Tag: Singapore

Singapore Inhibits Rightsholders Ability to Use Contracts to Prevent Unlicensed Text and Data Mining of Content

Image: Shutterstock (AI modified)

Singapore already has one of the most permissive text and data mining (TDM) exceptions in copyright law found anywhere, allowing AI developers to ingest copyrighted content for AI training purposes subject only to a very few limitations, all of which are pretty minimal. These provisions were introduced back in 2021 when several changes were made to the Copyright Act, some, frankly, better than others from the perspective of rightsholders. The TDM exception is not among the more positive outcomes. It is made worse by a provision in Singapore law that, for all practical purposes, prevents rightsholders from exercising contract terms to prevent the unlicensed appropriation of their content for the commercial benefit of third parties, namely, AI developers.

The term used to describe text and data mining in Singaporean law is “computational data analysis”.  As explained in this law firm blog post, this is defined as including;

“using a computer program to identify, extract and analyse information or data from the work; (or) using the work as an example of a type of information or data to improve the functioning of a computer program in relation to that type of information or data – a specific example being the use of images to train a computer program to recognise images”

The exception also permits supplying the works to other persons, provided this is for the purpose of “(i) verifying the results of the computational data analysis carried out by the latter; or (ii) collaborative research or study relating to the purpose of such analysis carried out by the latter.”

This is simply commercial, unlicensed web scraping of copyrighted content for AI development by another name, with very few limitations.

Singapore is now contemplating opening that door even wider by permitting the circumvention of digital locks (aka Technological Protection Measures, or TPMs) to allow text and data mining for AI training, as I wrote about here. (Singapore’s New Copyright Act Three Years On: There’s No Need to Open the AI Exception Door Even Wider). This proposal needs serious reconsideration as it would seriously tilt the copyright balance in favour of AI platforms to the detriment of rightsholders.

Another example of Singapore’s increasingly permissive approach to copyright protection is its undermining of the sanctity of contracts by limiting contractual terms that prevent unlicensed and unauthorized use of copyrighted content. In other words, the Act limits the ability of contractual terms to protect against, or override, the TDM exceptions. Rightsholders cannot “contract out” of the exceptions.

This provision was included when the Copyright Act was last updated (2021). The list of exceptions that cannot be restricted by contract was expanded to include the exception for computational data analysis use, as well as a couple of other uses; use of a work in judicial proceedings or for legal advice, and functions of galleries, libraries, archives, and museums, the so-called GLAM sector.  The computational data analysis exception is the one of concern. It requires that to be enforceable, a contract that limits the ability of a user to scrape content must be individually negotiated. In other words, standard contract terms on websites that limit use (so called shrink-wrap or clickwrap agreements) cannot be used to override the TDM exception. This has the effect of rendering standard contractual terms virtually unenforceable. They become the exception rather than the norm.

The term “shrink-wrap agreement” was originally applied to the preprinted agreement included as part of the packaging containing a software program. By opening the packaging, the user agreed to comply with the licence terms of the software. The term has since been expanded to include “clickwrap agreements” that take effect when a user accepts the terms and conditions of a website. This can be used to specify that the content about to be accessed can only be used for certain purposes or under certain conditions. One of these conditions could be a restriction on use of unlicensed content for AI development. The Singapore legislation eliminates what is a standard practice used by rights-holders in many parts of the world to protect and control use of their content. It also means that robot.txt files used by rightsholders to signal that their content should not be freely scraped (compliance is voluntary) are unlikely to be respected in Singapore. Robot.txt limitations are often included in clickwrap agreements.

Not only does the Singapore law allow for a broad undermining of contractual terms, and prevents “contracting out”, but its TDM exception is very wide in terms of application. In the UK, while contractual terms cannot override the TDM exception, unlike in Singapore allowable TDM use is much narrower. The exception in the UK can be used only “for the sole purpose of research for a non-commercial purpose”. No such restriction exists in Singapore. In the EU, contractual terms can override the general TDM exception (Article 4), unless the unlicensed access is “conducted by research organisations and cultural heritage organisations” or is “for the purposes of scientific research”, (Article 3). In these limited cases only, the contractual override does not apply. This still provides broader protection for rights-holders, and where the contractual override is disallowed, it is for very limited purposes. This is a much more nuanced approach than the one adopted by Singapore.

Contract law is generally seen as the oil that lubricates the wheels of business. In the digital age, shortcuts in the form of clickwrap agreements have been used to convey contractual terms to users. In some jurisdictions, explicit consent is required by clicking “I Agree”. Singapore’s current copyright legislation undermines the sanctity of contracts by imposing unrealistic conditions, particularly with respect to limiting the rights of rightsholders to prevent web-crawlers from ingesting copyrighted content without licence or permission. To say this is problematic is an understatement.

Singapore can do better. As an exemplar of rule of law in the region, it should be as assiduous in protecting the rights of copyright owners as it seems to be in advancing the interests of AI developers. The motivation, apparently, is to promote “innovation”. This is a misread of what brings about innovation. True innovation comes with a partnership between rights-holders and users that protects and compensates rights-holders for the time, effort and investment they have put into developing content that is clearly of value to the AI community. That content should be licensed, or at the very least, rights-holders should be given the option to opt-out through the ability to enforce contract terms, including overriding text and data mining exceptions when necessary.

© Hugh Stephens, 2024. All Rights Reserved.

Singapore’s New Copyright Act Three Years On: There’s No Need to Open the AI Exception Door Even Wider

Image: Shutterstock (modified)

The 2021 Revisions

Back in 2021 I wrote a mixed review of the revisions brought in to update Singapore’s Copyright Act. There were some significant positives for rights-holders, notably a new provision that made it illegal to sell ISDs (Illicit Streaming Devices, i.e. specially configured set top boxes) and associated software applications that offered access to pirated content (movies, television shows, sports events), or to advertise that the devices provided access to such content. Until the new law was brought in, Singapore had been a hotbed for streaming piracy with set top box retailers exploiting loopholes in the law to openly advertise and sell devices that could easily be modified to access pirated content. The ISD situation has improved markedly although illegal streaming of sports content remains a problem, with blocking of illegal offshore streaming sites (site blocking) being one solution. There were some other upsides for rights-holders as well, such as enhancing creator’s rights through a new right of attribution and providing a new public performance remuneration right for sound recordings. However, there were also some potential concerns and warning signs with regard to a broadening of copyright exceptions.

The Text and Data Mining (TDM) Exception

Some of the exceptions related to use by schools, galleries, museums and archives, and were suitably constrained to avoid negative commercial impact on content providers. Another more controversial exception was a Text and Data Mining (TDM) exception for “computational data analysis”. This was defined in the legislation as including the use of a computer program to identify, extract and analyze information or data from a work. Training for machine learning was used as an example. To me, that sounds a lot like scraping content to train AI algorithms. The 2021 TDM exception applied to both commercial and non-commercial uses (in many jurisdictions there are either no specific exceptions for TDM, as in Canada, or the use is limited to strictly non-commercial research purposes, as in the EU, where there is also an opt-out provision for rights-holders). The Singapore TDM exception is one of the broadest exceptions for AI training purposes anywhere. While it does have some safeguards, they are minimal.

Lawful Access Required

These safeguards require that the user have lawful access to the work. If the work is protected by a TPM (Technological Protection Method, aka digital lock) such as a paywall, and if the TPM is circumvented, or if the original was obtained from an online location known to have flagrantly infringed copyright, access would not be lawful. The Copyright Fact sheet explanatory brochure published by Singapore’s Intellectual Property Office provides the following illustration of this;

“A company is developing an artificial intelligence programme that can translate books from one language to another. To do so, the company carries out processes which involve making copies of various books in order to “teach” the programme how to recognise patterns. The company may rely on this exception to make the copies without first obtaining permission from the copyright owners of the books, provided that the company complies with the other conditions under this exception, such as not using the copies for any other purpose apart from computational data analysis. Moreover, the company must have lawful access to the materials that it copies. This means that it must first purchase the books it wants to use or subscribe to services that provide it with access to those books. The company should not circumvent paywalls to access the books.”

But Lawful Access Protection is Thin

Even this degree of protection does not really offer much comfort. For example, lawful access can be obtained by purchasing the cheapest or lowest minimal access possible and then using that access to justify commercial exploitation. Or, as in the example above, the company could purchase exactly one copy of each book that it wants to ingest. In the case of a paywall, instead of purchasing an institutional subscription, an employee of a company could purchase a single, individual subscription, gain lawful access and then proceed to exploit the content without licensing it. This is similar to the situation faced by the online journal in Canada, Blacklock’s Reporter, (which I wrote about here) where an employee of a government agency that employs 6000 people purchased a single subscription for about $150 and then shared the password for that subscription with whoever in that agency felt they needed access to the paywall-protected content. The Federal Court of Canada, in a controversial decision that is being appealed, found there was no infringement because the access was licitly obtained, and the use complied with fair dealing criteria.

Singapore has yet another loophole where an unlicensed user can mount an “I didn’t know” defence. Under the current law if the user did not know or could not have reasonably known that the copy of the work they were using was infringing, they could get a pass. (Gee, I didn’t know…and I didn’t try very hard to find out). Under another subsection, even if the first copy is an infringing copy “but the use of infringing copies is necessary for a prescribed purpose” and if the user “does not use the copy to carry out computational data analysis for any other purpose…”, a further exception to lawful use is created. In other words, using pirated content for TDM purposes is okay if the circumstances are right. All in all, the existing protection for rights-holders under Singapore’s laws is exceedingly thin.

Proposed Circumvention of Digital Protection for TDM Purposes

With respect to Text and Data Mining, this is the situation that has prevailed from 2021 until now. The exception was very broad but had some minimal guardrails (a requirement for legal access, with some exceptions) that provided a thin layer of protection for creators and content rights-holders. Now, however, even that minimal layer of protection is under threat. Pushed by the AI crowd, Singapore is now proposing to legalize circumvention of a TPM/digital lock for computational data analysis. In a consultation paper issued in April, the Ministry of Law sought input on the existing exceptions that currently allow for circumvention of digital locks (these include use of obsolete software, use of assistive technologies for audio and ebooks, educational uses of films in certified media courses, use of film clips for criticism or review in the making of a documentary, and a couple of others). Under the legislation, these must be reviewed every four years.

In addition to the review of existing exceptions that allow for the breaking of a digital lock, the Ministry is proposing three new purposes, on the premise that “circumvention should be allowed for… situations where access control measures are considered to adversely impair legitimate, non-infringing uses”. The new proposed exceptions to the prohibition on circumvention relate to use of copyrighted materials in a public collection for purposes of preservation and replacement, or for administrative purposes, (both unobjectionable) but also use of copyright works and protected performances for computational data analysis. In plain language, under this last provision, it would be legal in Singapore to hack a TPM (digital lock) that has been created by an author to protect access to copyrighted content as long as the intended use was for AI training purposes! Frankly, that is outrageous, and basically strips copyright owners of their rights to their property.

Of course, that is not how the Asia Internet Coalition sees it. Its brief to the Ministry it states that;

“The absence of an exception for computational analysis creates a disjointed regulatory landscape where conflicting laws may grant Text and Data Mining (TDM) rights but simultaneously prohibit their exercise. This ambiguity risks stifling investment and innovation in AI technologies. We emphasize the need for clear and coherent TPM laws that explicitly allow circumvention for fair use and TDM exceptions, ensuring legal certainty for stakeholders.”

The only “ambiguity” in the current law is that access must be lawful and hacking a digital lock to access the content renders the access unlawful. The AI crowd wants to strip away the sole, already-inadequate remaining protection for rights-holders.

Protecting TPMs (Digital Locks) and International Trade Commitments

On the other side of the equation, various rights-holder groups have weighed in, including IFRRO, the International Federation of Reproduction Rights Organizations (no, nothing to do with fertility or childbirth), the IIPA, the International Intellectual Property Alliance, the Copyright Alliance and the European Publishers Council, among others. IFRRO has based its comments on the lawful access argument. To hack a TPM that is protecting access to content is the antithesis of lawful access. Legalizing burgling doesn’t justify burgling. (My words, not IFRRO’s). IFRRO also points out that unduly permissive regimes have been rejected elsewhere, citing the EU and UK, and that they do nothing to sustain creative ecosystems over the long term. In addition to these arguments, the IIPA and the Copyright Alliance, both US based organizations, cite the 2004 Singapore-US Free Trade Agreement. That Agreement provides for a specified list of exceptions to the anti-circumvention rule (text and data mining is not among them) and while additional “temporary” exceptions (up to 4 years) are allowed, they must “not impair the adequacy of legal protection or the effectiveness of legal remedies that the Party (ie Singapore) provides against the circumvention of effective technological measures.” (Article 16.4.7 (f)).

There is also concern that, if adopted, the legalization of circumvention for AI scraping would violate the “three step test” of the Berne Convention, to which both Singapore and the US are parties. That provision requires that any exceptions to copyright protection be limited to (1) “certain special cases,” (2) “provided that such reproduction does not conflict with a normal exploitation of the work,” and (3) “does not unreasonably prejudice the legitimate interests of the author.” To strip away TPM access control protection for a copyrighted work so that another party can use the content for commercial purposes would clearly, in my view, be a violation of these terms of the Convention. That said, I am only aware of one situation where the three-step test has actually been litigated under WTO dispute settlement rules.

Don’t Throw Singapore’s Creators Under the Bus

Will these interventions have the desired effect of preventing the door allowing AI copyright exceptions from being kicked open even wider in Singapore? One would hope so, for the sake of creators everywhere, including those in Singapore. Throwing the creative industries under the bus to promote “innovation” in the name of generative AI development is a shortsighted and ultimately counterproductive strategy. One would hope that reasoning would have as much or more impact on Singapore policy makers as references to clauses in the US-Singapore FTA, but all arguments count.

Singapore’s Copyright Act review exercise needs to result in a balanced approach that protects creative industries while allowing for text and data mining within reasonable limits. The TDM exception door is already open very wide.  Kicking it open even wider would be a serious policy mistake.

© Hugh Stephens 2024. All Rights Reserved.