DPIIT Working Report: Lawful Access while ignoring Copyright Law

When I was reading the DPIIT working paper on AI & Copyright, my mind instantly went back to Bartz v Anthropic. To be specific, I recalled the following lines from the judgement- “The downloaded pirated copies used to build a central library were not justified by a fair use.”

As Ed Lee explained here, Bartz did not consider it relevant whether the pirated copies would be used for transformative use. The fact that a central library was built using pirated books disqualified Anthropic from using fair use as a defence. I use the word ‘disqualified’ since, effectively, that’s exactly what took place- that judgement never went into the question of whether such pirated works were put to transformative use. The fact that Anthropic did not license or purchase these books but instead chose to download from shadow libraries was the end of the infringement inquiry.

Unwittingly (or not), the Judge propounds a copyright-version of the ‘poisonous fruit’ theory- if the source of the evidence is tainted with illegality, the evidence thereby becomes inadmissible. Extend this logic to copyright, and we get a ‘poisonous use’ theory- if the source of where I got the book is illegal (Libgen, Scihub, Anna’s Archive), then my use of it to, let’s say write a book review, is also illegal. Although Section 52(1)(a)(ii) mentions writing a review is fair use, I cannot argue fair dealing because I used a pirated copy. 

Opening up fair dealing only to folks who can pay to purchase an authorised copy leaves the intended masses outside the protection of copyright regime. Author gets to decide who gets access to the work at the cost of user’s right to fair dealing, skewing the balance comprehensively in the former’s way. In the discussions I have had since Bartz came out, some people told me that it will blow over and, eventually, there will be course correction. 

Alas, this was not how it turned out. Bartz has given rise to, as Lee calls, a “shadow library strategy”- it shifts the focus from whether training is fair use to the manner of acquisition of the training copies at the initial stage. If acquisition was lawfully done, only then fair use is discussed. This post questions the soundness (and logic) of inserting a legal access requirement by the working group. But first, some thoughts on the report as a whole. 

Solving the Copyright Problem by Ignoring it

The committee report states that all “lawfully accessed” copyrighted content will be available as a matter of right. 

The logic for inserting this requirement can be found on page 62 of the report. Given the significance, I will excerpt it here in full- 

“It is proposed that lawful access be the prerequisite for the use of copyrighted works for the training of AI Systems. This means AI Developers would not be allowed to rely on the aforesaid mandatory license to bypass existing or future technological protection measures (TPMs) or to gain unauthorised access to works behind paywalls without making the necessary payment.”

Read the above para again- do you see the problem? 

Any AI model, trained on copies accessed and downloaded after bypassing TPMs, are plainly illegal. The working group simply elides the question of fair dealing (the report also does not say whether AI training is even infringement to begin as Akshat and Vishno highlight). Rather, on the question of whether AI training is fair dealing, it states:

“This working paper, therefore, does not attempt to resolve these questions or offer definitive conclusions on whether infringement is made out and/or the “fair dealing” exception applies. Instead, the goal is to propose a forward-looking legal and policy framework.”

The terms of reference of the committee, to be clear, had specifically mentioned–

“To examine the adequacy of existing provisions of the Copyright Act, 1957 in addressing these issues (use of Artificial Intelligence in the context of copyright).” 

How does the committee answer the above question- it doesn’t. It chooses to ignore copyright law itself. If you ignore it,  you don’t have to wade into the complicated questions of copyright law (precisely why the committee was constituted). Instead, the committee comes up with a jugaad of blanket licensing. 

Simply put, if a committee is constituted in face of a disruptive technology which, supposedly, challenges the foundation of copyright law, you don’t ignore the tough questions. You take them head on and give clarity on what the law says. Proposing “a forward-looking legal and policy framework” without even laying out what is wrong with the existing law is simply not done. 

Swaraj, Vishno, Ambika, Akshat, Shivam, Bharathwaj, in their respective posts, have already highlighted how the reports gets it wrong. So, without elaborating further, I invite you to read their views on the report. 

Lawful Access

Now that I have vented out my views on the report, let me cut to the chase: I completely disagree with the view that fair dealing is contingent on how you access the original work. I download it from annas archive or purchase it from a book store after paying, the right of fair dealing does not get defeated. If I bypass a TPM to access a work and use it for a purpose protected under section 52, nothing in the copyright Act prevents me from doing so. 

Section 65A(2) of the copyright act, which criminalises bypassing a TPM, states- 

• Nothing in sub-section (1) shall prevent any person from,—

(a) doing anything referred to therein for a purpose not expressly prohibited by this Act:

Swaraj, in 2012, on subsection (2), wrote– “the amendment adequately addresses legal concerns regarding possible sanctions against legitimate usage as per the fair dealing provisions of the Copyright Act.”  Amlan, too, had put faith on this provision back in 2010. I agree with both. This subsection, specifically, allows any individual to bypass a TPM and access a work locked behind a paywall. The only rider is that the work must be used for a purpose allowed under the copyright act. Clearly, it protects the fair dealing rights enumerated u/s. 52. 

The Government, when sec. 65A was introduced, was well aware of the impact it may have on fair dealing. 

In Lok Sabha, introducing 65A and 65B, Kapil Sibal (then HRD minister) had said– 

“These amendments are also necessary because we believe that there is a lot of piracy that is happening through technology in this country…People are using technology to take away somebody’s copyright and then use that material. We think the time has come to deal with piracy. Therefore, we have incorporated Section 65 (a).”

The intention of 65A was never to prevent fair dealing of a work. Rather, it was introduced specifically to target piracy. Although not defined anywhere in the copyright act, in R.G. Anand, the SC had said that “a violation of copyright amounts to an act of piracy.” 

Section 52, on the other hand, begins with a clear statement- “The following acts shall not constitute an infringement of copyright.” So, acts that are covered by fair dealing are not infringement in the first place. The question of piracy, therefore, never arises for actions covered by 65A(2). 

A Parliamentary standing committee was also constituted to suggest changes to the proposed amendments. In the report, Yahoo India had raised the concern that 65A “considerably interferes with consumer’s right to fair use.” In response, the Committee noted:

“the Committee agrees with the approach as enshrined in section 65 A  to give limited legislative guidelines and allow the judiciary to evolve the law based on practical situations, keeping in mind the larger public interest of facilitating access to work by the public.”

Notice the choice of words- facilitating access to work by the public. The committee was made aware that TPMs will prevent access to works even for lawful purposes under section 52. 65A, thus, was the safety valve which gave a breathing space for fair dealing of works. 

The working group, however, completely ignored the above context before inserting the lawful access requirement. It simply quotes other jurisdiction which have inserted a similar requirement as if that is a good enough justification. (to be clear, I am not a fan of Swadeshi Jurisprudence either)

A question could be asked here- what, then, is ‘unlawful access’ under 65A. 65A(1) requires two ingredients to be fulfilled- (1) act of bypassing the TPM with (2) an intention to infringe rights u/s. 14. So, if tomorrow, I bypass Oxford’s paywall and access a book/article with the intention to commercially redistribute, I have unlawfully accessed. The purpose to bypass, in this case, is not protected under 52. One could argue that my initial act of bypassing TPM itself demonstrates intent to infringe. The language is open-ended in that regard as acknowledged even by the Parliamentary committee. These are exactly the points the report could have raised when inserting lawful access as a pre-requisite.

The report could also have argued that AI developers operating for commercial motive are not covered by “any persons” under 65A. Maybe I would agree or maybe not. But at least, it would have answered a question it was tasked to solve in the first place: is copyright act, as it exists, adequate or not. The debates, then, would have brought the best out of copyright enthusiast (or critics). Unlike the report we have today, which, as Swaraj said, bulldozes copyright law. 

God forbid if the logic of this report finds takers and we have a fair dealing provision which becomes contingent on manner of access. I have that fear for a reason- the committee never suggested lawful access as an exception for AI training in the form of an amendment. It stated that requirement as if it was always the position or the norm. Or else, we might have a regime where fair dealing is meant for people who have the money to lawfully access. 

AI Sovereignty 

Just a small  point before I end- the committee, on pg. 62, noted that the requirement of lawful access is prospective i.e. this requirement is not applicable on companies who trained their AI models before this requirement kicks in. 

Think for a minute. The report itself noted that “access to large volumes of data and high-quality data is crucial” for AI. OpenAI has been collecting works for training since before 2020. Copyrighted or not, it kept on collecting and training. If the DHC decides that OpenAI’s use is fair use, even including the pirated work, is there a level playing field left for the Indian developers. Not at all. They are left to be the suckers who are paying to access the data and also pay a royalty on top of that. Can these developers, mostly startups, then, even imagine to compete with players like OpenAI, Meta or Google. 

This is a big if, I concede, but one which the Committee failed to consider in its report. The Indian Government has committed Rs. 10,371.92 crore for the AI mission. Against that level of investment and commitment, this report woefully falls short.