Australia - Got A Secret? Can You Keep It? Using Trade Secrets To Protect AI.
Legal News & Analysis - Asia Pacific - Australia - Intellectual Property
29 May 2020
In this fourth article of our series on artificial intelligence (AI) and intellectual property (IP), we consider the use of trade secrets to protect AI and the unique challenges to trade secret protection presented by AI.
The recipe for Coca Cola, KFC's 11 secret herbs and spices and Google's search algorithm are all trade secrets which have been vehemently guarded by the companies which rely on them.
However, to gain a legally enforceable patent monopoly over an AI system or AI-created invention, companies face an expensive and potentially uncertain patent application process. As AI becomes an increasingly important business asset, are companies considering protecting their AI by treating them as trade secrets rather than applying for patent protection?
In the technology sector, innovation is an important source of competitive advantage, which drives short product lifecycles. To keep pace, AI products are often built upon existing technology using open-source (i.e., freely available) code. When developed, AI products are scaled and distributed widely. These factors, together with high employee mobility and extensive use of consultants at technology companies, presents a unique set of challenges to trade secret protection for AI and AI inventions.
In this article, we examine the unique challenges that AI systems present to securing and maintaining trade secret protection.
What is a trade secret and how is it protected in Australia?
Australia has no common law or statutory cause of action directed specifically to the protection of trade secrets. Instead, Australia complies with its obligations under article 39 of the 1995 Trade-Related Aspects of Intellectual Property Rights Agreement (TRIPS) by giving trade secrets legal protection through the equitable doctrine of "breach of confidence" and, where an appropriate clause exists, breach of contract.
A "trade secret" is a category of information that has a quality of intrinsic confidentiality, the secrecy of which is maintained by its owner. Protection of trade secrets is achieved through actions for breach of confidence or, if a non-disclosure agreement or confidentiality clause exists, breach of contract.
An action for breach of confidence requires that:
the information is confidential;
the information has been imparted in circumstances importing an obligation of confidence; and
there has been unauthorised use of the information to the detriment of the person imparting it. 1
An action for breach of confidence is distinguishable from a confidentiality clause within a contract. A confidentiality clause is a contractual obligation to keep information which is expressly identified confidential.
Similar to an action for breach of confidence, a confidentiality clause will not, typically, extend to information which has been made public (other than through a breach of the clause). However, unlike an action for breach of confidence, while the recipient must be aware of the obligation (by express wording in the clause), there is no requirement for damage.
Whether by way of breach of confidence or a contractual provision, trade secret protection is afforded to a broad range of subject matter including both technical and non-technical information. Moreover, there is no originality, distinctiveness, novelty or inventiveness requirement. There is also no application or registration (nor related fee(s)) required. Importantly, trade secret protection through either means can last forever, so long as the secrecy is maintained and the information is not publicly disclosed.
As is clear from the above, the confidentiality of the trade secret is imperative to a successful breach of confidence and/or contract cause of action. Therefore, companies must act very quickly when they suspect there has been any potential unauthorised access to their information. While a company may be able to seek damages for an unauthorised breach, if the information is made public, it will not be protectable going forward.
Is trade secret protection suitable for AI?
Trade secret protection has been afforded to confidential information contained within software. 2 With this protection, the owner of the software may enjoin a third party from copying or using the source code to create new software or to use the information as a springboard for development. 3 It follows that the same protection would be given to other elements of AI (for example, the source code, the training data and the learning algorithms) provided that they are kept confidential and cannot be reverse-engineered.
Additionally, AI may be particularly well suited to trade secret protection given some of its unique development and lifecycle features which we discuss further below.
Rapid AI lifecycles
In the technology sector, innovation is an important source of competitive advantage. The constant drive to innovate leads to efficiency, productivity and increasingly short product lifecycles.
When products are launched, there is often quick, initial uptake by the market. However, as competitive products are released and improvements made, uptake of the original product can quickly taper-out. This trend motivates companies to leverage current technologies to produce innovations before a product moves to market maturity and eventual decline.
The rapid pace of product development and improvement in the technology industry is arguably out of step with the patent system. According to IP Australia, it can take anywhere between 6 months to several years for a patent to be granted (depending on the circumstances). While a granted patent enables a lengthy monopoly, if a technology is likely to become obsolete within a few years and cannot be reverse-engineered, companies may not be willing to expend the time and money seeking a patent.
Uncertainty of patent protection
While traditional IP protection may protect different aspects of AI (for example, copyright protection of the source code as a literary work), recent decisions from the Federal Court in Australia cast doubt on the patentability of computer-implemented inventions.
The difficulty in securing patent protection of computer-implemented inventions was highlighted in the recent decision of Encompass Corporation v Infotrack (see our article here).
Given the uncertainty for companies considering patent protection of computer-implemented inventions, companies may not be willing to undergo the extensive patent application disclosure requirements that risk publicly disclosing their innovations when the likelihood of a patent being granted is uncertain. These circumstances may make trade secret protection a more appealing alternative.
Challenges of trade secret protection for AI
Notwithstanding the advantages of trade secret protection, there are also unique challenges that AI presents.
Use of open-source software
Open-source software is a critical component in technology development. Open-source software is publicly available source code provided online pursuant to a licence. The licence commonly allows users to view and alter the source code and to distribute derivative works.
Due to the rapid development environment in which AI is created, developers regularly rely on open-source software to program fundamental components of their technology, for example, the operating system and the application program interface. Developers can then focus on the core, innovative aspects of their technology.
However, using open-source software presents risks. There are different types of open-source software licences; some are highly permissive, for example, the MIT License, 4 which imposes very few, if any, restrictions on the legal use of the software and allows for the software to be re-licenced as proprietary software. Other open-source software licences are more restrictive, for example, the GNU General Public License, 5 which is a "copyleft" licence that requires any derivative work to be distributed under the same or equivalent open-source licence terms.
It is possible to keep a company's proprietary source code confidential while also using open-source software but a company must be careful to adhere to open-source licence terms. Therefore, it is important to be aware of the terms of any open-source software before this software is used to ensure that the company is not required to release their proprietary code to the public under the same or equivalent open-source licence terms thereby destroying the confidentiality of the proprietary code.
Companies should maintain, and continually review, their open-source software use policy. It is also important to restrict the use of open-source software that is licenced under restrictive terms, such as the GNU General Public Licence. This will enable companies to continue to use open-source code where useful (for example, utilising open-source analytics tools) but enable the companies to maintain secrecy over elements of their systems that are unique.
Wide distribution and reverse-engineering
When any technology is placed on the market, there is a risk it may be reverse-engineered. This is because new technology is often, and almost immediately, studied and analysed by competitors. While, at present, it may be difficult to reverse-engineer complex AI, this may not be the case in future. In circumstances where AI can be reverse-engineered, treating it as a trade secret is redundant.
Where the software or an AI system is made available to the public, this overrules any right to confidentiality and anyone who can break the encryption could reverse-engineer the software. 6 As a result, any trade secret protection over the software is lost (although there may be other causes of action against individuals who deliberately circumvent encryption systems, for example, under the Copyright Act 1968 (Cth)).
High mobility & atypical employment relationships
The technology sector's high mobility makes for a particularly susceptible environment for ex-employees or consultants to steal trade secrets.
However, without evidence of an ex-employee inappropriately exceeding their authorised use of a past employer's AI (for example, by actively copying information prior to resigning), it would be difficult to determine whether a competitor's AI or software was created using a past employer's confidential information or simply an ex-employee's skill or know how.
Similarly, the situation can be unclear where the AI or software is created by independent contractors but tailored for the benefit of the employer. Both parties may claim ownership to different parts of the software and, without a clear agreement, there are likely to be disputes regarding whether contractors can use what they have created for other clients (such as competitors).
Given the high employee mobility and atypical employment relationships associated with the technology sector, companies should consider proactive measures (for example, executing non-disclosure agreements, consulting contracts, implementing "need to know" policies and incorporating restrictive covenants in employment contracts) to prevent the inadvertent or deliberate disclosure of valuable trade secrets.
A different approach – the US
In Australia, breach of confidence litigation to protect trade secrets is rare. By contrast, trade secret litigation is on the rise in the US. The passing of the Defend Trade Secrets Act (DTSA) has resulted in several recent cases alleging misappropriation of trade secrets protecting technologies.
The DTSA is a reaction to reports that trade secret theft in the US was reaching unprecedented levels. The DTSA creates a cause of action for trade secret misappropriation. Since its commencement in 2016, the DTSA has reinvigorated the law of trade secrets in the US leading to a 24% increase in trade secret litigation. By contrast, other IP litigation has remained steady or slightly declined during the same period.
The broad reach of the DTSA was recently demonstrated by a US District Court decision in favour of Motorola Solutions against Hytera. After determining that the DTSA applies 7 to acts conducted outside the US, so long as there is a requisite nexus to the US, Motorola was awarded US$746.6 million in punitive and compensatory damages after a jury found that 3 former Motorola employees had taken technical documents to develop digital radios at Hytera.
Without equivalent legislation with such a broad reach in Australia, damages of this scale are unlikely to be awarded by an Australian Court. However, if companies do increase treating AI as trade secrets, breach of confidence, and potentially also breach of contract, litigation is likely to increase.
Actions for breach of confidence in Australia are, arguably, under-utilised and may prove a useful avenue to protect a company's key business assets. While AI certainly poses several unique challenges, if appropriate measures are taken, a company could effectively use trade secrets to protect the confidential elements of their AI systems.
Companies developing AI should consider a holistic approach to their IP protection and consider factors such as the likely lifespan of the technology, the value of the AI, and its importance to the business. IP protection is complementary, so utilising patents, copyright and trade secrets to protect different aspects of AI is likely to be the most effective strategy.
For further information, please contact:
Nina Fitzgerald, Partner, Ashurst
Australian Medic-Care Co ltd v Hamilton Pharmaceutical Pty Ltd (2009) 261 ALR 501 at .
Ibcos Computers Ltd v Barclays Mercantile Highland Finance Ltd (1994) 29 IPR 25 at 39.
See Fraser v Evans  1 QB 349, see also Admar Computers Pty Ltd v Ezy Systems Pty Ltd (1997) 38 IPR 659.
The MIT License is a software licence which grants permission, free of charge, to any person obtaining a copy of the licenced software to deal with the software without restriction. The MIT License pertains only to the licenced software (not derivative works) and is subject only to the condition that a copyright notice and permission notice be included in all copies or substantial portions of the licenced software.
The GNU General Public License is a software licence which grants permission, free of charge, to any person obtaining a copy of the licenced software to redistribute and/or modify the licenced software. The GNU General Public pertains to the licenced software and all derivative works and is subject to the condition that, when distributing or publishing derivative work, the derivative work must be licensed as a whole, free of charge, to all third parties under the same terms of the licence (among other conditions).
Mars UK Ltd v Teknowledge Ltd  All ER (D) 600; (1999) 46 IPR 248;  FSR 138.
Pub.L. 114–153, 130 Stat. 376, enacted May 11, 2016, codified at 18 U.S.C. § 1836, et seq.