Europe is spending billions trying to keep its data out of American hands. But there’s a problem nobody’s talking about: it’s baked into the processor.
There’s a conversation happening across Europe about digital sovereignty: about keeping data out of the reach of American tech giants and US law. Governments are funding alternatives. France has built one of the world’s most rigorous cloud certification frameworks. The EU is pouring billions into infrastructure that isn’t subject to American jurisdiction.
It’s a worthwhile conversation. But it has a significant blind spot, and that blind spot is sitting inside every modern Intel and AMD processor on the planet.
A computer inside your computer
Since 2006 every Intel chip has shipped with something called the Management Engine (ME), now known as the Converged Security and Manageability Engine (CSME). AMD has an equivalent called the Platform Security Processor (PSP). These aren’t obscure features. They’re in virtually every laptop, desktop, and server you can buy.
The Management Engine is, in effect, a separate computer running inside your computer. It has its own processor, its own memory, its own operating system, and its own network connection. It runs whether or not your main operating system is running and can keep running even when the machine appears to be switched off.
That last point is worth contemplating: Your laptop is in your bag. The lid is shut. The power light is off. And somewhere inside it, a small, independent computer is still running. It’s one that you have no visibility of, and no control over.
Below the operating system
Security researchers describe the Management Engine as operating at “Ring -3” – a privilege level so deep that it sits below the operating system, below virtualisation software, below anything your security tools can see or log.
If you’re not a technical person, here’s the simplest way to think about it: everything you do on a computer happens on a stage. The Management Engine is underneath the stage, in the basement, with its own entrance.
It shares your computer’s network address, which means any traffic it generates looks identical to your own traffic. Your firewall can’t tell the difference. Your antivirus can’t see it. Your VPN doesn’t cover it.
This isn’t theoretical. In 2017, a nation-state hacking group known as PLATINUM was found to have used Intel’s management features as a covert data exfiltration channel. The did not by exploit a flaw, they used the feature exactly as designed, with default credentials that many organisations had never changed.
The sovereignty problem
Now here’s where this connects to the bigger picture. Intel and AMD are American companies.
Under US law (specifically a provision updated in 2024) hardware manufacturers can be compelled by American intelligence agencies to cooperate with secret surveillance orders.
Europe’s sovereign cloud frameworks certify who controls the data centres, who can access the servers, and which laws apply. What they don’t certify is what’s happening inside the chip.
France’s SecNumCloud framework, widely regarded as the gold standard, has nearly 1,200 technical requirements. According to the researchers who designed it, it contains no direct requirement addressing the Intel Management Engine or AMD’s equivalent.
As one security researcher put it plainly: “There is no direct requirement for firmware backdoor prevention.”
This isn’t new – but the stakes are higher
The Intel Management Engine has been around for many years. It has long been a concern in privacy and open-source hardware circles, and security researchers began publishing on it as far back as 2009. The Free Software Foundation has described it as an attack on users’ freedom. The Snowden revelations in 2013 confirmed that intelligence agencies were already exploiting exactly this class of hardware-level capability in the field. In some cases, the capabilities described in leaked documents matched academic proof-of-concept research almost exactly.
What’s changed is the political context. With Europe now spending seriously on digital sovereignty and with US policy becoming an increasingly live concern for organisations that depend on American technology, the gap between what sovereignty frameworks promise and what they actually deliver has become harder to ignore. As someone who works with organisations on ethical IT choices every day, I find it frustrating that this conversation is still largely confined to technical circles.
What can be done?
In the short term, not much at the hardware level, at least. There is no fully supported way to disable the Management Engine entirely on a standard machine. The long-term hope is RISC-V, an open-source processor architecture that wouldn’t carry this kind of embedded management system. But researchers are blunt about the timeline: we’re talking decades, not years.
What can be done is to name the problem honestly. Sovereignty frameworks that certify the legal and operational layer whilst ignoring the silicon layer are incomplete. Not worthless, but incomplete.
Organisations making decisions about where to host data, which hardware to buy, and how much to trust vendor assurances about security ought to know that there is a computer inside their computer that nobody certified, nobody assessed, and nobody can switch off.
The best thing any of us can do right now is stay informed, talk about these issues, and push back against the idea that sovereignty is solved once the data centre is in the right country. It isn’t. These conversations need to happen in boardrooms and council chambers, not just in security research papers.
The Register covered this in depth this week in an excellent piece by Kim Loohuis. If you want to go further down the rabbit hole, it’s well worth your time.