Privacy Enhancing Technologies (PETs) promise something that for a long time seemed difficult to reconcile: extracting value from sensitive data without revealing sensitive information. The technologies have existed for years and are increasingly moving from research into practice. Yet despite successful implementations and pilots in areas such as healthcare, public administration and artificial intelligence, large-scale adoption has yet to materialize.
According to Sarah van Drumpt, researcher and consultant at CoE-DSC/TNO, and Dave Buckley, technology policy consultant at the OpenMined Foundation and co-chair of the UN PET Lab, who both took the stage at the Data Sharing Festival, the challenge is shifting. The question is increasingly not whether the technologies work, but whether the surrounding conditions — governance, legal certainty, institutional readiness — exist for organisations to adopt them.
From promising technology to scalable deployment
PETs are not one particular technology, but an umbrella term for different approaches to protecting sensitive data while enabling it to be used or shared. These approaches can rely on statistical methods, such as synthetic data and differential privacy, cryptographic techniques, such as secret sharing, homomorphic encryption and zero-knowledge proofs, or secure hardware, such as Trusted Execution Environments. Multi-party computation is a broader concept that can make use of cryptographic techniques such as secret sharing or homomorphic encryption. “Every flavour has its benefits,” Sarah explains. “Synthetic data serves a different purpose than multi-party computation. One allows you to create a dataset that is no longer sensitive and can be stored for longer. The other enables you to perform computations on encrypted data so that sensitive information in the raw data remains confidential. Your best choice of PETs depends on the purpose you want to achieve.”
Dave believes the underlying promise of PETs is best understood by comparing them with more traditional approaches to data protection. “If you think about more traditional ways of protecting data, you often pseudonymise or anonymise datasets by essentially removing information from them,” he says. “But by removing information, you necessarily reduce their potential utility. The promise of PETs is not that this privacy-utility trade-off disappears, but that it can be managed in a far more precise, context-specific way, enabling organisations to make use of data whilst protecting privacy.”
From trust to assurance
Not every technology is equally mature. But there are enough viable technologies available today to shift attention towards the next question: how can they be combined into systems that organisations can actually deploy and that fit their individual needs? That transition towards scale introduces a different challenge. Organisations need to be confident that the technology not only works, but that data is processed correctly, agreements are respected and regulatory requirements are met.
For Sarah, trust may not even be the most useful word to describe what is needed. “I actually like the word assurance a bit more,” she says. “Trust is something quite personal. Every organisation or person will have to decide for themselves whether they trust something enough to act upon it. What you can do is provide assurances. These provide a tangible basis for this trust, but can also prove that other ethical, legal, and societal requirements are met.”
Technology itself can increasingly provide part of that assurance. Zero-knowledge proofs, for example, make it possible to prove something without revealing any of the information behind that claim. Sarah sees growing interest in precisely this type of application, particularly around what she describes as Automated Compliance: using technology not only to comply with laws and regulations, but also to demonstrate that compliance without creating an additional administrative burden. “You don’t want to have to share everything to prove that your claim is right. I’m noticing more and more interest in that, specifically when it relates to compliance. It’s a way of creating trust because you can demonstrate that someone complies with the rules without having to disclose any of the underlying information.”
Dave sees the same development. At OpenMined, this is approached through the concept of structured transparency. “We frame a lot of our work around this concept of structured transparency,” he explains. “It’s a framework for constructing information flows to meet the privacy aims of a specific context. It enables much finer control over how information is accessed and processed, moving away from the binary idea that data has to be either open or closed.”
This also changes the nature of trust. Instead of simply asking another organisation to promise that data will be handled correctly, technical and organisational mechanisms can increasingly provide verifiable guarantees. “I think trust isn’t something you should aim for directly,” Dave says. “You aim for trustworthiness — verification, standards, assurance mechanisms — and trust follows on its own.”
Healthcare shows what is possible
The value of that approach becomes particularly tangible in healthcare, where data is distributed across organisations but is often too sensitive to simply bring together. Sarah points to pilots in which hospitals have combined information without exposing the identities of individual patients. This makes it possible to create much richer datasets for research. “If you combine data across different organisations, you get a clear idea of patient journeys,” she explains. “You can see where somebody entered the healthcare system, what medication they received and where they eventually ended up. If you have that information for many people, you can say much more about the effectiveness of treatments or interventions.”
Dave points to rare diseases as an example of where international data sharing could make an especially large difference. “The challenge with rare diseases is that, by definition, you have very few data points,” he explains. “Any one country trying to do that analysis is quite limited. If you could bring together data across different countries, you could do research with much more statistical power. And if you could extend that into countries in the Global South, you would also have much more diverse data, instead of research skewed towards Western populations. Biovault is a great open-source example of this approach being applied to enable new research on real-world data”
Evaluating AI systems
A similar challenge is emerging around AI: independent organisations and governments increasingly want to evaluate AI systems, but doing so can require access to models, training data or user logs containing commercially or personally sensitive information. “OpenMined has been working with AI Safety Institutes and other third-party evaluation organisations to devise safe methods for evaluating different types of AI systems,” Dave says. “PETs can enable independent evaluation with real verification behind it, providing assurance to everyone involved without exposing the underlying models or data.”
The hardest barriers are not technical
Successful pilots, however, do not automatically lead to widespread adoption. Sarah sees an important cultural barrier, particularly among organisations that currently hold valuable datasets. “Especially in the Netherlands, we have this culture of ‘better safe than sorry’. Organisations tend to hold on to their data. In some cases, datasets are also seen as valuable organisational assets with potential commercial value, creating an additional disincentive to share them. Sharing data requires a different way of working, built on trust and mutual dependence”
Dave sees a similar problem in demonstrating the business case. “The potential economic value is real and tangible, but it can be very hard to articulate and show upfront,” he says. “You’re talking about how making more use of data can create more economic value, but putting numbers on it upfront is genuinely difficult.”
“Everyone in the chain should see the value of collaborating,” Sarah adds. “Before you reach that point, you need successful use cases that demonstrate what organisations can actually gain by making their data available for collaboration.”
Creating room for adoption
Dave sees a particular role for governments in accelerating adoption. “Governments have a role in supporting these public-interest use cases,” he says. “They’re in a position where their incentives aren’t just about the bottom line. They can also take some of the perceived risk out of experimenting with PETs and apply them to social-good use cases.”
Clearer regulatory guidance could further reduce that risk. Established compliance processes were often not designed with PETs in mind, which can make organisations hesitant to adopt a new approach. Sandboxes are one way of creating more room to experiment. Dave’s experience working on the UK-US PETs Prize Challenges showed that even direct access to regulators can make a significant difference. “Having experts from the regulator able to provide direct guidance to people building these systems was probably one of the most helpful things. If you can do something like that at scale, you lower the risk for organisations and make it much easier to move forward.”
A European opportunity
Europe could play an important role in creating effective regulatory conditions. Sarah points to developments such as the European Health Data Space, which requires organisations to work out how sensitive data can be shared responsibly across borders.
Dave takes that argument one step further. In the debate about sovereign AI, Europe is often compared to the United States and China in terms of computing infrastructure and AI models. But he believes Europe’s data governance could become an advantage if combined effectively with PETs. “I think a really promising way that Europe can compete is on data,” he says. “If you combine the regulatory regime with these technologies and the emerging data spaces, you could create an ecosystem where AI draws on far more high-quality data without compromising privacy.”
Turning that opportunity into practice will also require communities that connect technical expertise with the wider conditions for adoption. Sarah sees complementary roles for OpenMined and CoE-DSC.
“OpenMined is also about democratising knowledge about PETs and making that knowledge more accessible,” she says. “CoE-DSC helps to go beyond technology and bridge the gap towards governance and ecosystem building. I think such communities are really important if we want to make PETs scalable.”
When PETs become invisible
For Dave, the ultimate goal is for PETs to become trusted infrastructure that works largely unnoticed by the people who ultimately benefit from it. He sees a clear precedent in another technology: encryption. “Think about encryption on the internet. Nobody thinks about it anymore. You see the little padlock in your browser and you trust that the underlying infrastructure works and is safe. We developed the standards, assurance mechanisms and the open infrastructure, and now end users don’t have to think about it.”
PETs aren’t there yet. But that is precisely the direction in which Dave believes the field should develop. “I don’t think we want a world where everybody needs to understand what homomorphic encryption is in order to use it. We want these technologies to become beneficial in a way that’s almost invisible to end users.”
If that happens, the success of PETs may ultimately be measured by how little people have to think about them. The technology will remain important, but largely beneath the surface. Above it will sit something much more recognisable: organisations that can share and use sensitive data with enough assurance to collaborate, and enough trust to actually do so.
