexity of AI crimes may lead to complications for prosecution in national and international domains.
• Strengthening of the legal frameworks such as inclusion of AI crimes in the IT Act and also updating the current IT act with the definition of AI crimes which may include defining of crimes such as deepfakes, AI-frauds, inclusion of penalties and fines. • There is a need to establish statutory bodies to specially combat the crimes committed through artificial intelligence. Although the government must suggest the criteria of
adequate knowledge of technological advancement in order to outperform the certain tasks like detection, investigation of AI crimes. • The government of India should set the legislative agenda in order to curb the malpractices of AI crimes but along with it establish educational and also building up of centres to track as well as monitor the movement of offenses. • The government of India should also collaborate at the international level to establish a council or organization for assistance in detecting AI illegal activities and also provide general public with AI literacy.