Introduction
Innovation in technology is quickly incorporating artificial intelligence. Patent applications are increasingly including machine learning systems, autonomous technologies, AI-assisted healthcare, and human-computer interaction.
However, the foundation of patent law remains mainly conventional. When an innovator appropriately discloses their innovation to the public, they are allowed a limited monopoly.
This poses a challenging question for AI innovations.
What happens when an invention works, but the inventor cannot adequately explain how it works?
In Caleb Suresh Motupalli v. Controller of Patents, C.M.A. (PT) No. 2 of 2024, 2025:MHC:293, ruled on January 29, 2025, the Madras High Court addressed a variation on this issue. The Court maintained the denial of a patent application for a system that aimed to combine artificial intelligence, human intelligence, and digital and physical settings. The Court discovered many flaws in the specification, especially with regard to adequate disclosure, claim clarity, and technical impact demonstration.
The ruling is noteworthy because it shows how difficult it is to incorporate sophisticated AI discoveries into India’s current patent disclosure system, rather than because it considers AI inventions unpatentable. The ruling is noteworthy because it shows how difficult it is to incorporate sophisticated AI discoveries into India’s current patent disclosure system, rather than because it considers AI inventions unpatentable.
This becomes increasingly more pertinent in light of the growing black box issue in AI, especially in industries like healthcare where disclosing private technical details and confidential datasets may be necessary to explain or replicate an AI system.
The facts of the Case
Indian Patent Application No. 5606/CHENP/2012 was submitted by Caleb Suresh Motupalli. “Necktie Persona-Extender/Environment-integrator and method for Super-Augmenting a Persona to Manifest a Pan-Environment Super-Cyborg” was the title of the application.
The overall goal of the idea was to combine artificial and human intelligence and increase a person’s potential through contact with both digital and physical settings.Persona extension, augmentation, black box modernisation, and human interaction with AI and surrounding objects were among the topics included in the specification.
The Patent Office issued several objections, which included inadequate disclosure as per Section 10(4), ambiguity under Section 10(5), lack of an inventive step according to Section 2(1)(ja), and objections related to Sections 3(k) and 3(m) of the Patents Act, 1970. The application was ultimately denied by the Controller. The applicant later requested a review under Section 77(1)(f) and (g). After reconsideration, the Controller again declined the application. The applicant then turned to the Madras High Court.
The initial challenge: Was the appeal permissible?
Prior to examining patentability, the Court tackled a significant procedural matter.
Section 117A of the Patents Act allows for appeals to the High Court regarding specific orders from the Controller. The issue at hand was whether an appeal could be pursued against an order made in a review application. The Court determined that an appeal under Section 117A concerning an order issued in a review was not permissible on its own. The Court also looked into the substance of the case. The Controller reassessed the patent application and effectively conducted a review of the substantive matters. Consequently, the Court moved forward to analyze the patentability issues that arose from the contested order. This element of the ruling is significant as it highlights the difference between the formal aspect of an order and the substantive evaluation carried out by the Patent Office. Thus, a procedural error did not hinder the Court’s ability to review the foundational patentability questions.
Sufficiency Of Disclosure Under Section 10(4)
The Court also looked into the substance of the case. The Controller reassessed the patent application and effectively conducted a review of the substantive matters. Consequently, the Court moved forward to analyze the patentability issues that arose from the contested order. This element of the ruling is significant as it highlights the difference between the formal aspect of an order and the substantive evaluation carried out by the Patent Office. Thus, a procedural error did not hinder the Court’s ability to review the foundational patentability questions.
The AI disclosure problem
AI systems can make enablement considerably more complex. A conventional mechanical invention can often be described through its components, their arrangement and their method of operation.
An AI system may involve:
- training data;
- model architecture;
- preprocessing;
- training methodology;
- parameters;
- optimisation;
- computational infrastructure; and
- Inference mechanisms.
The relationship between these elements may be extremely complicated.A patent applicant may therefore be able to demonstrate that a model produces a particular result without being able to explain every individual decision made by the trained model.On its own, that fact should not lead to the conclusion that every invention by an AI is unpatentable.The basic issue is whether the technical mechanism described in the specification is enough to enable a skilled person to reproduce the invention claimed.Motupalli is important at this point.
The Court was not content with the applicant simply pointing out technological concepts which could be linked to the invention; it was necessary for the specification to explain how those concepts would in fact achieve the result claimed.
The judgment therefore provides a useful lesson for AI patent applicants: Just because it is stated what an AI system is intended to achieve does not mean that it also reveals how the invention achieves it.
Section 10(4)(b): Best mode and reproducibility
Section 10(4)(b) mandates that the person making the application should disclose the method which they know to be the best for carrying out the invention.
The applicant maintained that the best mode had been adequately disclosed by the architecture set out in the specification.The Court disagreed.It discovered that the specification contained no adequate technical criteria or practical instructions for achieving the result claimed.This brings up an interesting question about AI patents.
What is the best method for an AI invention?
The model’s architecture, the training method, the relevant features of the dataset, the preprocessing techniques, the optimisation method or other technical parameters could all be involved, depending on the type of the invention.The situation becomes especially difficult when some of these elements are valuable trade secrets.He or she may thus have to make a hard decision.If you disclose too little you may fail to meet the sufficiency requirement.If you disclose too much you might jeopardise information which has commercial value.When the invention is based on information for which the inventor has good reasons to keep it confidential, the traditional patent arrangement becomes much more complicated.
Section 10(5): The problem of broad AI claims
The Court also looked at Section 10(5), which mandates that patent claims should be clear, concise and fairly based on the material disclosed in the specification. The Court identified issues concerning the clarity of the claims and the relationship between the general claims and the technical disclosure.
This is particularly relevant to AI.
There is a possibility that applicants who are seeking patents for AI might have an incentive to draw up their claims in such a way as to cover a number of different implementations of the underlying technology.It is not legitimate for a patent to claim the whole of a technological field just because the specification shows a single implementation.For instance, a specification might show a particular machine learning architecture which can produce a given result, whereas the claim is written so as to include every AI system that can achieve that result.There is a discrepancy between the invention described and the monopoly that is being sought.
The principle is simple:The scope of the patent must match the technical contribution which is actually disclosed.
The problem is that AI inventions generally function at several levels of abstraction, a narrow technical implementation lying underneath a more general functional concept.The problem then becomes, when drafting a patent, to determine the point at which a functional claim for an AI becomes so general that it is no longer properly supported by the specification.
Section 3(k): AI is not automatically patentable
The third major point was subsection 3(k) of the Patents Act.
Section 3(k) excludes from patentability mathematical or business methods, computer programmes per se and algorithms.
Even so, Indian courts have acknowledged that inventions which are based on computers can be granted patents if they show a real technical effect or make a technical contribution.
The applicant cited Ferid Allani v. Union of India, The Delhi High Court, in the case of Ferid Allani, acknowledged that computer programmes are not automatically excluded if the invention involves a technical effect or makes a technical contribution.
The importance of Ferid Allani lies in the fact that Section 3(k) cannot be applied in a mechanical way just because software is concerned.However, Motupalli shows the other aspect of the principle.
The fact that AI is present does not by itself produce a technical effect. He has to show what the technical contribution actually consists of.
The Court looked into whether the invention in question had improved the functioning or efficacy of the system and whether it had produced a specific technical effect. It finally concluded that the specification did not adequately show such an effect.The result is an important distinction:The fact that AI is involved does not mean that it has made a technical contribution.Just because an algorithm is included in a more technologically advanced description does not mean that the invention becomes patentable. This point brings us to the issue of the black box.The importance of Motupalli goes beyond the facts of the case when the issue is viewed in the light of the growing problem of black box AI.The situation is especially complicated when it comes to medical AI.Considering an AI system which has been trained to detect a certain disease from medical images.In order to demonstrate enablement, the applicant will have to provide meaningful information regarding the model, the training process and the data on which the system is based.Yet medical datasets frequently contain sensitive personal information.Information which makes it easier to reproduce an AI invention might therefore be information that cannot just be made available to the public.This creates a problem with disclosure.Patent law demands disclosure.The development of AI usually relies on keeping things confidential.Privacy is now involved in the matter through the use of medical AI.
The result is a three way tension between:
- patent disclosure,
- business secrecy
- data protection
One of the most important issues in the field of AI patents remains unresolved.
Is explainability now the standard that must be met for a patent to be granted?
From the case of Motupalli arises an interesting question as to whether AI patent law will one day demand some level of explainability.Motupalli does not state that all AI inventions must be fully explainable if they are to be granted patent protection.There is a significant difference between explainability and enablement.It is possible for an AI model to be technically reproducible even though it is difficult to explain on a per-prediction basis, whereas an applicant might give a very detailed conceptual explanation without supplying enough information for a skilled person to carry out the invention.The legal requirement is therefore not necessarily: Make sure that the AI explains each and every decision it makes. It is closer to:Make it clear by giving sufficient details about the technical aspects so that the claimed invention can be carried out.The importance of this distinction will increase as AI systems become more complex.
Critical analysis
What makes the Motupalli judgment so strong is the way it stresses that patent protection must be based on a technical contribution that can be demonstrated.It is especially important in the case of AI since general conceptual claims could in principle lead to extensive monopolies concerning technological outcomes.A patent could allow exclusivity without giving the public the relevant technological knowledge if a specification only states the desired outcome and fails to resolve the technical means by which it is to be achieved.The Court’s requirement of adequate disclosure therefore advances an important public interest.The case also reveals a limitation of the present patent system in India.The traditional principle of enablement holds that the inventor provides enough explanation so that a person skilled in the art can carry out the invention.AI complicates this assumption.
Even if a machine learning model is reproducible in terms of its architecture and methodology it may still be difficult to explain when looking at individual outputs. This raises a question that Indian patent law has not yet fully answered:
What level of disclosure is sufficient in the case of an AI invention when the invention is reproducible but not completely explainable?
Motupalli does not give a clear answer.That could be since the issue before the Court was more fundamental in a factual sense. The Court concluded that there had been insufficient technical teaching and a lack of adequate working examples showing that the proposed techniques could in fact achieve the result claimed.The distinction is important. It therefore ought not to be regarded as establishing a general rule that AI-generated inventions need to be completely explainable.
In fact, it strengthens the current principle that technical assertions must be accompanied by adequate disclosure and evidence of implementation. What Motupalli means for AI patent applicantsThe judgment contains a number of practical lessons.
1. Do not use only abstract descriptions.
The specifications for AI patents should describe the real technical architecture and explain how the desired result is obtained.
2. Include working examples
Wherever possible, applicants should give specific examples to show that the invention functions as stated.
3. Set out the technical contribution
It is not enough just to state that AI is being used.The specification must describe the technical problem which the AI is addressing and explain how the system in question achieves a technical effect.Make sure that the statements are in agreement with the disclosure. Broad functional claims must be accompanied by sufficient technical disclosure.At the stage of drafting, attention should be paid to confidentiality. Applicants should expect the difficulty involved in balancing patent disclosure with the need to protect their proprietary AI models, datasets and training methods.
6. Regard Section 3(k) as a substantial issue
Inventions based on AI should not suppose that the use of advanced technology automatically places them outside the exclusion that applies to the computer programme itself.
The way forward
Since AI is being used more and more in creative activities, the Indian Patent Office may need to provide more detailed guidance.
AI patent applicants should be given more clear guidance as to the level of disclosure that is expected in the case of different types of AI inventions.Such guidance could address: model architecture; training methodology;dataset characteristics;reproducibility; technical effect;model parameters; working examples; and The handling of confidential or privacy-sensitive information.
At the same time, it would not be sufficient just to require the disclosure of all the information concerning an AI system. Patent law has to strike a balance between the public interest in having information disclosed and the legitimate interests concerned with trade secrets, privacy, and commercial confidentiality.The aim is to specify just that amount of technically significant disclosure which is needed for a skilled person to carry out the invention as claimed.
Conclusion
Caleb Suresh Motupalli v. Controller of Patents is an important case in the developing relationship between artificial intelligence and Indian patent law.
Its significance does not lie in establishing that AI inventions are unpatentable.Instead, the judgment demonstrates that AI inventions remain subject to the fundamental architecture of patent law: the inventor must disclose enough of the invention to justify the monopoly being sought.
The difficulty is that AI increasingly challenges the assumptions underlying that bargain.AI systems can depend upon enormous datasets, complex architectures and learned parameters. Their outputs may sometimes be difficult to explain even when the underlying system can be technically reproduced.This creates a difficult question for patent law:
How can the law demand meaningful disclosure from technologies whose operation may itself be opaque?
For now, Motupalli provides a relatively clear answer.
An AI patent application cannot rely upon technological vocabulary, ambitious objectives or the mere invocation of artificial intelligence. The specification must demonstrate the technical pathway from the invention claimed to the result promised.
The future challenge for Indian patent law will be determining how that principle should operate when the technology itself becomes increasingly difficult to explain. The patent system has always operated on a bargain between monopoly and disclosure.
