The global communications agency Ogilvy is currently in the middle of a project that combines Robotic Process Automation (RPA) with Microsoft’s Vision AI to solve a business problem. Yuri Aguiar, Chief Innovation and Transformation Officer, is already thinking about how he can protect the resulting algorithms and processes from theft: “I doubt that this is a patent in the classic sense – but the algorithms give us a competitive advantage and reduce the time-to-market considerably. I consider algorithms as modern software modules and that’s exactly how they should be protected. “
The theft of intellectual property (IP) has now become a real problem for global companies: In February 2020, the FBI identified around 1,000 cases of technology theft – in connection with China alone. It is not just governments and secret services that are looking for intellectual property: direct competitors, partners or their own employees also use intellectual property.
It is part of the routine of security teams to take appropriate measures to protect intellectual property such as software, design concepts or marketing plans. But what if the intellectual property is not a document or a database, but an algorithm?
For years, legal experts have argued that algorithms are not patentable because they only represent mathematical methods and are therefore not covered by patent law. In the age of AI and machine learning, however, algorithms ensure that software “learns” from results – without the intervention of a developer. In this way, algorithms can open up competitive advantages and no longer serve mere data processing, but solve technical problems. The United States Patent Office has recently revised its guidelines on patent protection for algorithms and is now making it much easier to obtain such a patent. However, there are downsides to this: after all, there is nothing stopping the competition from developing a new algorithm that leads through the same steps.
Even under German law, there may be the possibility of patenting an algorithm – if it is used as part of a control unit for a complex, technical machine and is “protected” in this context. However, this is rather an exception. Copyright, on the other hand, protects computer programs, but not the algorithm itself.
Many companies therefore choose to treat algorithms as business or company secrets and to protect them accordingly. This does not require an application and no transparency – on the contrary: confidentiality comes first. Already with the conception.
In the following we have summarized the individual measures for you in order to best protect company secrets and intellectual property in algorithm form:
Zero trust approach
The best practices for IT-side protection of algorithms are based on the principles of the zero trust approach. Algorithms that are business or company secrets should be kept in a virtual safe. Access should be given to as few people as possible and if only with the absolutely necessary privileges to perform their tasks. In addition, the use of two-factor authentication as well as permanent logging and monitoring of all access processes is recommended.
Organizations should ensure that everyone who has access to the algorithm or trade secrets signs a confidentiality agreement.
Think carefully about who actually needs to know how much about the algorithm or other trade secrets. In larger companies in particular, it is often an advantage if fewer people know about trade secrets. Even if you only initiate small teams, it also makes sense here to rely on two-factor authentication and process guidelines, for example when it comes to remote work or the use of external storage media.
Algorithm protection meets specialist areas
IT decision-makers should also raise awareness in the specialist areas of what is required to comprehensively protect a trade secret such as an algorithm. For example, sales experts should know exactly which parts of a product are to be treated confidentially.
In any case, employees leaving the company should be prevented from taking business secrets away from the competition. It is advisable to conduct an exit interview, in the context of which it is once again possible to expressly refer to the contractual obligations and confidentiality agreements.
Preservation of evidence
Despite all precautionary measures: If someone is targeting your algorithm (or some other trade secret), sooner or later he will find a way to access it. Then it is important that you can prove that it is your property. This works in the form of digital watermarks, for example: IBM and other companies are researching ways to anchor such watermarks in deep neural networks. A problem with this method is that third parties may be able to smuggle their own watermark into models that have already been given a different one. For this reason, a team of researchers from the University of Chicago introduced the so-called “zero embedding” process in February 2020, which can be used to anchor forgery-proof, digital watermarks in deep neural networks. However, the concept is in an early development phase. (fm)
This article is based on an article from our US sister publication CSO Online.