MIT CSAIL researchers have identified a new attack that circumvents Spectre v2 defenses on both Intel and AMD processors. The technique, called Interrupt Injection, exploits a timing window in how operating systems apply branch predictor sanitization.
The attack works by having an unprivileged Linux program time a hardware interrupt to land between two critical moments. The processor first sanitizes its branch predictor as part of Spectre v2 mitigation, then the kernel attempts to use it. By injecting an interrupt at precisely the right moment, attackers re-poison the branch predictor after the defense has already executed, restoring the vulnerability.
Researchers Daniël Trujillo and Mengjia Yan demonstrated the attack on an AMD Zen 2 system running Linux 6.14 with all default Spectre v2 mitigations enabled. This reveals a fundamental gap in how current defenses coordinate processor and kernel-level protections.
Spectre v2 attacks manipulate branch prediction units to trick processors into speculatively executing code paths that leak sensitive data. Since its discovery in 2018, vendors have deployed multiple mitigations across firmware and operating systems. However, Interrupt Injection exposes a race condition in the implementation of these defenses.
The timing window appears to be difficult but feasible to hit reliably. An attacker needs only unprivileged code execution to trigger the interrupt at the exact moment the kernel uses the branch predictor after sanitization.
This discovery carries implications for Linux systems and potentially Windows machines running Intel processors, as the underlying vulnerability stems from processor architecture rather than OS-specific code. Organizations relying on existing Spectre v2 defenses should treat this as a potential gap, though practical exploitation still requires local code execution.
Intel and AMD will likely need to coordinate with kernel maintainers to patch