The New Draft Regulations: A Landmark Development

The integration of Artificial Intelligence (AI) within Indian courts is no longer a distant prospect. The Supreme Court has initiated a significant development by introducing the “Regulations for the Use of Artificial Intelligence in Courts (2026)”. The adoption of AI is imperative, considering the substantial backlog of cases currently burdening the Indian judicial system.

Here Regulation 4 talks about “human primacy”, basically saying AI should only be used for the task which are assistive in nature and must remain subservient to human judgement. It also admits that Large Language Models (LLMs) can “hallucinate”, which means they seem reliable but the outcome generated is ultimately a byproduct of an inevitable technological error. And also, it takes into consideration the principle of “public data sovereignty” which means that the procurer/vendor of a technology can't claim his intellectual property over the models after feeding the public judicial data. Also, Regulation 39, which talks about setting up an AI Incident Database, feels like better management, because AI systems seem to commit the same kind of mistakes over and over, recording and sharing these incidents with other peer models and to different courts will boost the productivity.

But good laws need more than just good intent. The draft has a big contradiction when looked closely. The rules push for using AI in courts, but then they make officers personally liable for “AI Incidents” even ones they couldn't have seen coming or inevitable because of the technology used.

Like, imagine someone using an AI tool for legal research and if the AI model brings up a fake case or overruled ruling, that's a “hallucination” per Regulation 3(z). With this draft, officers can be held liable for no “fault liability” it's all on them. So, who pays when a machine commits error and causes harm?

Key Issues for consideration

The Strict Human Liability Model:

The Regulation 8 of the draft expressly states that the liability for every decision assisted by AI will rest solely on the officer. The regulation precludes an individual from invoking three defenses.

(a) AI generated Outputs

(b) Opaqueness of “black box”

(c) Hallucination of algorithms, where the judgement is harmful, illegal, or erroneous.

Additionally, Regulation 43(6) extends this standard to litigants and lawyers, who are similarly precluded from invoking these technological limitations as a defense and hence will suffer for No fault liability. Similarly, these provisions establish a strict liability regime for officers, litigants, and lawyers. And are held responsible for the errors committed by Court-approved AI models, and where the exercise of “reasonable care” does not constitute a valid defense.

This regulatory framework elicits serious concerns regarding Art. 14 of the Constitution of India, which prohibits arbitrary state action. The Supreme Court, in E.P. Royappa v. State of Tamil Nadu (1974) 4 SCC 3, held that “there cannot be harmony between arbitrariness and equality.' If an officer is held responsible for a judiciary approved inevitable “Algorithmic deficiency”, and where all the reasonable precaution has been taken, then imposing liability without any fault appears “ex facie” arbitrary. Under such conditions, the regulations risk deeming individuals liable for risks that are entirely beyond their control.

Furthermore, Art. 21 of the Constitution demands not just the procedure est. by law, but also strict requirement of “due process of law”, as laid down in Maneka Gandhi v. Union of India (1978) 1 SCC 248; Under these regulations, a court officer may act bona fide on an approved AI Model and nonetheless incur liability for an undetectable error, therefore questioning due process. Essentially, the litigants are forced to bear the consequences, due to a failure of Algorithm a litigant has to compromise their access to justice.

A “Black Box” has been defined under Regulation 3(n) as an AI system whose outcome cannot be determined. This indistinctness is explicitly recognized in Regulation 7(3), which demands heightened vetting. However, under Regulation 8(1), a judicial officer cannot take the defence of opacity. Therefore, in practical terms, the draft acknowledges that AI models cannot be fully comprehensible however, it continues to place the full obligation on human officers.

This system directly clashes with the established precedent pertaining to liability in M.C. Mehta v. Union of India (1987) 1 SCC 395; The principle of “Absolute Liability” was formulated for entities involved in hazardous undertakings that capitalized on those activities. As per Absolute liability principle it will never to hold responsible a judicial officer for the errors committed by Algorithmic models sanctioned by courts. Where the damage lies within a black box and the officer himself is not equipped to examine the system reasonably, placing sole liability on such officers jeopardizes shifting accountability into a mere “fictio juris”.

“The Liability Void”: Accountability Lacking Effective Remedies

So, the Regulation 8 says if an officer messes up using AI, they're on the hook. You can't blame the AI, or say it was a “black box” or “hallucinated” outcome to escape the responsibility for a algorithmic error. Then Regulation 43(6) comes up, about documents and proof that if you file something fake or wrong made by artificial intelligence, then the onus is on that person itself. The AI's output isn't an excuse. It sounds like strict liability for AI generated outcomes.

But the weird part is, it's silent about the penalties imposition framework to decide the liability. similarly, Regulations 20 and 46, also nails down who's liable for using AI, but there's no actual punitive guidelines been laid down.

The Digital Divide and Judicial Access

As per the Regulation 13 of the Draft regulations which reiterates that “AI systems shall encourage inclusivity and must expand equitable Access to Justice and shall not create or widen digital divides”. This aspiration is constitutionally grounded post Anita Kushwaha v. Pushap Sudan (2016) 8 SCC 509, where “Access to justice was explicitly recognized as a part and parcel of the right to life under Art. 21 of the Constitution” (para 29). However, this regulatory aspiration is not accompanied with a well-grounded implementation mechanism, for the enforcement of these draft regulations.

The Draft Regulations envisions for AI-assisted court processes, including digital filing, and online scheduling, all of which presuppose digital access, device ownership, internet connectivity, and digital competence to every citizen of this country, However the digital divide in India is significant as internet penetration ranges from 37 % to 67% in rural and urban areas, respectively (as per NSSO data), and digital literacy remains substantially lower among elderly citizens, women, and marginalized groups of the society.

Although the vision underlying these rules are ambitious, existing legal gaps must be addressed prior to finalization. So, first thing first, when an officer utilizes an authority-approved AI system, acts in good faith without negligence, and adheres to mandated procedures, they should be protected from liability, and punitive measures should not be imposed for algorithm errors (AI incidents). Such protection is consistent with the principle of “Actus non facit Reum nisi men's sit rea” which requires that liability be based on fault instead of solely on outcomes. Failure to implement this safeguard would result in arbitrariness and contravene the Principles of Article 14 of the Constitution.

Secondly, when a court-approved AI system produces erroneous outcomes, both the AI Secretariat and the approving authority should assume significant liability. The draft should include an explicit provision establishing joint liability.

Thirdly, the immunity granted to vendors under Chapter VI of the regulation should be extended to individual officers. If a faulty system results in liability for officers, vendors should also be subject to joint and several liabilities. Officers' liability should be parallel to that of vendors.

Fourthly, the grievance handling mechanism under Regulation 52 should assess whether it goes beyond the prescribed use. At present, the framework lacks a legal remedy for litigants adversely affected by court-authorized AI tools. This situation imposes prejudicial and untenable consequences on litigants as this system is fundamentally intended to promote access to justice. It should safeguard not only institutions but also officers, litigants, and lawyers who rely on the legal system.

And finally the lack of digital infrastructure in Indian courts specially in district courts is a major hurdle for integration of artificial intelligence in judicial system as most of these courts doesn't even have the basic video conferencing facilities then this expectation is a far too much, and will require a infrastructural boost and training etc.

Views are personal.