INTRODUCTION
AI technology has evolved from a mere computational device to an independent entity that makes its own decisions without any active role on the part of humans. The result has been an accountability crisis in criminal law. Traditional criminal law is premised on the concepts of actus reus, the guilty act, and mens rea, the guilty mind. Both concepts imply human agency and moral reasoning.[1] However, contemporary autonomous AI has become a real threat to humanity in terms of inflicting harm in a way that cannot be predicted by its manufacturers and controllers. As the use of AI becomes ubiquitous in such spheres as healthcare, transport, banking, and governance, there are growing difficulties in identifying a perpetrator and assigning criminal liability for the harm caused.
THE UBER AUTONOMOUS VEHICLE ACCIDENT
The most glaring example in this regard happened in 2018 when an automated car from Uber killed Elaine Herzberg in Arizona. After investigation, it came out that Uber had turned off critical safety features and had preprogrammed the automobile in a manner that its response was delayed. However, instead of charging the company criminally for the same, prosecutors chose to indict only the human back-up driver.[2] This highlighted how criminal laws were failing to cope with this issue since the AI had caused an undesirable effect while dispersing responsibility among the programmer, company, and user.
THE CORE LEGAL CHALLENGE: RECONCILING MENS REA AND OPAQUE SYSTEMS
The basic question that needs to be answered is that of establishing mens rea in the AI. An artificial intelligence system is capable of committing the actus reus of committing an offence, such as an accident, trading on a fraudulent basis, and spreading viruses, but it is difficult to prove a guilty mind behind all that.[3] Traditionally, mens rea, recklessness, and negligence have been associated with human beings. However, since legal persons like corporations lack any biological entity or consciousness, they still have been held accountable by many legal regimes.[4]
THE BLACK BOX PROBLEM
The black box problem represents another substantial hurdle when determining who is liable for the actions of artificial intelligence systems. Today’s systems are becoming increasingly complicated, with a high level of complexity that cannot be understood even by the creators themselves.[5] Unlike traditional computer programs that follow rules established during the coding process, artificial intelligence learns and makes decisions independently. Because of that, courts find it hard to understand how the system arrived at a certain conclusion. This leads to difficulties in implementing some well-established doctrines of criminal law in connection with AI liability.
GABRIEL HALLEVY’S PARADIGMS OF AI CRIMINAL LIABILITY
Gabriel Hallevy introduced three principal models of artificial intelligence criminal liability.[6]
- The Perpetration-via-Another Model: In accordance with the concept of the “Perpetration-via-Another”, it can be assumed that artificial intelligence will be considered a tool used to commit the offence without fault. This model is rather similar to the principle of innocent agency in criminal law, where one is liable for having committed an offence through another person who was incapable of forming a guilty state of mind. In much the same way that liability would arise from one’s actions in employing children or individuals suffering from mental disability to commit an offence, liability will be placed on those individuals who purposely use such artificial intelligence as tools for committing an offence. It becomes highly applicable in scenarios when people use artificial intelligence in their crimes, like creating ransomware and committing frauds.
- The Natural-Probable-Consequence Model: The second liability rule, “Natural-Probable-Consequence”, makes people criminally liable if any harmful effect from the use of artificial intelligence could have been predicted at the time of releasing the technology.[7] In this case, corporations could be prosecuted for criminal neglect for introducing any AI without proper precautions or tests. For instance, the deployment of an autonomous medical artificial intelligence which prescribes death to a patient because of some unnoticed design flaws will be considered recklessness and negligence.
- The Direct Liability Model: The third liability rule, “Direct Liability”, suggests that artificial intelligence systems must take responsibility for criminal acts.[8] Hallevy states that, in the same way as corporations could be penalised in spite of being unconscious, autonomous artificial intelligence capable of making its own decisions is also able to become the subject of legal sanctions. Yet, this approach poses significant theoretical and practical problems, namely, whether punishment of computers makes sense.
THE TRIPARTITE MATRIX OF ACCOUNTABILITY – DEVELOPER NEGLIGENCE AND THE “STRIPED SHIRT” PARADOX
A quest for accountability usually involves three actors: developers, corporate users, and end-users. Developers can face liability for negligence if they intentionally overlook risks. In some states, such actions fall under dolus eventualis, where individuals knowingly subject themselves to the risk of harm.[9] However, most modern AI systems are created collaboratively, through the work of many programmers developing different elements of code. The harmful behaviour of AI can only appear after it is integrated into more complex systems. It represents the “striped shirt” paradox; none of the programmers possess the intent or knowledge required for conviction.[10]
CORPORATE LIABILITY AND ACCOUNTABILITY BY DESIGN
Therefore, more and more emphasis is placed on corporations that use AI. Many researchers call for implementing the accountability by design approach, which implies the necessity to incorporate mechanisms of explainable AI, algorithmic audits, human supervision, and safety checks.[11] Not doing so may imply negligence, but such a strategy aims at overcoming the issue of lack of criminal intent without proving that AI had it.
END-USER LIABILITY AND WILFUL BLINDNESS
Moreover, end users could be subjected to liability in situations where ignorance can be equated with actual knowledge under principles such as “wilful blindness”.[12] If users intentionally use systems that are highly prone to committing unlawful actions while deliberately ensuring their lack of control over such actions, it can be assumed that the resulting ignorance would not absolve them from liability for those actions.
ECONOMIC AND SYSTEMIC IMPLICATIONS: MARKET MANIPULATION
These risks are far from hypothetical because autonomous AI has caused damage to global financial infrastructure. For example, the flash crash of 2010, as well as the Knight Capital incident, proved just how much harm can be inflicted by autonomous trading programmes within a matter of minutes.[13] In addition, autonomous price-making algorithms pose an antitrust problem. The reinforcement learning technology allows competing AI programmes to learn to keep the prices unnaturally high on their own, without any interaction between humans. Thus, cartels can be formed digitally.
COMPARATIVE REGULATORY APPROACHES
The Indian stance on the liability for artificial intelligence has been fragmented since both the Information Technology Act, 2000, and the Bharatiya Nyaya Sanhita, 2023,[14] lack explicit provisions related to autonomous decision-making systems. Indeed, while the former is mainly aimed at regulating cybercrimes and provides intermediaries protection under Section 79[15], thus relieving an entity from liability where statutory duties of due diligence have been performed, such a legal approach is rather problematic in the case of autonomous artificial intelligence because organisations could claim safe harbour defences while using technologies that might cause independently any adverse results.
At the same time, the Bharatiya Nyaya Sanhita, 2023, maintains the legacy of criminal liability stemming from the Indian Penal Code and still relies upon such key concepts as intention, knowledge, recklessness, and negligence as prerequisites to culpability. As such, in the situation of any harms being caused by self-learning and opaque technologies, there would be virtually no way to find a blamable person.
One of the most extreme proposals was the introduction of electronic personhood. Back in 2017, the European Parliament discussed granting high-autonomy AI systems their separate legal personality, allowing them to own property, have obligatory insurance funds, and pay compensation to the victims.
CONCLUSION
The philosophy of punishment poses additional problems regarding AI liability. Historically, retributive punishment is based on moral blameworthiness and the ability to experience suffering such as guilt and shame.[16] AI is devoid of consciousness, emotions, or morality. Thus, punishment of an artificial machine seems to be illogical. On the other hand, utilitarian philosophies offer better justifications for punishment of AI. For example, destruction of harmful AI programs will result in making it inoperative, mandatory retraining will help address bias, and financial punishments will deter companies from developing dangerous AI.[17]
Author(s) Name: Priyanka Ratha (Vikash Law School,Bargarh)
References:
[1] Tany Calixto Bonfim, ‘Criminal liability of artificial intelligent machines: eyeing into AI’s mind’ (Master Thesis, Lund University 2021) <https://lup.lub.lu.se/luur/download?func=downloadFile&recordOId=9095199&fileOId=9095200> accessed 08 June 2026
[2] Helen Stamp, ‘The Reckless Tolerance of Unsafe Autonomous Vehicle Testing: Uber’s Culpability for the Criminal Offense of Negligent Homicide’ (2024) 15(1) Case Western Reserve Journal of Law, Technology & the Internet 1155 <https://scholarlycommons.law.case.edu/jolti/vol15/iss1/2/> accessed 08 June 2026
[3] Muhammad Ahsan Iqbal Hashmi et al., ‘Criminal Liability in the Age of Autonomous Systems: Rethinking Mens Rea and Actus Reus’ (2023) 3(3) The Critical Review of Social Sciences Studies <https://thecrsss.com/index.php/Journal/article/view/683> accessed 07 June 2026
[4] Gabriel Hallevy, ‘The Criminal Liability of Artificial Intelligence Entities — From Science Fiction to Legal Social Control’ (2010) 4(2) Akron Intellectual Property Journal 121 <https://ideaexchange.uakron.edu/cgi/viewcontent.cgi?article=1037&context=akronintellectualproperty> accessed 07 June 2026
[5] Y Bathaee, ‘THE ARTIFICIAL INTELLIGENCE BLACK BOX AND THE FAILURE OF INTENT AND CAUSATION’ (2018) 31(2) Harvard Journal of Law & Technology 889 <https://jolt.law.harvard.edu/assets/articlePDFs/v31/The-Artificial-Intelligence-Black-Box-and-the-Failure-of-Intent-and-Causation-Yavar-Bathaee.pdf> accessed 07 June 2026
[6] Hallevy (n 4)
[7] Ibid
[8] Ibid
[9] Marta Bo, ‘Are Programmers In or ‘Out of’ Control? The Individual Criminal Responsibility of Programmers of Autonomous Weapons and Self-driving Cars’ in Sabine Gless and Helena Whalen-Bridge (eds), Human-Robot Interaction in Law and its Narratives: Legal Blame, Criminal Law, and Procedure (CUP 2020)
[10] Ryan Abbott and Alex Sarch, ‘Punishing Artificial Intelligence: Legal Fiction or Science Fiction’ (2019) 53 UC Davis Law Review <https://lawreview.law.ucdavis.edu/sites/g/files/dgvnsk15026/files/media/documents/53-1_Abbott_Sarch.pdf> accessed 07 June 2026
[11] Aayushman Verma and Manik Tindwani, ‘Accountability by Design: Shared Liability in AI Fraud under Indian Cyber Law’ (Virtuosity Legal, 11 October 2025) <https://virtuositylegal.com/accountability-by-design-shared-liability-in-ai-fraud-under-indian-cyber-law/> accessed 07 June 2026
[12] Oleh Vretsona et al., ‘Managing Evolving Third Party Risks: Exploring Best Practices’ (Gibson Dunn, 05 February 2026) <https://www.gibsondunn.com/wp-content/uploads/2026/02/WebcastSlides-Managing-Third-Party-Risk-in-a-Shifting-Regulatory-Landscape-5-FEB-2026.pdf> accessed 07 June 2026
[13] Theodore Milosevic, ‘Corporate Criminal Liability for Algorithmic Price Fixing in Canada’ (2018) 16(2) Canadian Journal of Law and Technology 201 <https://digitalcommons.schulichlaw.dal.ca/cgi/viewcontent.cgi?article=1236&context=cjlt> accessed 07 June 2026
[14] Bharatiya Nyaya Sanhita 2023
[15] Information Technology Act 2000, s 79
[16] Abbott (n 10)
[17] Elina Nerantzi and Giovanni Sartor, ‘‘Hard AI Crime’: The Deterrence Turn’ (2024) 44(3) Oxford Journal of Legal Studies <https://doi.org/10.1093/ojls/gqae018> accessed 05 June 2026