Is Demand for Data Sovereignty Back?
AI Is Reviving an Old Fight Over Who Gets to Read Whose Data
Like low-rise jeans and Crocs, data sovereignty is back, though not for the usual reasons.
We went a couple of years without hearing much about data sovereignty, but in the last few months, it’s come up frequently. Demand for sovereign solutions from US software companies dropped off in 2023 after the EU-US Data Privacy Framework (DPF) took effect. Now the SaaS companies we talk to are getting requests for data sovereign options all over again, and customers tell us it’s dominating Q&A at their user conferences.
On a related note, I’ve been meeting with international senior government officials in charge of cybersecurity lately and learning about their concerns. Data sovereignty is at the top of the list right next to the impact of AI on security.
The combination of these two things is also turning up in a new term, “AI Sovereignty,” which is a subset of data sovereignty and a major reason behind the latest push.
How we got here
Let me rewind a moment for those not familiar. I’ll speed run you through what I’m talking about:
- GDPR Adequacy: Europe passed a seminal privacy law in 2016, and the law started being enforced in 2018. The primary goal is to protect the data of EU citizens. A key provision in it doesn’t allow other countries (or their software providers) to hold EU citizen data unless some conditions are met. That usually means being on a list of blessed countries with adequate privacy laws.
- EU-US Privacy Shield: In 2016, the EU and US struck an agreement called the EU-US Privacy Shield, which granted the US that adequacy status. It replaced a prior agreement (“Safe Harbor”), which was struck down by a court case called “Schrems I.”
- Schrems II: In 2020, a court case brought by Max Schrems struck down the EU-US Privacy Shield. This started a period of high demand for data sovereign solutions.
- EU-US Data Privacy Framework (DPF): In 2023, a new framework replaced Privacy Shield, ending three years of uncertainty and problems around US companies having access to the private data of EU citizens.
For around three years, between #3 and #4, European companies were reticent to use the services of US companies because of the legal uncertainties, which led to a big push for encryption-driven sovereignty solutions. The encryption approach provides a throughline of trust so companies aren’t at the mercy of the whipsaw of international agreements.
Residency vs. sovereignty
There’s one more piece of history that explains the vocabulary we use now:
- At one point, it was deemed okay for a US company (or any non-EU company) to hold EU citizen data as long as the data stayed within EU (or even per-country) borders. This concept was called “data residency.”
- In 2013, the US served a federal warrant on Microsoft demanding data from its Irish servers.
- Microsoft pushed back saying US laws didn’t govern the data in Ireland, and they won the case on appeal. The government appealed that decision, and the case went to the Supreme Court.
- In 2018, the US Congress made the case moot by passing the CLOUD Act, which compels US companies to respond to warrants regardless of where the data is stored.
The CLOUD Act made data residency an obsolete concept and changed the conversation to instead be about data sovereignty, which comes down to whose courts govern lawful access to the data.
I should note that this isn’t limited to just the US and EU, though that’s the biggest example. Many countries are passing data sovereignty laws across Asia, South America, and the Middle East.
Why AI changed the math
The US is leading the world in frontier AI models. China isn’t far behind, but the rest of the world is left looking to one of those two for their frontier-level AI.
To really make use of AI, you need to give it access to your data. And if that data is sensitive government files, or if you’re a company who values data security, then you have a quandary. Which is why cryptographic solutions are once again gaining in popularity.
On October 13, the CEO of Opaque announced that Anthropic was semi-acquiring them (the details are murky here, as Anthropic seems to have licensed the tech and hired just part of the team, shutting down the business in the process). Opaque built confidential computing solutions and focused in recent times on running private AI and compute workloads in cryptographic isolation chambers in the cloud. Anyone running the servers couldn’t see what was being fed in or what came out, which is a pattern that can preserve sovereignty even if the computers are hosted in the US and run by US companies.
Anthropic’s move underscores how demand is forcing the largest model providers to find solutions that preserve customer privacy in provable ways.
Anthropic already has a thriving enterprise business that lets companies run their own copies of Anthropic models inside AWS Bedrock. This trades trust in Anthropic for trust in AWS, which for many is an improvement. But even with the Bedrock approach, US authorities with valid warrants can gain access to logs and data. A cryptographic approach is still needed to keep data sovereign when selling to international customers.
How cryptography keeps data sovereign
There are basically two ways to keep data sovereign. In the first case, for example, if Germany wants its citizens’ data to remain sovereign, then that happens automatically if the data is held in Germany and is solely under the control of German companies. This is a very tricky proposition for foreign companies who have to license out their tech and manage expensive islands of infrastructure. It’s also problematic for customers who aren’t free to purchase the best-of-breed solutions.
The second approach uses cryptography, which is what we do here at IronCore. The stored data is application-layer encrypted so peeking at it or exporting it produces garbage. The key that unlocks the data is stored in the target jurisdiction in a non-exportable manner (HYOK) so a US vendor served with a CLOUD Act warrant can only hand over ciphertext. See how SaaS companies avoid compelled access with encryption for more details.
We have software and integration patterns that give US SaaS companies the ability to offer data sovereign solutions to their customers, including ways to limit where data can be decrypted. As a bonus, the same integrations bring Enterprise-caliber data protections and assurances, which helps companies sell not just internationally, but up-market as well.
The full picture
Data getting fed through AI models is a major risk for people trying to keep their data safe and secure. But the LLM isn’t the only risk in the system. The LLM providers have largely ephemeral systems (if you opt out of letting them train on your queries and if you ignore the logs they keep for at least 30 days and sometimes 2+ years). But AI systems by nature are data hungry, and the workflows and systems around them store a ton of data. Vector databases, for example, contain full copies of many organizations’ most sensitive data repositories, but are largely ignored or not understood by security teams.
For those who do understand and who are looking for AI Sovereignty, they need protections for the rest of their stored AI data. We have VectorLens, a free-to-try tool for scanning vectors for sensitive data. And we offer encryption-in-use solutions for classic keyword search and for vector search products. Deploying these solutions together preserves privacy from start to finish when dealing with AI.
So, is it back?
So to answer the question we pose in the title: yes, data sovereignty is very much back, and this time it’s being driven by both law and technology. The legal pressure never went away, and there are real doubts about whether the DPF will survive a long-anticipated Schrems III. But the bigger driver is the incredible leaps forward that we’re seeing in AI and the lack of data privacy and security options to go with them.
