Privacy Enhancing Technologies (PETs) are commanding the attention of the international data innovation community, promising a step-change in what is possible in data collaboration by mathematically solving longstanding data collaboration challenges. They are becoming increasingly mainstream in the AdTech and mobile hardware industries, drawing interest from prominent institutions such as the Royal Society, the ICO, and the Centre for Data Ethics and Innovation (CDEI). 

Yet, government has been slow to recognise the value of PETs.

Earlier this month, PETs use cases for government entered the spotlight with the announcement of the UK-US Privacy Enhancing Technologies Prize Challenge, which calls for PETs innovators to develop solutions for financial crime detection and disease surveillance.

As PETs providers prepare their solutions and the visibility of government PETs use cases increases, this blog shares PUBLIC’s views on practical steps to take government PETs solutions from the lab to the real world. Our insights are based on PUBLIC’s extensive research on PETs for the public sector and our engagement with numerous PETs providers, including Bitfount, Gradient0, and Duality. 

Tackle a whole data collaboration problem

The first step in the PETs journey is developing a clear understanding of why a problem requires collaborating with sensitive data. To make the business case for a new class of technologies, such as PETs, public sector data owners must show not only how the technology makes an incremental improvement – but how it can unlock significant value by opening up a new way of using data entirely. 

Based on PUBLIC’s work delivering data strategy and transformation to government clients, we have identified four key use cases: 

• Cross-departmental data access requires two or more organisations to access each other’s data when differences in organisational jurisdictions pose security, regulatory, or ethical risks. For example, UK and European police forces may need to coordinate responses to sophisticated, cross-border threats by sharing real-time intelligence data, and could use homomorphic encryption to enable rapid threat intelligence. 
• Central oversight of disparate entities requires a ‘central’ source of truth based on de-centralised data, a typical requirement of national government departments with an oversight function over many local organisations (e.g., Home Office’s relationship to local police forces), or regulatory authorities (e.g., the Financial Conduct Authority’s remit over regulated financial institutions). Regulated entities, for instance, could use a Trusted Execution Environment to allow their regulators to access sensitive data for compliance auditing, without sacrificing data controllership. 
• Monitoring risks to vulnerable populations requires data from a variety of sources to predict adverse outcomes and deliver interventions. For instance, a federated learning system could enable a service delivery organisation to deliver support to older adults in social care who are at risk of falling by analysing data from the nursing home, the NHS, and the local authority – all without any party seeing the others’ underlying dataset.
• Safely opening sensitive data assets reduces the risks of publishing data by reducing the impact of potential re-identification. In order to reduce the privacy risks of its statistical products, the U.S. Census Bureau used differential privacy techniques to add sufficient ‘noise’ to prevent re-identification while preserving much of the analytic utility for researchers.

Each technology has different strengths and limitations, and should be seen as an important layer of your organisation’s evolving “privacy stack” – a set of foundational capabilities for managing data privacy risk. 

Conduct a Discovery 

Once a high-level use case is identified, data owners should start with a rapid Discovery to baseline their own data capabilities, privacy posture, and benefits of PETs solutions. This can help simplify future PETs procurement, enable data owners to be smarter customers and ease supplier onboarding.

A Discovery should answer several strategic questions, while educating and exciting stakeholders about the potential of PETs solutions in the process. Key strategic questions include:

• Outcome: What is the end goal?
• Stakeholders: Are the users internal or external? 
• User Persona: Who are the all of the end users of the solution e.g., data scientists, business users, senior leadership? What kind of data does each need to see?

A Discovery should also map current and potential future risks: determining severity and likelihood to these is key to justifying why PETs, rather than more commonplace solutions, are needed to get the job done:

• Data security risks: Can the data “leave”? What security measures are required?
• Data privacy risks: What effect will the proposed solution have on the data subject? Would the data subject be comfortable with their data being processed this way?

Finally, questions examining data infrastructure and datasets can help organisations to diagnose which PETs techniques are most appropriate for the problem:

• Data Structure: Is the data structured or unstructured? Has it already been assessed for sensitive elements? Where is it stored? How will it be prepared?
• Data Volume: How big are the relevant datasets? Are there thousands, millions, or billions of records?
• Data Fidelity: What is an acceptable tradeoff between data accuracy and confidentiality of data subjects? Is synthetic data or raw data needed? 
• Sophistication: Do users need to run basic SQL queries, or more advanced ML models?

At PUBLIC, we have found that the biggest challenge facing public sector data transformation projects is aligning diverse stakeholders on an acceptable risk-benefit trade-off. As a result, the Discovery process should also convene and empower a centralised team with remit over data governance and security to own the project going forward. 

Select a reliable supplier

In a nascent technology segment such as PETs, it can be difficult to tell which suppliers have the credentials to deliver. There are a few rules of thumb to be a smart, critical customer of PETs solutions. 

• Data controls: If the provider’s architecture allows data to be stored in a local location, and controlled from another, they will lend themselves to a more seamless, policy-driven UX and will scale faster than other providers.
• Open Source: Engaging vendors with open source solutions can help your organisation’s security team examine and experiment with solutions to ensure they are ‘quantum-safe,’ while indicating the maturity of a PETs technique. 

Similarly, it is worth being aware of a few red flags when evaluating PETs suppliers. 

• AI all over their marketing: PETs typically are leveraged to enable privacy-preserving AI/ML, but be wary if there are frequent mentions of AI as the solution on how they do things without further explanation of how AI improves data privacy. 
• Claims of scale: PETs are getting faster all the time and many PETs providers are innovating to cut down on computation time. Make sure they are not cutting corners on privacy in doing so. If they claim they can process billions of records with cryptographic methods in minutes, they are not being 100% transparent with you.

Conclusion

The Government Digital Service (GDS) standard requires new digital services to solve a “whole problem” for users: PUBLIC believes this principle applies to internal data analytics use cases as well. This means that the new user journey enabled by a PETs solution must alleviate a real pain point and make intuitive sense to the user, regardless of the technology that is powering it. PUBLIC helps public sector organisations both build the right solution and build the solution right to ensure it actually ‘sticks’ beyond implementation. 

Unleashing the power of PETs and unlocking new data collaboration opportunities doesn’t require a leap of faith. PUBLIC can help your organisation on its journey to build trust in this powerful new set of technologies. Get in touch with PUBLIC’s Privacy, Security, and Online Safety (PSOS) and Data Services’ teams to find out how to get a Discovery for your data collaboration project off the ground. Get in touch with Daniel Fitter to find out more.

‍