Researchers Sneak Past Spectre v2 by Slipping Between the Kernel's Own Defenses

Key points

• MIT CSAIL researchers Daniël Trujillo and Mengjia Yan disclosed a new attack called Interrupt Injection that bypasses Spectre v2 defenses.
• The technique works on an AMD Zen 2 processor running Linux 6.14 with every default Spectre v2 mitigation switched on.
• The attack times a hardware interrupt to land in the microscopic gap between the CPU clearing its branch predictor and the kernel using it.
• No unusual privileges are needed: an ordinary, unprivileged Linux program can pull it off.
• The finding revives a class of processor-level attacks that chip makers and Linux maintainers have been patching against since 2018.

A new academic attack shows that even a fully patched Linux machine can still be tricked into leaking secrets from its own kernel, the core program that runs the operating system.

The technique is called Interrupt Injection. It was published by Daniël Trujillo and Mengjia Yan at MIT's Computer Science and Artificial Intelligence Laboratory, and first reported by The Hacker News.

It is a fresh twist on Spectre v2, a family of processor flaws disclosed in 2018 that let a program guess at data it should not be allowed to see.

What is Spectre v2, in plain terms?

Modern processors try to work faster by guessing what they will be asked to do next. That guessing machine is called the branch predictor. Spectre v2 attacks poison those guesses so the chip briefly runs code the attacker chose, leaving fingerprints in memory that reveal secrets like passwords or encryption keys.

Chip makers and Linux developers have spent seven years bolting on defenses. One of them tells the processor to sanitize its branch predictor, wiping the attacker's poisoned guesses, before the kernel starts running sensitive code.

How does Interrupt Injection get around that?

By squeezing into the gap. There is a tiny sliver of time between the moment the CPU cleans the predictor and the moment the kernel actually uses it. An unprivileged program, meaning any normal app with no special permissions, can time a hardware interrupt (a signal from a device like a network card that briefly grabs the processor's attention) to land inside that sliver.

When the interrupt fires, it re-poisons the branch predictor after the defense has already run. The kernel then walks straight into the trap.

Which systems are affected?

The researchers demonstrated the attack on an AMD Zen 2 machine running Linux 6.14 with every default Spectre v2 mitigation enabled. They say the underlying idea is not specific to one vendor, and both Intel and AMD chips are in scope.

Should ordinary people be worried?

Not in the immediate, panic-today sense. This is a local attack, which means the person running it needs to already be running code on your machine. It is not something a random website can fire off at your laptop from across the internet.

The real risk sits with cloud providers and shared servers, where many customers' programs live on the same physical processor. That is exactly the setting Spectre-class attacks were built to worry about.

For home users, the practical advice is dull and familiar: install operating system updates when your vendor ships them. Fixes for Interrupt Injection, once they exist, will arrive that way.

What comes next?

Chip vendors and Linux kernel maintainers now have to work out how to close a timing gap they may not have known was exploitable. History suggests the fix will be a mixture of microcode updates from Intel and AMD, and changes in the kernel to make sure interrupts cannot land in that vulnerable window.

Interrupt Injection is not a smash-and-grab bug. It is a reminder that the deep, quiet defenses inside a processor are only as strong as their worst-timed millisecond.