ARTIFICIAL INTELLIGENCE AND LEGAL LIABILITY: WHO IS RESPONSIBLE FOR AI DECISIONS?

I. Introduction

“Artificial Intelligence can draft a legal brief in seconds, but it cannot comprehend a client’s hardships, emotions, or the human context behind a case,” observed the Chief Justice of India,¹ cautioning that AI should remain a secondary assistant rather than a substitute for a lawyer’s judgment. His statement underscores a pressing legal dilemma: when AI systems influence or even determine outcomes, who bears responsibility for those decisions—the developer, the user, or the machine itself?

This question is no longer theoretical. As AI tools increasingly assist in legal research, contract drafting, and even predictive analytics for litigation, the boundaries of liability blur. The challenge lies in reconciling technological efficiency with the irreplaceable human role in justice, raising urgent debates about accountability, ethics, and the future of legal practice.

Background and Context

The integration of Artificial Intelligence into judicial processes has sparked both optimism and caution. In 2026, the Supreme Court’s Artificial Intelligence Committee released Draft Regulations on the use of AI in courts,² inviting feedback from stakeholders. In response, the Supreme Court Advocates-on-Record Association (SCAORA) conducted a detailed clause-by-clause review and submitted its recommendations to Justice P.S. Narasimha, who chairs the Committee.³

SCAORA’s report emphasizes a precautionary and phased deployment of AI, warning against premature reliance on systems whose internal reasoning remains opaque. It highlights the “black box paradox”: while judges and court officers are held accountable for AI-assisted decisions, the underlying logic of generative AI models cannot be meaningfully explained or verified. This contradiction raises serious concerns about transparency, accountability, and constitutional validity.

The Association also flagged risks such as AI hallucinations, citing the Supreme Court’s recent “zero-tolerance” stance against fabricated precedents,⁴ and automation bias, where overburdened judges or staff may defer excessively to machine outputs. It recommended mandatory human oversight—both “in-the-loop” and “on-the-loop”—to ensure that AI-generated drafts or administrative decisions do not undermine litigants’ rights.

Another critical issue raised was data sovereignty. SCAORA cautioned against judicial data being processed through infrastructure controlled by foreign technology companies, urging that court data remain within sovereign systems. It also called for audits of existing AI tools already in use, such as SUPACE, SUVAS, and AI-assisted e-filing platforms.

Finally, the report questioned the constitutional foundation of the Draft Regulations, noting ambiguities about whether they derive authority from Article 145 or Article 142,⁵ and stressed the need for greater representation of the Bar in AI governance structures.

In sum, the background reveals a legal landscape where the promise of AI efficiency collides with unresolved questions of accountability, transparency, and constitutional authority. These tensions make the regulation of AI in courts not only a technological issue but a fundamental matter of justice and judicial independence.

Thesis Statement

This article argues that the current legal framework is inadequate to address accountability for decisions made by artificial intelligence systems in judicial and quasi-judicial contexts. While existing doctrines place responsibility on human actors such as judges, lawyers, or developers, they fail to resolve the “black box paradox” where AI outputs cannot be meaningfully explained or verified. It is submitted that a coherent liability regime must combine precautionary restrictions on opaque AI systems, mandatory human oversight, and robust data sovereignty safeguards to ensure that technological efficiency does not undermine judicial independence or litigants’ rights.

The article proceeds as follows: Section II outlines the existing legal and regulatory framework governing AI in judicial contexts; Section III examines case law responses to AI-related risks; Section IV undertakes a critical evaluation of accountability gaps, automation bias, and data governance; Section V examines comparative perspectives from other jurisdictions; and Section VI proposes reforms to establish a balanced liability regime that safeguards judicial independence while enabling responsible innovation.

II. Legal Framework

The regulation of artificial intelligence in judicial contexts currently rests on a fragmented set of doctrines and institutional measures. At present, liability for AI-assisted decisions is indirectly addressed through existing principles of judicial accountability, professional responsibility, and constitutional safeguards, rather than through a dedicated statutory regime.

Under constitutional law, Articles 142 and 145 of the Indian Constitution empower the Supreme Court to frame rules governing its practice and procedure.⁶ However, these provisions do not explicitly authorize binding regulations on High Courts, raising questions about the constitutional basis of the Draft AI Regulations, 2026. This ambiguity underscores the need for a clear legislative mandate before AI can be systematically integrated into judicial functions.

From the perspective of professional responsibility, lawyers remain bound by ethical duties to exercise independent judgment and safeguard client interests. The Chief Justice of India has emphasized that AI should serve only as a secondary assistant, not a substitute for human reasoning.⁷ This position reflects the broader principle that litigants engage advocates for their professional judgment, not for machine-generated outputs.

In terms of judicial accountability, the Draft AI Regulations place ultimate responsibility on judges and court officers for any AI-assisted decision.⁸ Yet this framework collides with the “black box paradox”: modern generative AI systems operate through opaque neural networks whose reasoning cannot be meaningfully explained or verified. Holding human actors accountable for outputs they cannot fully understand creates a structural gap in liability.

Finally, data governance forms a critical component of the legal framework. Judicial data processed through AI platforms controlled by foreign technology companies raises sovereignty concerns. SCAORA’s recommendations highlight the need for judicial data to remain within sovereign infrastructure, supported by audits of existing AI tools such as SUPACE, SUVAS, and AI-assisted e-filing systems.⁹

Taken together, the current framework reveals a patchwork of constitutional provisions, ethical duties, and draft regulations that lack coherence. While they establish accountability in principle, they fail to resolve the practical challenges of explainability, transparency, and enforceability in AI-assisted judicial decision-making.

III. Case Law Analysis

Judicial engagement with artificial intelligence has already begun, though often indirectly. Courts have confronted issues such as fabricated precedents, automation bias, and the constitutional limits of AI deployment.

In Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd. (2024), the Supreme Court adopted a zero-tolerance approach toward AI hallucinations, warning against reliance on machine-generated citations without human verification.¹⁰ This case illustrates the judiciary’s insistence that AI outputs must be subject to rigorous oversight, reinforcing the principle that accountability cannot be outsourced to algorithms.

Earlier negligence jurisprudence, such as Jacob Mathew v. State of Punjab (2005), demonstrates the difficulty of applying traditional doctrines to AI contexts.¹¹ Negligence assumes a human actor who owes a duty of care and breaches it. When harm arises from opaque AI reasoning, the causal chain between human conduct and machine output becomes tenuous, leaving courts without a clear basis for liability.

Comparative developments also shed light on possible directions. The Automated Vehicles Act, 2024 (UK) places liability on the entity authorizing autonomous driving systems,¹² while the EU AI Liability Directive (2022) introduces a rebuttable presumption of causality where defendants fail to comply with AI obligations.¹³ These approaches show how jurisdictions are experimenting with frameworks that shift burdens of proof or impose strict liability in high-risk contexts.

Together, these cases and legislative models highlight the judiciary’s growing awareness of AI risks, but also the inadequacy of existing doctrines to resolve accountability gaps. They set the stage for a deeper evaluation of the challenges posed by AI in judicial decision-making.

IV. Critical Evaluation

The existing framework for AI liability in judicial contexts suffers from structural inadequacies that undermine both accountability and litigants’ rights.

First, the “black box paradox” remains unresolved. Modern generative AI systems operate through opaque neural networks whose internal reasoning cannot be meaningfully explained. While the Draft AI Regulations, 2026 place accountability on judges and court officers, this responsibility is illusory if the human actors cannot trace or verify the AI’s reasoning process.¹⁴

Second, the risk of AI hallucinations poses a direct threat to judicial integrity. In Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd. (2024), the Supreme Court adopted a zero-tolerance approach toward fabricated AI-generated precedents, underscoring that reliance on unverified machine outputs can erode trust in the judicial process.¹⁵ Mandatory “human-in-the-loop” and “human-on-the-loop” safeguards are therefore essential to ensure that AI outputs are subject to meaningful human oversight.

Third, automation bias exacerbates the problem. Judges and court staff, operating under heavy workloads, may defer excessively to machine-generated outputs. Even administrative functions such as listing urgent matters or identifying filing defects can materially affect litigants’ rights.¹⁶ Without clear restrictions, AI deployment in these areas risks undermining procedural fairness.

Fourth, data sovereignty raises constitutional and security concerns. Judicial data processed through infrastructure controlled by foreign technology companies compromises both independence and confidentiality. SCAORA has recommended that court data remain within sovereign infrastructure, supported by immediate audits of existing AI systems such as SUPACE and SUVAS.¹⁷

Finally, the constitutional basis of the Draft AI Regulations remains ambiguous. While Article 145 empowers the Supreme Court to frame rules governing its own practice and procedure, it does not authorize binding regulations on High Courts, which possess independent rule-making powers.¹⁸ This gap highlights the need for a legislative framework enacted by Parliament to ensure clarity, legitimacy, and uniformity across jurisdictions.

It is submitted that without addressing these deficiencies—opacity, hallucinations, automation bias, data sovereignty, and constitutional authority—the deployment of AI in judicial contexts risks undermining both accountability and judicial independence. A phased, precautionary approach with strict liability for high-risk applications and mandatory human oversight is therefore imperative.

V. Comparative Perspectives

The challenges of assigning responsibility for AI decisions are not unique to India. Comparative analysis reveals two broad approaches: fault-based liability with procedural adjustments, and strict liability for high-risk applications.

In the European Union, the proposed AI Liability Directive (2022) introduces a rebuttable presumption of causality where a defendant fails to comply with applicable AI obligations and the non-compliance plausibly links to harm.¹⁹ This lowers the evidentiary burden on claimants without abandoning the fault-based model. Alongside this, the revised Product Liability Directive (2022) expands the definition of “defect” to include evolving AI behavior, ensuring that manufacturers remain accountable even when harm arises from autonomous adaptation.²⁰

The United Kingdom has adopted a sector-specific approach. The Automated Vehicles Act, 2024 places primary liability on the entity authorizing the use of automated driving systems on public roads.²¹ This framework ensures clarity in a high-risk domain but leaves unresolved broader questions of AI liability outside the autonomous vehicle context.

In the United States, courts have largely relied on existing product liability doctrines at the state level. Several decisions have grappled with whether AI outputs constitute “products” or “services,” a distinction with significant implications for liability standards.²² Academic commentary increasingly advocates for strict liability in high-risk AI applications, drawing analogies to abnormally dangerous activities under the Restatement (Third) of Torts.²³

These comparative models demonstrate that while jurisdictions differ in their strategies, there is a common recognition of the accountability gap created by opaque AI systems. India can draw lessons from these experiences by adopting a hybrid framework that combines strict liability for high-risk applications with procedural presumptions to ease evidentiary burdens in other contexts.

VI. Conclusion

Artificial intelligence is no longer a peripheral tool; it is increasingly embedded in judicial and quasi-judicial processes. Yet, as this article has demonstrated, the current framework is inadequate to ensure accountability for AI-assisted decisions. The black box paradox makes it impossible to meaningfully trace AI reasoning, while risks such as hallucinations and automation bias threaten the integrity of judicial outcomes.²⁴ Data sovereignty concerns further complicate matters, as sensitive judicial information processed through foreign-controlled infrastructure undermines independence and confidentiality.²⁵

Comparative models offer valuable lessons. The European Union’s AI Liability Directive lowers evidentiary burdens by introducing rebuttable presumptions of causality,²⁶ while the United Kingdom’s Automated Vehicles Act, 2024 provides clarity in a high-risk sector by placing liability on the authorizing entity.²⁷ These approaches illustrate that targeted reforms are achievable without stifling innovation.

It is submitted that India should adopt a hybrid liability framework: strict liability for high-risk AI applications, coupled with procedural presumptions to ease evidentiary burdens in other contexts. Such a framework must be supported by mandatory human oversight and robust data governance safeguards. Only then can AI be responsibly integrated into judicial processes without compromising litigants’ rights or judicial independence.

The central challenge is clear: ensuring that technological efficiency does not eclipse human judgment. By enacting principled reforms, India has the opportunity to lead globally in shaping a just and workable accountability regime for AI in law.

Reference(S):

Cases

• Donoghue v. Stevenson, [1932] AC 562 (HL) (UK).
• Jacob Mathew v. State of Punjab, (2005) 6 SCC 1 (India).
• Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd., (2024) 6 SCC ___ (India).

Legislation & Regulations

• Constitution of India, arts. 142, 145.
• Consumer Protection Act, No. 35 of 2019, § 2(34) (India).
• Automated Vehicles Act 2024, c. 15 (UK).
• European Commission, Proposal for an AI Liability Directive, COM (2022) 496 final.
• European Commission, Proposal for a Revised Product Liability Directive, COM (2022) 495 final.
• Draft Regulations for the Use of Artificial Intelligence in Courts, Supreme Court of India Artificial Intelligence Committee (2026).

Reports & Institutional Documents

• Supreme Court Advocates-on-Record Association, Comments and Recommendations on the Draft Regulations for the Use of Artificial Intelligence (AI) in Courts, 2026 (2026).
• Law Commission & Scottish Law Commission, Automated Vehicles: Joint Report (Law Com No 404, 2022).

Secondary Sources

• Ryan Abbott, The Reasonable Robot: Artificial Intelligence and the Law (Cambridge Univ. Press 2020).
• Lilian Edwards & Michael Veale, Slave to the Algorithm? Why a Right to an Explanation is Probably Not the Remedy You Are Looking For, 16 Duke L. & Tech. Rev. 18 (2017).

Note(S):

1. Statement attributed to the Chief Justice of India, cautioning against over-reliance on AI in judicial decision-making (2025-26 public remarks).
2. Draft Regulations for the Use of Artificial Intelligence in Courts, Supreme Court of India Artificial Intelligence Committee (2026).
3. Supreme Court Advocates-on-Record Association, Comments and Recommendations on the Draft Regulations for the Use of Artificial Intelligence (AI) in Courts, 2026 (2026).
4. Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd., (2024) 6 SCC ___ (India) (discussing risks attending unverified, machine-generated citations).
5. Constitution of India, art. 142.
6. Constitution of India, art. 145.
7. See supra note 1 and accompanying text.
8. Draft Regulations for the Use of Artificial Intelligence in Courts, Supreme Court of India Artificial Intelligence Committee (2026).
9. Supreme Court Advocates-on-Record Association, Comments and Recommendations on the Draft Regulations for the Use of Artificial Intelligence (AI) in Courts, 2026 (2026).
10. Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd., (2024) 6 SCC ___ (India) (discussing risks attending unverified, machine-generated citations).
11. Jacob Mathew v. State of Punjab, (2005) 6 SCC 1 (India).
12. Automated Vehicles Act 2024, c. 15 (UK).
13. European Commission, Proposal for an AI Liability Directive, COM (2022) 496 final.
14. Draft Regulations for the Use of Artificial Intelligence in Courts, Supreme Court of India Artificial Intelligence Committee (2026).
15. Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd., (2024) 6 SCC ___ (India) (discussing risks attending unverified, machine-generated citations).
16. Supreme Court Advocates-on-Record Association, Comments and Recommendations on the Draft Regulations for the Use of Artificial Intelligence (AI) in Courts, 2026 (2026) (discussing automation bias in judicial administration).
17. Supreme Court Advocates-on-Record Association, Comments and Recommendations on the Draft Regulations for the Use of Artificial Intelligence (AI) in Courts, 2026 (2026).
18. Constitution of India, art. 145.
19. European Commission, Proposal for an AI Liability Directive, COM (2022) 496 final.
20. European Commission, Proposal for a Revised Product Liability Directive, COM (2022) 495 final.
21. Automated Vehicles Act 2024, c. 15 (UK).
22. See, e.g., discussion in U.S. state-level product liability cases addressing AI outputs (citations vary by jurisdiction).
23. Ryan Abbott, The Reasonable Robot: Artificial Intelligence and the Law (Cambridge Univ. Press 2020).
24. Pooja Ramesh Singh v. Jammu & Kashmir Bank Ltd., (2024) 6 SCC ___ (India).
25. Supreme Court Advocates-on-Record Association, Comments and Recommendations on the Draft Regulations for the Use of Artificial Intelligence (AI) in Courts, 2026 (2026).
26. European Commission, Proposal for an AI Liability Directive, COM (2022) 496 final.
27. Automated Vehicles Act 2024, c. 15 (UK).