Privacy enhancing technologies ensure that the security risks of processing customer data remain manageable, your customer data is protected against old and new threat scenarios in compliance with the law, and your data use initiatives stay on track.

In the first part of this article, we noted that data protection laws are getting stricter and customers are less likely to suffer a data breach. There is also more pressure to do more with data and the tools we have used for years in data protection have many weaknesses. Therefore, their use is very limited. In the second part, we will get to know some innovative technologies that will enable us to deal with customer data in an active and legally compliant way in the future.

Data protection – being in the middle instead of just being there

Data protection has always been a highly interdisciplinary activity. The bouquet of coordinated contractual, organisational and technological measures has ensured that data protection risks are reduced to a minimum. Whereas in the past, technological measures tended to play a subordinate role in this mix, and legal departments were in the lead when it came to data protection, the situation is changing with the rapid spread of privacy-by-design principles. This approach, whose origins can be found in the 1995 report on Privacy Enhancing Technologies[1], assumes that privacy is best respected when it is technically integrated from the outset in the development of products, services and the supporting processes. One of the major challenges in implementing privacy-by-design principles over the past 25 years has been the lack of market-ready technologies, which:

• would guarantee complete anonymisation of the data and the results of the data processing, or
• would provide strong protection of the data during processing.

These technologies are now market-ready and available either as commercial solutions or as open source frameworks.

The light at the end of the tunnel

In the last 5 years, important technological breakthroughs have been made that allow us to effectively solve the problems mentioned above.These new technologies are often called Privacy Enhancing Technologies (PETs)[2]. PETs comprise a set of hardware or software solutions designed to enable privacy-compliant data processing without compromising the privacy and security of the data itself. In doing so, PETs eliminate the two biggest weaknesses of traditional data protection tools:

• the need to have trust in the party processing the data; and
• the risk that clients can be re-identified from alienated data.

The main PETs that enable complete anonymisation of data or anonymisation of the results of a data analysis are:

• AI-generated synthetic data
• Today, it is becoming increasingly easy to re-identify customers whose data has been alienated through classical anonymisation and pseudonymisation. Synthetic data generated by an artificial intelligence maintains the statistical properties and usability of the original data and can in many cases replace the real customer data. The synthetic data generator first trains with “real world” data and then generates synthetic data without direct connection to the individual data points of the original data. AI-generated data is therefore completely anonymous and its use does not fall under data protection legislation.Differential Privacy

Differential Privacy is a criterion that measures how much a result of a calculation reveals about individual input values. Put more simply, differential privacy allows us to say whether the result of a data processing operation can be used to determine whether or not a person was preserved in the data set used for that purpose. Differential privacy ensures that an outside observer who sees the result of a calculation is not able to recognise whether a certain individual data record (e.g. a customer) was used as an input value, regardless of which other data records are available to the observer. The most important PETs that enable data processing with original data – even in situations where one does not have full confidence in the parties processing the data – are:

• Secure Multiparty Computation

Normally, data collaboration requires that the data be physically assembled in a central location. Secure multiparty computation is a cryptographic technology that allows different parties to analyse data together without sharing it. This technology promises to play a key role in creating multi-stakeholder ecosystems that add value from the decentralised analysis of shared data sets without the need to share the data.

• Homomorphic Encryption

Conventional encryption methods make it impossible to use data while it is encrypted. Homomorphic Encryption (HE), a software-based method, on the other hand, allows third parties to process and in some cases even manipulate encrypted data without seeing the underlying data in an unencrypted content format. Therefore, data secured with HE can remain confidential while being processed, allowing useful tasks to be performed on the data in untrusted environments.

• Confidential Computing

Use of a third-party hosted service, increases data security risks. Confidential computing is a hardware-based method that protects data during use by performing computations in a Trusted Execution Environment or TEE. These secure and isolated environments prevent unauthorised access to or modification of applications and data during use. Which is the right PET for your organisation? This depends largely on what data protection problem you are most concerned about solving. Do you want to initiate data collaboration with third parties or rid your testing process of production data? Some questions you need to answer on the way to your first PET are:

• What kind of data processing do you want to perform?
• What personal data do you need?
• What is the context of the data processing? Who is processing the data? Where? How? For what purpose?

Many PETs described above will soon be part of a company’s standard technology package. They will help us to establish privacy-friendly data use in the company and privacy-by-design principles in the company. However, they will never become a one-size-fits-all solution to all data protection problems. Data protection will remain an interdisciplinary discipline in the future, which will certainly rely much more on technological data protection solutions such as PETs, but will not be able to do without compatible and organisational measures.

This article is part 2 after part 1 How companies use our data, but also protect it, which was about the challenges companies face.

A CAS for the use of AI in companies

How to exploit the potential of artificial intelligence in your own company is taught in the second edition of the CAS AI for Business from BFH Wirtschaft. The course offers a strongly practice-oriented introduction to Artificial Intelligence (AI) and Machine Learning (ML) for managers without technical knowledge. It supports specialists and managers from business, administration and non-profit organisations in better assessing the potential of AI and communicating effectively with data scientists. More information and registration.

References

[1] Hes, R. & Borking, John. (1995). Privacy-Enhancing Technologies: The Path to Anonymity. [2] Digital Banking Blind Spot – PETs (Mobey Forum 2021)