Some years ago, I had traced out two decades of the ‘confusingly confounding’ regulation of software patents in India. The “ping-ponging” journey traced in that post seems to have been a precursor for the judicial fragmentation that Section 3(k)’s interpretation would soon see. Last week, as a few of us (Bharathwaj Ramakrishnan, Yogesh Byadwal, Anushka Dhankar and myself) were putting together comments for the 2025 Draft Computer Related Inventions (CRI) Guidelines and after some grueling work going through as many orders as we could from the recent few years – it really stood out to us as to how many different ‘tests’ were being used in interpreting Section 3(k).
There is a lot to say on CRIs, from detailed policy based critiques, to more nuanced internal critiques such as on disclosure requirements and algorithmic transparency. However, for the limited purposes of this call for comments, we decided to eschew all of that and ask for the simple (yet perhaps unlikely) “disclosure” of how fragmented the interpretative landscape of Section 3(k) has been. After all, the patent office is bound by judicial precedent. If judicial precedent doesn’t demonstrate a clear way forward (with clear reasoning), then the question of adequately following processes in examining patent applications gets stymied at the get go. In the draft guidelines, some excerpts from some cases are mentioned. However, why are those mentioned in the sections they are? What about those cases is authoritative in terms of guidance that can be taken? What about the ones, from equal judicial authority, that are not mentioned? Why are some being chosen over others?
The Patent Office it seems has been put in a difficult place on this front. After all, they are unlikely to want to be the ones to point out this fragmentation. Yet, if this fragmentation is not acknowledged, what is the way forward for examiners/controllers when they need to lay out the interpretation and reasoning that they are so often (rightly) criticised for not giving?
Our submitted comments are available here.
For the more immediate purposes of the above point though, the two charts below (created by Anushka) presents what we could make of the interpretative inconsistencies just over the last few years. The first chart below shows all the different ‘technical’ tests that have been used in relation to ‘computer programmes per se’ (meanwhile, to me, its still unclear what ‘technical’ even is supposed to mean, especially in the “technical” field of computers!). And the second chart below deals with the other subject matters under Section 3(k). While this area continues to evolve, we hope these (along with the annexure in the comments) are useful for anyone who is looking to understand what’s happening on this front!
(We had inserted two charts conveying the same information in our submitted comments, however, they were redesigned a bit to be made accessible as well as a bit easier to read)
You guys have done a great job in compiling a set of comments. I am in agreement with you on seeing how the document is jumping from one topic to another. The simplest mode seems to be to take a flow chart and start defining how the computer related inventions (CRIs) are treated. I miss that type of clarity and precision from the entire guidelines document. But I am sure the patent office heard your point, and hopefully, incorporate this.
There is however something new about this document from the patent office. I found the sections dealing with sufficiency and the best mode requirement under sections 10(4) and (5) of the Patents Act to be quite explicit in listing the enablement or disclosure requirements. Clearly, those passages have been drafted taking the rise of the artificial intelligence (AI) related inventions and the corresponding patent applications into account. The disclosure requirements are quite specific for the AI and is set at a higher threshold, which is quite understandable.
Everyone including the patent offices are quite overwhelmed how quickly it all changed. In India I notice a plethora of use cases (very innovative) being developed by using the available LLMs. However, if these use cases start coming to the patent offices, the applicants or patent owners are going to ask also the benefit of disclosing the training dataset and more detailed information in the patent applications. For instance, the following section from the first paragraph of page 25 is quite instructive from the guidelines document:
“The invention must disclose specific implementation elements critical to reproducing the AI model’s functionality, such as training data sources, data pre-processing steps, chosen learning models (e.g., neural networks, decision trees), and any applied loss functions.”
That particular section of the document asks for everything from the applicant in order to get a grant. We can argue all day long what are the benefits and disadvantages of disclosing the details of training data. But this is a different kind of approach which have not been demanded from the generic computer related inventions.
Things are about to get really complex. Thus, I don’t see the wisdom of dealing the generic computer related inventions and the artificial intelligence and its allied fields in the same document. But I will settle for at least the start of the debate in India on this.
There is a challenge to the patent practitioners all over. On a different note, how are we going to enforce the AI related inventions in the courts? Are the current pre-trial procedures in the Indian courts enough for addressing the challenges posed in enforcement of the AI related inventions? I am not sure. Unless you can convince the applicants and patentees on their enforcement options, asking for training data and the details of the AI related inventions will be a hard sell. But that is a topic for a different day. Everyone is having at the moment a black-box approach. We may have to take a step back and ask whether the current state of the Patents Act, 1970 is equipped to address such new age inventions in which Indian inventors may gain some edge?
Too many open questions.
Hello Anon,
Thank you very much for your kind comments. I agree with you on the question of disclosure, wherein it has some strict requirements regarding disclosure for AI-related inventions. As a general rule, more disclosure is better, as it is part of the Patent bargain, that in return for the 20-year right to exclude, the applicant has to give their pound of flesh (full and complete disclosure so that the person skilled in the art can work the invention).
Now, coming to the question as to whether the pre-trial procedures in India, as they exist, are sufficient to facilitate such extensive disclosures?
My feeling is that it might be robust enough because in this blog itself, many bloggers have criticised the Courts for being trigger-happy with interim injunctions, which are usually granted pre-trial. The Injunctions are harsh to the extent it might cripple an business. This trend is across the board for all IP matters.
But then, this position I have taken above might not stand if there are some unique problems that AI-related inventions present that might prevent the grant of this interim injunction or any other interim measure, for that matter. I am unable to think of any specific problem(s) that AI-related inventions might present to the Courts. If you think there are such issues, kindly let me know in the comments.
As to the question of whether the current Patent Act is equipped to handle AI-related inventions?
If I can reframe this question a bit as follows, ‘whether the current Patent Act has sections that signal specific policy preferences towards AI-related inventions?’
It might not be, since I find it difficult to believe the legislature would have envisioned anything close to the current AI-related inventions when the substantial provisions of the Patent Act were written down. Yet, unfortunately, the current Govt’s stance seems to be that India’s IP statutes are quite capable of dealing with the AI disruption (https://www.pib.gov.in/PressReleasePage.aspx?PRID=2004715).
Since there might not be specific guidance from the legislature, our Courts will have to interpret the sections, and I have a feeling that legal developments in the EU and the US will have a disproportionate effect on how the Courts will decide. I am not sure whether this is a good or a bad thing.
Now, the question is what specific policy measures can be incorporated into the Indian Patent Act that will give Indian inventors an edge?
The easiest answer you would get from people is to extend clear and unambiguous patent protection to AI-related inventions. Yet, I am highly suspicious whether these claims can be backed by any empirical data and are contrary to Indian patent history. The Patent Act initially withheld product patent protection to pharmaceutical inventions, which paved the way for the growth of the generic companies in India. This policy measure that India took itself was informed by history, wherein Germany chose to deny product patents for Pharmaceuticals to cultivate its pharmaceutical industry. I am not suggesting that we should deny patent protection to AI-related inventions. But there should be a more empirically grounded approach to understand what specific measures can give Indian inventors an edge, instead of simply stating that patent protection will solve all problems. These measures might even be outside patent law (for example, making available compute facilities by the government for inventors or entrepreneurs to build and train their models, or more public R&D investment into open source AI development projects, which democratizes knowledge and skills etc)
Maybe I should have clarified further on the disclosure requirements from the IPO further. What the IPO asks seems to go beyond the curent understanding of the Patent bargain. The Patent office asks for Applicants to “disclose specific implementation elements critical to reproducing the AI model’s functionality”. Reproducability sounds like a higher threshold of disclosure for AI related inventions. What does this mean for patent practice and patent agents involved in drafting AI cases? We have to see what kind of objections the FERs raise under sections 10(4) and 10(5) and I hope the Examiners at the the IPO are up for the challenge.
Regarding the enforcement under current pre-trial procedures, injunctions are not what I meant to be frank, even that may not be easy for the AI products offered by competitors. I meant more in evidence gathering measures related to the AI products. There appear some serious challenges in relation to evidence gathering for enforcing AI related inventions. At the moment, the AI products with specific use cases are not available for everyone to perform, for instance, infringement analysis. So traditional search and siezure measures need to be flexible for the AI related patent rights, if the ask from the patent office is “reproducibility”.
Even simple embedded systems used in current automobiles are not prone to flashing ECUs. Reading the software inside is almost close to impossible with specialized crypto and associated hardware tools. If the AI products are cloud-based with security measures in place, how are the patent holders going to see what is being used in a competitor’s product without a quick search and siezure? It gets even further tricky with machine learning algorithms. By definition, these algorithms also prone to self-correction and/or self-improve over time during multiple iterations.
This has to be studied in detail, in my view. Whether the right holders bite the bullet on higher threshold of disclsoure for AI related inventions depends on how easy it is to enforce the rights generated out of the AI related patents.
Respected Anon,
You have given me a lot to think about, and I agree with you that more thought should go into this. Thank you for the detailed response!
Hey there, I think the Indian patent office has heard you in some respects:
https://ipindia.gov.in/writereaddata/Portal/Images/pdf/Draft_CRI_Guidelines_2025_v2.0.pdf