‘Life would be intolerable in personal and commercial terms, if information could not be given or received in confidence and the right to have the information respected by the force of law’.1 This is Sir John Donaldson’s remark in the appellate judgment in Attorney General v Guardian Newspapers (1990). The case involved the disclosure in a book, by a retired MI5 agent, of alleged irregularities and illegal activities within MI5.2 Sir Donaldson’s remark highlights the need to uphold the protection of confidential information, for personal or commercial life as well as governments to function. Nevertheless, he recognised that this protection is not absolute as ‘… there will be just cause or excuse for breaking confidence when there are countervailing public interests supporting publication which outweigh those supporting the right to confidentiality’.3 So the law should protect secrecy and facilitate disclosure when there is a compelling interest. In the era of advanced algorithms used for automated decision-making (ADM) impacting the rights of individuals and public interest at large, the extent to which commercial secrets should be protected has been a matter of longstanding controversy.

Trade secret protection of algorithms as a limiting factor in ensuring transparency in algorithmic decisions has been widely analysed.4 The dilemma arises from two potentially conflicting but legitimate interests. On the one hand, businesses have a legitimate interest in safeguarding their proprietary algorithm potentially giving them a competitive advantage over rival firms.5 On the other hand, the members of the public that allege to be unfairly impacted by an algorithmic decision may wish to understand how the decision was made, requiring some level of transparency, including potential rights to examine the data involved in the decision-making as well as the decision-making process itself. The law must strike a delicate balance between these two opposing expectations.

On several occasions where disclosure of information regarding algorithmic decisions was demanded, trade secret has been invoked as a defence to limit disclosure.6 This has caused concerns among scholars, advocacy groups and policy makers, as the combination of opaque algorithmic decision systems and trade secret law might lead to relevant actors making decisions for which they are not held accountable.

In the European Union (EU), the conflict between transparency in ADM and trade secret protection has been addressed by the General Data Protection Regulation (GDPR)7 and the Trade Secret Directive.8 Recently, the EU Artificial Intelligence (AI) Act and the now-withdrawn Draft AI Liability Directive have introduced some provisions dedicated to trade secret protection in the context of AI. However, these recent legal developments have not brought about significant changes in addressing the concerns raised in academic literature and policy papers. In the USA, both the legal basis for trade secret protection and the rule for balancing disclosure and secrecy have largely been established by state and case laws.9 The US Congress passed the first federal trade secret legislation, the Defend Trade Secret Act (DTSA), in 2016.10 Neither the EU’s relatively tailored system, nor the US’s sparse legal regime contain hard and fast rules on how judges could reconcile the conflict between transparency in ADM and trade secret protection of proprietary information in concrete cases.

There are two primary problems with the existing literature. The first one is the call for abolishing trade protection for algorithms that impact individuals and the public.11 Such a call, unsupported by a robust analysis of the existing legal frameworks, is unrealistic. Second, while moderate positions recognised the need for a balancing exercise, they fail to provide guidance on how such exercise can be done in practical terms.12 This article first articulates the conflict between the demand for transparency and maintaining trade secret protection for algorithms. More importantly, it argues for the first time that the existing legal literature that advocate for the abolition of trade secrecy for certain ADM is erroneous due to the lack of strong legal reasoning justifying such a prescription. To that effect, the article analysis how balancing exercises introduced by courts in the USA can aid in applying the existing laws to allow the enforcement of trade secret protection in such a way that the individual rights and public interests are also protected. The article also analyses why patent protection for algorithms, especially in the age of AI, is an unattractive method of protection, giving businesses little incentive to pursue it as an alternative. Thus, the call for removal of trade secret protection for algorithms or for its replacement by patent protection is neither justifiable, nor necessary.

The remaining part of the article is divided into three sections. Section 2 provides a brief overview of the preference for trade secrecy over other IP rights for software. Section 3 analyses the conflict between trade secrecy and algorithmic transparency by focusing on ADM under the GDPR. It also discusses judicial decisions from the USA to exemplify how judges have tried to balance the two interests in concrete cases. Section 4 investigates the possibility and feasibility of mandating patent protection for computer programmes including AI, as this is one of the solutions presented as an alternative in the existing literature. It discusses the economic arguments against patenting computer programmes as well the legal challenges to obtaining patents for them. This section highlights that the combination of legal and practical challenges makes patenting computer programmes a difficult choice, strongly supporting the argument that businesses should not be forced to make that choice by law. Section 5 provides concluding remarks.

Source: https://academic.oup.com/jiplp/advance-article/doi/10.1093/jiplp/jpaf038/8216795