Different definitions of trade secrets exist in various jurisdictions around the world. However, they share some common elements as enumerated below:

1. It is not generally known among or ‘readily accessible to circles that generally deal with the kind of information in question.’¹ Absolute secrecy is therefore not required.

2. They have ‘actual or potential commercial value attributed to their secrecy.’²

3. They should be subject to ‘reasonable steps taken by the rightful holder of the information to keep it secret.’³

Therefore any information that has secrecy, commercial value and reasonable preventive steps qualifies as trade secrets.⁴ AI components which meet the threshold above can therefore be protected by trade secrets. This includes algorithms, training datasets, software codes, machine learning methods, AI generated outputs and other components.⁵ Trade secrets are best suited for algorithm protection and since they are mathematical instructions, they are ineligible for patent and copyright protections and therefore are best protected by trade secrets which will keep other people from using algorithms created by others.⁶

Technological know-how is described as a ‘licence of confidential information.’⁷ It has also been defined by the EU competition law as “a package of practical information obtained from experience and testing which is secret, substantial and identified.”⁸ It is also best suited for protection by trade secrets since it involves how the AI innovation works which is considered as valuable business information.⁹ An innovator may gain competitive edge over other competitors if the means of getting the AI to work or how he implements machine learning is better than other competitors. In order to prevent others from taking advantage of the know-how, an innovator may decide to protect the know-how behind the AI innovation by keeping the knowledge secret.

Copyright protection has its own limitations for instance for AI to be protected, it must meet the threshold of originality.¹⁰ Patents also may not be compatible for AI developers because they are ‘unavailable for certain facets of AI’¹¹ such as data compilations like AI training sets and a ‘programmer’s particular expression of source code.’¹²Therefore, trade secrets are the most desirable tool for AI protection.¹³Compared to patent protection, trade secret protection has its own benefits thus being preferred by AI developers. Some of the advantages of using trade secret protection are:

1. Threshold for protection

Compared to patent law, there is no stringent and costly application process for trade secrecy.¹⁴ Trade secrecy requirements are also regarded as less demanding compared to patent law novelty and non-obviousness standards.¹⁵ When it comes to trade secrets, several parties can maintain the same trade secrets simultaneously while in patents multiple parties cannot maintain patents on the same invention.¹⁶

In some jurisdictions like the United States, the scope of what parties can protect as trade secrets has increased because of changes in federal law, and thereby providing AI developers with a broader scope of protection unlike when using patent protection. The advantage of this is that legal protection can be provided to parts of AI systems that are outside the scope of patent protection such as training data that is relied upon by many AI systems.¹⁷ Additionally, even for AI systems that are patentable, the wide scope provided by trade secret protection encompasses ‘many [more] elements of AI systems than patent law.’¹⁸

2. Requirements on (Non) Disclosure

Unlike patent protection, trade secret protection requires that parties maintain secrecy for their inventions.¹⁹ Failure to do so eliminates trade secret protection.²⁰ Therefore this is best suited for AI protection since AI innovators ‘derive significant competitive advantages by maintaining the secrecy of their systems.’²¹

3. Duration

Another advantage of using trade secret protection for AI compared to patent protection is in the duration of protection offered. Patent protection is limited to twenty years from the date of filing. On the other hand, trade secrets are not subject to statutory term limits²² as long as the information ‘remains valuable, is not generally known or readily ascertainable and the trade secret owner continues to employ reasonable precautions in protecting the information.’²³An example of this would be the owners of Coke who have protected their recipe as a trade secret for around 100 years.²⁴ The long duration provided by trade secrecy may seem more attractive to parties dealing with AI especially if they desire long term protection for their inventions.

4. AI technologies difficult to reverse engineer or do an Independent discovery

Where the innovation would be difficult to reverse engineer, the technology would be well suited for trade secret protection. This is because the value and attractiveness of trade secrets increases when the protected technology does not allow reverse engineering to take place or independent discovery.²⁵ The amounts of input data such as testing data and validation data determine eventually how the AI output will be shaped and therefore making an independent discovery would be impracticable.²⁶ This is also applicable to reverse engineering since the combination of input data and the categorization of the data sets involved is not likely to be reverse engineered.²⁷

Also, the complexity of algorithms involved in the AI technologies make independent discovery and reverse engineering highly unlikely. An illustration of this is the case of ClearOne Communications v Bowers²⁸ where it was held that since the development of trade secret protected acoustic echo cancellation software entails several algorithmic and programming choices, it is hard or impossible for two people working independently to come up with a similar algorithm.²⁹

By keeping the technology a trade secret, the inventor would be preventing others from having access to and eventually using the technology.³⁰ The innovations may also include the know-how and method for extracting appropriate information from raw data sets and eventually generating models using the information that has been extracted.³¹ It also includes methods and know-how ‘for training and utilizing models.’³²

5. Low barrier to entry

The advantage of using trade secret protection for AI innovations is the fact that trade secret protection comes into effect immediately and they need not be registered at all. This is because the registration would alert others and ‘provide for public disclosure.’³³ Patents on the other hand require time and money to ‘prosecute and obtain.’³⁴ The benefit of this for AI innovators is that unlike patents, which are costly because of the fees involved for instance filing fees, trade secret protection does not require ‘hefty preparation or filing fees.’³⁵

Challenges of Trade Secrets protection for AI

1. Trade Secrets theft

In the United States, studies that have been conducted have shown that companies have incurred losses in billions of dollars because of trade secrets theft.³⁶One of the most dangerous means of conducting trade secrets theft is through cyber-attacks.³⁷ Considering that cyberspace use has increased over the last couple of years, cyber-attacks have also increased and without proper cybersecurity measures in place, it becomes hard to detect.³⁸

2. Limited protection provided by Trade Secrets Law

Trade Secrets do not prevent ‘independent discovery or reverse engineering of technology.’³⁹ Companies can implement precautionary measures so as to provide data secrecy protection but they can’t prevent others from ‘legitimately reverse engineering inventions represented by protected information.’ Therefore, in cases where there is a high possibility of discovering the company’s secret information, the AI inventor should use other kinds of protection such as patents.⁴⁰

3. Difficulty in enforcing rights and proving Trade Secret infringement

Trade secrets pose a number of challenges that AI innovators must take into consideration when protecting their innovations. These include:

1. ‘There is need for sufficient proof that the alleged trade secret infringed is actually ‘a trade secret especially when enforcing the right.’⁴¹

2. There is no formal registration of trade secrets with government authorities and therefore companies lack a written document they can rely on. This means that they should sufficiently prove that their trade secret meets the requirements for defining trade secrets in the relevant jurisdiction.⁴²

3. There is a high evidentiary burden on companies to prove misappropriation. This is the case especially in countries where ‘sufficient discovery resources are lacking.’⁴³

4. The ‘technical complexity and evolving nature of AI systems make proving infringement difficult.’⁴⁴

Technology is advancing at a very fast pace and Artificial Intelligence has become one of the important pillars of the modern economy. This is because companies have discovered the need of using Artificial Intelligence to be globally competitive with other companies. As illustrated, trade secrets are ideal in the protection of AI. However, as the advancement of AI takes shape every day it is important to ponder these questions: What will be the impact on trade secrets if an AI can perform an independent discovery or reverse engineer? Can trade secret protection of AI hinder innovation?

Image used is from knowledgewebcasts.com